#include "llvm/Pass.h"
#include "llvm/Analysis/CFG.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/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"
// 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> PrintLiveSetSize("spp-print-liveset-size",
- cl::Hidden, cl::init(false));
+static cl::opt<bool> PrintLiveSetSize("spp-print-liveset-size", cl::Hidden,
+ cl::init(false));
// Print out the base pointers for debugging
-static cl::opt<bool> PrintBasePointers("spp-print-base-pointers",
- 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
+static bool ClobberNonLive = false;
+#endif
+static cl::opt<bool, true> ClobberNonLiveOverride("rs4gc-clobber-non-live",
+ cl::location(ClobberNonLive),
+ 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();
}
"Make relocations explicit at statepoints", false, false)
namespace {
+struct GCPtrLivenessData {
+ /// Values defined in this block.
+ DenseMap<BasicBlock *, DenseSet<Value *>> KillSet;
+ /// Values used in this block (and thus live); does not included values
+ /// killed within this block.
+ DenseMap<BasicBlock *, DenseSet<Value *>> LiveSet;
+
+ /// Values live into this basic block (i.e. used by any
+ /// instruction in this basic block or ones reachable from here)
+ DenseMap<BasicBlock *, DenseSet<Value *>> LiveIn;
+
+ /// Values live out of this basic block (i.e. live into
+ /// any successor block)
+ DenseMap<BasicBlock *, DenseSet<Value *>> LiveOut;
+};
+
// The type of the internal cache used inside the findBasePointers family
// of functions. From the callers perspective, this is an opaque type and
// should not be inspected.
// Generally, after the execution of a full findBasePointer call, only the
// base relation will remain. Internally, we add a mixture of the two
// types, then update all the second type to the first type
-typedef std::map<Value *, Value *> DefiningValueMapTy;
+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
- std::set<llvm::Value *> liveset;
+ StatepointLiveSetTy liveset;
/// Mapping from live pointers to a base-defining-value
DenseMap<llvm::Value *, llvm::Value *> PointerToBase;
- /// Any new values which were added to the IR during base pointer analysis
- /// for this safepoint
- DenseSet<llvm::Value *> NewInsertedDefs;
-
- /// The bounds of the inserted code for the safepoint
- std::pair<Instruction *, Instruction *> SafepointBounds;
+ /// The *new* gc.statepoint instruction itself. This produces the token
+ /// that normal path gc.relocates and the gc.result are tied to.
+ Instruction *StatepointToken;
/// 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;
};
-}
+} // namespace
+
+/// Compute the live-in set for every basic block in the function
+static void computeLiveInValues(DominatorTree &DT, Function &F,
+ GCPtrLivenessData &Data);
+
+/// Given results from the dataflow liveness computation, find the set of live
+/// Values at a particular instruction.
+static void findLiveSetAtInst(Instruction *inst, GCPtrLivenessData &Data,
+ StatepointLiveSetTy &out);
// TODO: Once we can get to the GCStrategy, this becomes
// Optional<bool> isGCManagedPointer(const Value *V) const override {
return false;
}
-/// Return true if the Value is a gc reference type which is potentially used
-/// after the instruction 'loc'. This is only used with the edge reachability
-/// liveness code. Note: It is assumed the V dominates loc.
-static bool isLiveGCReferenceAt(Value &V, Instruction *loc, DominatorTree &DT,
- LoopInfo *LI) {
- if (!isGCPointerType(V.getType()))
- return false;
-
- if (V.use_empty())
- return false;
-
- // Given assumption that V dominates loc, this may be live
- return true;
+// Return true if this type is one which a) is a gc pointer or contains a GC
+// pointer and b) is of a type this code expects to encounter as a live value.
+// (The insertion code will assert that a type which matches (a) and not (b)
+// is not encountered.)
+static bool isHandledGCPointerType(Type *T) {
+ // We fully support gc pointers
+ if (isGCPointerType(T))
+ return true;
+ // We partially support vectors of gc pointers. The code will assert if it
+ // can't handle something.
+ if (auto VT = dyn_cast<VectorType>(T))
+ if (isGCPointerType(VT->getElementType()))
+ return true;
+ return false;
}
#ifndef NDEBUG
-static bool isAggWhichContainsGCPtrType(Type *Ty) {
+/// Returns true if this type contains a gc pointer whether we know how to
+/// handle that type or not.
+static bool containsGCPtrType(Type *Ty) {
+ if (isGCPointerType(Ty))
+ return true;
if (VectorType *VT = dyn_cast<VectorType>(Ty))
return isGCPointerType(VT->getScalarType());
- else if (ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
- return isGCPointerType(AT->getElementType()) ||
- isAggWhichContainsGCPtrType(AT->getElementType());
- } else if (StructType *ST = dyn_cast<StructType>(Ty)) {
- bool UnsupportedType = false;
- for (Type *SubType : ST->subtypes())
- UnsupportedType |=
- isGCPointerType(SubType) || isAggWhichContainsGCPtrType(SubType);
- return UnsupportedType;
- } else
- return false;
+ if (ArrayType *AT = dyn_cast<ArrayType>(Ty))
+ return containsGCPtrType(AT->getElementType());
+ if (StructType *ST = dyn_cast<StructType>(Ty))
+ return std::any_of(
+ ST->subtypes().begin(), ST->subtypes().end(),
+ [](Type *SubType) { return containsGCPtrType(SubType); });
+ return false;
}
-#endif
-// Conservatively identifies any definitions which might be live at the
-// given instruction. The analysis is performed immediately before the
-// given instruction. Values defined by that instruction are not considered
-// live. Values used by that instruction are considered live.
-//
-// preconditions: valid IR graph, term is either a terminator instruction or
-// a call instruction, pred is the basic block of term, DT, LI are valid
-//
-// side effects: none, does not mutate IR
-//
-// postconditions: populates liveValues as discussed above
-static void findLiveGCValuesAtInst(Instruction *term, BasicBlock *pred,
- DominatorTree &DT, LoopInfo *LI,
- std::set<llvm::Value *> &liveValues) {
- liveValues.clear();
-
- assert(isa<CallInst>(term) || isa<InvokeInst>(term) || term->isTerminator());
-
- Function *F = pred->getParent();
-
- auto is_live_gc_reference =
- [&](Value &V) { return isLiveGCReferenceAt(V, term, DT, LI); };
-
- // Are there any gc pointer arguments live over this point? This needs to be
- // special cased since arguments aren't defined in basic blocks.
- for (Argument &arg : F->args()) {
- assert(!isAggWhichContainsGCPtrType(arg.getType()) &&
- "support for FCA unimplemented");
-
- if (is_live_gc_reference(arg)) {
- liveValues.insert(&arg);
- }
- }
-
- // Walk through all dominating blocks - the ones which can contain
- // definitions used in this block - and check to see if any of the values
- // they define are used in locations potentially reachable from the
- // interesting instruction.
- BasicBlock *BBI = pred;
- while (true) {
- if (TraceLSP) {
- errs() << "[LSP] Looking at dominating block " << pred->getName() << "\n";
- }
- assert(DT.dominates(BBI, pred));
- assert(isPotentiallyReachable(BBI, pred, &DT) &&
- "dominated block must be reachable");
-
- // Walk through the instructions in dominating blocks and keep any
- // that have a use potentially reachable from the block we're
- // considering putting the safepoint in
- for (Instruction &inst : *BBI) {
- if (TraceLSP) {
- errs() << "[LSP] Looking at instruction ";
- inst.dump();
- }
-
- if (pred == BBI && (&inst) == term) {
- if (TraceLSP) {
- errs() << "[LSP] stopped because we encountered the safepoint "
- "instruction.\n";
- }
-
- // If we're in the block which defines the interesting instruction,
- // we don't want to include any values as live which are defined
- // _after_ the interesting line or as part of the line itself
- // i.e. "term" is the call instruction for a call safepoint, the
- // results of the call should not be considered live in that stackmap
- break;
- }
-
- assert(!isAggWhichContainsGCPtrType(inst.getType()) &&
- "support for FCA unimplemented");
-
- if (is_live_gc_reference(inst)) {
- if (TraceLSP) {
- errs() << "[LSP] found live value for this safepoint ";
- inst.dump();
- term->dump();
- }
- liveValues.insert(&inst);
- }
- }
- if (!DT.getNode(BBI)->getIDom()) {
- assert(BBI == &F->getEntryBlock() &&
- "failed to find a dominator for something other than "
- "the entry block");
- break;
- }
- BBI = DT.getNode(BBI)->getIDom()->getBlock();
- }
+// Returns true if this is a type which a) is a gc pointer or contains a GC
+// pointer and b) is of a type which the code doesn't expect (i.e. first class
+// aggregates). Used to trip assertions.
+static bool isUnhandledGCPointerType(Type *Ty) {
+ return containsGCPtrType(Ty) && !isHandledGCPointerType(Ty);
}
+#endif
static bool order_by_name(llvm::Value *a, llvm::Value *b) {
if (a->hasName() && b->hasName()) {
}
}
-/// Find the initial live set. Note that due to base pointer
-/// insertion, the live set may be incomplete.
-static void
-analyzeParsePointLiveness(DominatorTree &DT, const CallSite &CS,
- PartiallyConstructedSafepointRecord &result) {
+// Conservatively identifies any definitions which might be live at the
+// given instruction. The analysis is performed immediately before the
+// given instruction. Values defined by that instruction are not considered
+// live. Values used by that instruction are considered live.
+static void analyzeParsePointLiveness(
+ DominatorTree &DT, GCPtrLivenessData &OriginalLivenessData,
+ const CallSite &CS, PartiallyConstructedSafepointRecord &result) {
Instruction *inst = CS.getInstruction();
- BasicBlock *BB = inst->getParent();
- std::set<Value *> liveset;
- findLiveGCValuesAtInst(inst, BB, DT, nullptr, liveset);
+ StatepointLiveSetTy liveset;
+ findLiveSetAtInst(inst, OriginalLivenessData, liveset);
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
// by name
- std::vector<Value *> temp;
+ 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";
result.liveset = liveset;
}
-/// True iff this value is the null pointer constant (of any pointer type)
-static bool isNullConstant(Value *V) {
- return isa<Constant>(V) && isa<PointerType>(V->getType()) &&
- cast<Constant>(V)->isNullValue();
+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) {
+ assert(I->getType()->isVectorTy() &&
+ cast<VectorType>(I->getType())->getElementType()->isPointerTy() &&
+ "Illegal to ask for the base pointer of a non-pointer type");
+
+ // Each case parallels findBaseDefiningValue below, see that code for
+ // detailed motivation.
+
+ if (isa<Argument>(I))
+ // An incoming argument to the function is a base pointer
+ return I;
+
+ // We shouldn't see the address of a global as a vector value?
+ assert(!isa<GlobalVariable>(I) &&
+ "unexpected global variable found in base of vector");
+
+ // 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;
+
+ // Due to inheritance, this must be _after_ the global variable and undef
+ // checks
+ if (Constant *Con = dyn_cast<Constant>(I)) {
+ assert(!isa<GlobalVariable>(I) && !isa<UndefValue>(I) &&
+ "order of checks wrong!");
+ assert(Con->isNullValue() && "null is the only case which makes sense");
+ return Con;
+ }
+
+ if (isa<LoadInst>(I))
+ return I;
+
+ // 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.
+ 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");
}
/// Helper function for findBasePointer - Will return a value which either a)
assert(I->getType()->isPointerTy() &&
"Illegal to ask for the base pointer of a non-pointer type");
- // There are instructions which can never return gc pointer values. Sanity
- // check
- // that this is actually true.
- assert(!isa<InsertElementInst>(I) && !isa<ExtractElementInst>(I) &&
- !isa<ShuffleVectorInst>(I) && "Vector types are not gc pointers");
- assert((!isa<Instruction>(I) || isa<InvokeInst>(I) ||
- !cast<Instruction>(I)->isTerminator()) &&
- "With the exception of invoke terminators don't define values");
- assert(!isa<StoreInst>(I) && !isa<FenceInst>(I) &&
- "Can't be definitions to start with");
- assert(!isa<ICmpInst>(I) && !isa<FCmpInst>(I) &&
- "Comparisons don't give ops");
- // There's a bunch of instructions which just don't make sense to apply to
- // a pointer. The only valid reason for this would be pointer bit
- // twiddling which we're just not going to support.
- assert((!isa<Instruction>(I) || !cast<Instruction>(I)->isBinaryOp()) &&
- "Binary ops on pointer values are meaningless. Unless your "
- "bit-twiddling which we don't support");
-
- if (Argument *Arg = dyn_cast<Argument>(I)) {
+ // This case is a bit of a hack - it only handles extracts from vectors which
+ // trivially contain only base pointers or cases where we can directly match
+ // the index of the original extract element to an insertion into the vector.
+ // See note inside the function for how to improve this.
+ 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;
+ }
+
+ 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
- assert(Arg->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return Arg;
- }
+ return I;
- if (GlobalVariable *global = dyn_cast<GlobalVariable>(I)) {
+ if (isa<GlobalVariable>(I))
// base case
- assert(global->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return global;
- }
+ return I;
// inlining could possibly introduce phi node that contains
// undef if callee has multiple returns
- if (UndefValue *undef = dyn_cast<UndefValue>(I)) {
- assert(undef->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return undef; // utterly meaningless, but useful for dealing with
- // partially optimized code.
- }
+ if (isa<UndefValue>(I))
+ // utterly meaningless, but useful for dealing with
+ // partially optimized code.
+ return I;
// Due to inheritance, this must be _after_ the global variable and undef
// checks
- if (Constant *con = dyn_cast<Constant>(I)) {
+ if (Constant *Con = dyn_cast<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(con->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- assert(con->isNullValue() && "null is the only case which makes sense");
- return con;
+ assert(isa<ConstantPointerNull>(Con) &&
+ "null is the only case which makes sense");
+ return Con;
}
if (CastInst *CI = dyn_cast<CastInst>(I)) {
- Value *def = CI->stripPointerCasts();
- assert(def->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- if (isa<CastInst>(def)) {
- // If we find a cast instruction here, it means we've found a cast
- // which is not simply a pointer cast (i.e. an inttoptr). We don't
- // know how to handle int->ptr conversion.
- llvm_unreachable("Can not find the base pointers for an inttoptr cast");
- }
- assert(!isa<CastInst>(def) && "shouldn't find another cast here");
- return findBaseDefiningValue(def);
+ Value *Def = CI->stripPointerCasts();
+ // If we find a cast instruction here, it means we've found a cast which is
+ // not simply a pointer cast (i.e. an inttoptr). We don't know how to
+ // handle int->ptr conversion.
+ assert(!isa<CastInst>(Def) && "shouldn't find another cast here");
+ return findBaseDefiningValue(Def);
}
- if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
- if (LI->getType()->isPointerTy()) {
- Value *Op = LI->getOperand(0);
- (void)Op;
- // Has to be a pointer to an gc object, or possibly an array of such?
- assert(Op->getType()->isPointerTy());
- return LI; // The value loaded is an gc base itself
- }
- }
- if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
- Value *Op = GEP->getOperand(0);
- if (Op->getType()->isPointerTy()) {
- return findBaseDefiningValue(Op); // The base of this GEP is the base
- }
- }
+ if (isa<LoadInst>(I))
+ return I; // The value loaded is an gc base itself
- if (AllocaInst *alloc = dyn_cast<AllocaInst>(I)) {
- // An alloca represents a conceptual stack slot. It's the slot itself
- // that the GC needs to know about, not the value in the slot.
- assert(alloc->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return alloc;
- }
+ if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I))
+ // The base of this GEP is the base
+ return findBaseDefiningValue(GEP->getPointerOperand());
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
switch (II->getIntrinsicID()) {
+ case Intrinsic::experimental_gc_result_ptr:
default:
// fall through to general call handling
break;
case Intrinsic::experimental_gc_result_float:
case Intrinsic::experimental_gc_result_int:
llvm_unreachable("these don't produce pointers");
- case Intrinsic::experimental_gc_result_ptr:
- // This is just a special case of the CallInst check below to handle a
- // statepoint with deopt args which hasn't been rewritten for GC yet.
- // TODO: Assert that the statepoint isn't rewritten yet.
- return II;
case Intrinsic::experimental_gc_relocate: {
// Rerunning safepoint insertion after safepoints are already
// inserted is not supported. It could probably be made to work,
// We assume that functions in the source language only return base
// pointers. This should probably be generalized via attributes to support
// both source language and internal functions.
- if (CallInst *call = dyn_cast<CallInst>(I)) {
- assert(call->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return call;
- }
- if (InvokeInst *invoke = dyn_cast<InvokeInst>(I)) {
- assert(invoke->getType()->isPointerTy() &&
- "Base for pointer must be another pointer");
- return invoke;
- }
+ if (isa<CallInst>(I) || isa<InvokeInst>(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.
assert(!isa<LandingPadInst>(I) && "Landing Pad is unimplemented");
- if (AtomicCmpXchgInst *cas = dyn_cast<AtomicCmpXchgInst>(I)) {
+ 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. We loaded a pointer from a address in memory, that value
- // had better be a valid base pointer.
- return cas->getPointerOperand();
- }
- if (AtomicRMWInst *atomic = dyn_cast<AtomicRMWInst>(I)) {
- assert(AtomicRMWInst::Xchg == atomic->getOperation() &&
- "All others are binary ops which don't apply to base pointers");
- // semantically, a load, store pair. Treat it the same as a standard load
- return atomic->getPointerOperand();
- }
+ // like a load.
+ return I;
+
+ assert(!isa<AtomicRMWInst>(I) && "Xchg handled above, all others are "
+ "binary ops which don't apply to pointers");
// The aggregate ops. Aggregates can either be in the heap or on the
// 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 (ExtractValueInst *ev = dyn_cast<ExtractValueInst>(I)) {
- return ev;
- }
+ if (isa<ExtractValueInst>(I))
+ return I;
// We should never see an insert vector since that would require we be
// tracing back a struct value not a pointer value.
// return a value which dynamically selects from amoung several base
// derived pointers (each with it's own base potentially). It's the job of
// the caller to resolve these.
- if (SelectInst *select = dyn_cast<SelectInst>(I)) {
- return select;
- }
- if (PHINode *phi = dyn_cast<PHINode>(I)) {
- return phi;
- }
-
- errs() << "unknown type: " << *I << "\n";
- llvm_unreachable("unknown type");
- return nullptr;
+ assert((isa<SelectInst>(I) || isa<PHINode>(I)) &&
+ "missing instruction case in findBaseDefiningValing");
+ return I;
}
/// Returns the base defining value for this value.
-static Value *findBaseDefiningValueCached(Value *I, DefiningValueMapTy &cache) {
- Value *&Cached = cache[I];
+static Value *findBaseDefiningValueCached(Value *I, DefiningValueMapTy &Cache) {
+ Value *&Cached = Cache[I];
if (!Cached) {
Cached = findBaseDefiningValue(I);
}
- assert(cache[I] != nullptr);
+ assert(Cache[I] != nullptr);
if (TraceLSP) {
- errs() << "fBDV-cached: " << I->getName() << " -> " << Cached->getName()
+ dbgs() << "fBDV-cached: " << I->getName() << " -> " << Cached->getName()
<< "\n";
}
return Cached;
/// Return a base pointer for this value if known. Otherwise, return it's
/// base defining value.
-static Value *findBaseOrBDV(Value *I, DefiningValueMapTy &cache) {
- Value *def = findBaseDefiningValueCached(I, cache);
- auto Found = cache.find(def);
- if (Found != cache.end()) {
+static Value *findBaseOrBDV(Value *I, DefiningValueMapTy &Cache) {
+ Value *Def = findBaseDefiningValueCached(I, Cache);
+ auto Found = Cache.find(Def);
+ if (Found != Cache.end()) {
// Either a base-of relation, or a self reference. Caller must check.
return Found->second;
}
// Only a BDV available
- return def;
+ return Def;
}
/// 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)) {
+static bool isKnownBaseResult(Value *V) {
+ if (!isa<PHINode>(V) && !isa<SelectInst>(V)) {
// no recursion possible
return true;
}
- if (cast<Instruction>(v)->getMetadata("is_base_value")) {
+ if (isa<Instruction>(V) &&
+ cast<Instruction>(V)->getMetadata("is_base_value")) {
// This is a previously inserted base phi or select. We know
// that this is a base value.
return true;
}
PhiState(Value *b) : status(Base), base(b) {}
PhiState() : status(Unknown), base(nullptr) {}
- PhiState(const PhiState &other) : status(other.status), base(other.base) {
- assert(status != Base || base);
- }
Status getStatus() const { return status; }
Value *getBase() const { return base; }
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
// class that implementes the meet operation. The meat of the meet
// operation is implemented in MeetPhiStates::pureMeet
class MeetPhiStates {
public:
// phiStates is a mapping from PHINodes and SelectInst's to PhiStates.
- explicit MeetPhiStates(const std::map<Value *, PhiState> &phiStates)
+ explicit MeetPhiStates(const ConflictStateMapTy &phiStates)
: phiStates(phiStates) {}
// Destructively meet the current result with the base V. V can
PhiState getResult() const { return currentResult; }
private:
- const std::map<Value *, PhiState> &phiStates;
+ const ConflictStateMapTy &phiStates;
PhiState currentResult;
/// Return a phi state for a base defining value. We'll generate a new
case PhiState::Base:
assert(stateA.getBase() && "can't be null");
- if (stateB.isUnknown()) {
+ if (stateB.isUnknown())
return stateA;
- } else if (stateB.isBase()) {
+
+ if (stateB.isBase()) {
if (stateA.getBase() == stateB.getBase()) {
assert(stateA == stateB && "equality broken!");
return stateA;
}
return PhiState(PhiState::Conflict);
- } else {
- assert(stateB.isConflict() && "only three states!");
- return PhiState(PhiState::Conflict);
}
+ assert(stateB.isConflict() && "only three states!");
+ return PhiState(PhiState::Conflict);
case PhiState::Conflict:
return stateA;
}
- assert(false && "only three states!");
+ llvm_unreachable("only three states!");
}
};
-}
+} // namespace
/// 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
/// relocation.) On failure, returns nullptr.
-static Value *findBasePointer(Value *I, DefiningValueMapTy &cache,
- DenseSet<llvm::Value *> &NewInsertedDefs) {
+static Value *findBasePointer(Value *I, DefiningValueMapTy &cache) {
Value *def = findBaseOrBDV(I, cache);
if (isKnownBaseResult(def)) {
// analougous to pessimistic data flow and would likely lead to an
// overall worse solution.
- std::map<Value *, PhiState> states;
+ 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
- // PERF: Yes, this is as horribly inefficient as it looks.
+ // 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;
+ 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)) {
- unsigned NumPHIValues = phi->getNumIncomingValues();
- assert(NumPHIValues > 0 && "zero input phis are illegal");
- for (unsigned i = 0; i != NumPHIValues; ++i) {
- Value *InVal = phi->getIncomingValue(i);
+ 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();
// have reached conflict state. The current version seems too conservative.
bool progress = true;
- size_t oldSize = 0;
while (progress) {
- oldSize = states.size();
+#ifndef NDEBUG
+ size_t oldSize = states.size();
+#endif
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?");
- assert(isa<SelectInst>(v) || isa<PHINode>(v));
if (SelectInst *select = dyn_cast<SelectInst>(v)) {
calculateMeet.meetWith(findBaseOrBDV(select->getTrueValue(), cache));
calculateMeet.meetWith(findBaseOrBDV(select->getFalseValue(), cache));
- } else if (PHINode *phi = dyn_cast<PHINode>(v)) {
- for (unsigned i = 0; i < phi->getNumIncomingValues(); i++) {
- calculateMeet.meetWith(
- findBaseOrBDV(phi->getIncomingValue(i), cache));
- }
- } else {
- llvm_unreachable("no such state expected");
- }
+ } else
+ for (Value *Val : cast<PHINode>(v)->incoming_values())
+ calculateMeet.meetWith(findBaseOrBDV(Val, cache));
PhiState oldState = states[v];
PhiState newState = calculateMeet.getResult();
}
// 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());
for (auto Pair : states) {
- Instruction *v = cast<Instruction>(Pair.first);
- PhiState state = Pair.second;
+ Value *V = Pair.first;
+ Keys.push_back(V);
+ }
+ 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()) {
- 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);
- NewInsertedDefs.insert(phi);
- // 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 if (SelectInst *sel = dyn_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);
- NewInsertedDefs.insert(basesel);
- // 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);
- } else {
- assert(false);
- }
+ 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);
+ // 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);
}
}
assert(!isKnownBaseResult(v) && "why did it get added?");
assert(!state.isUnknown() && "Optimistic algorithm didn't complete!");
- if (state.isConflict()) {
- if (PHINode *basephi = dyn_cast<PHINode>(state.getBase())) {
- PHINode *phi = cast<PHINode>(v);
- unsigned NumPHIValues = phi->getNumIncomingValues();
- for (unsigned i = 0; i < NumPHIValues; i++) {
- Value *InVal = phi->getIncomingValue(i);
- BasicBlock *InBB = phi->getIncomingBlock(i);
-
- // If we've already seen InBB, add the same incoming value
- // we added for it earlier. The IR verifier requires phi
- // nodes with multiple entries from the same basic block
- // to have the same incoming value for each of those
- // entries. If we don't do this check here and basephi
- // has a different type than base, we'll end up adding two
- // bitcasts (and hence two distinct values) as incoming
- // values for the same basic block.
-
- int blockIndex = basephi->getBasicBlockIndex(InBB);
- 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!");
- assert(NewInsertedDefs.count(base) &&
- "should have already added this in a prev. iteration!");
- }
-
- // In essense 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() &&
- "sanity -- findBaseOrBDV should be pure!");
-#endif
- continue;
- }
+ if (!state.isConflict())
+ continue;
- // Find either the defining value for the PHI or the normal base for
- // a non-phi node
+ if (PHINode *basephi = dyn_cast<PHINode>(state.getBase())) {
+ PHINode *phi = cast<PHINode>(v);
+ unsigned NumPHIValues = phi->getNumIncomingValues();
+ for (unsigned i = 0; i < NumPHIValues; i++) {
+ Value *InVal = phi->getIncomingValue(i);
+ BasicBlock *InBB = phi->getIncomingBlock(i);
+
+ // If we've already seen InBB, add the same incoming value
+ // we added for it earlier. The IR verifier requires phi
+ // nodes with multiple entries from the same basic block
+ // to have the same incoming value for each of those
+ // entries. If we don't do this check here and basephi
+ // has a different type than base, we'll end up adding two
+ // bitcasts (and hence two distinct values) as incoming
+ // values for the same basic block.
+
+ int blockIndex = basephi->getBasicBlockIndex(InBB);
+ 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.
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());
- NewInsertedDefs.insert(base);
- }
- basephi->addIncoming(base, InBB);
+
+ // In essense 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() &&
+ "sanity -- findBaseOrBDV should be pure!");
+#endif
+ continue;
}
- assert(basephi->getNumIncomingValues() == NumPHIValues);
- } 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);
- NewInsertedDefs.insert(base);
- }
- basesel->setOperand(i, base);
+
+ // 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!");
}
- } else {
- assert(false && "unexpected type");
+ 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);
+ }
+ assert(basephi->getNumIncomingValues() == NumPHIValues);
+ } else {
+ SelectInst *basesel = 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);
}
}
}
// post condition: PointerToBase contains one (derived, base) pair for every
// pointer in live. Note that derived can be equal to base if the original
// pointer was a base pointer.
-static void findBasePointers(const std::set<llvm::Value *> &live,
- DenseMap<llvm::Value *, llvm::Value *> &PointerToBase,
- DominatorTree *DT, DefiningValueMapTy &DVCache,
- DenseSet<llvm::Value *> &NewInsertedDefs) {
- for (Value *ptr : live) {
- Value *base = findBasePointer(ptr, DVCache, NewInsertedDefs);
+static void
+findBasePointers(const StatepointLiveSetTy &live,
+ DenseMap<llvm::Value *, llvm::Value *> &PointerToBase,
+ DominatorTree *DT, DefiningValueMapTy &DVCache) {
+ // For the naming of values inserted to be deterministic - which makes for
+ // much cleaner and more stable tests - we need to assign an order to the
+ // live values. DenseSets do not provide a deterministic order across runs.
+ SmallVector<Value *, 64> Temp;
+ Temp.insert(Temp.end(), live.begin(), live.end());
+ std::sort(Temp.begin(), Temp.end(), order_by_name);
+ for (Value *ptr : Temp) {
+ Value *base = findBasePointer(ptr, DVCache);
assert(base && "failed to find base pointer");
PointerToBase[ptr] = base;
assert((!isa<Instruction>(base) || !isa<Instruction>(ptr) ||
cast<Instruction>(ptr)->getParent())) &&
"The base we found better dominate the derived pointer");
- if (isNullConstant(base))
- // 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.
- llvm_unreachable("the relocation code needs adjustment to handle the"
- "relocation of a null pointer constant without causing"
- "false positives in the safepoint ir verifier.");
+ // 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.
+ assert(!isa<ConstantPointerNull>(base) &&
+ "the relocation code needs adjustment to handle the relocation of "
+ "a null pointer constant without causing false positives in the "
+ "safepoint ir verifier.");
}
}
const CallSite &CS,
PartiallyConstructedSafepointRecord &result) {
DenseMap<llvm::Value *, llvm::Value *> PointerToBase;
- DenseSet<llvm::Value *> NewInsertedDefs;
- findBasePointers(result.liveset, PointerToBase, &DT, DVCache, NewInsertedDefs);
+ findBasePointers(result.liveset, PointerToBase, &DT, DVCache);
if (PrintBasePointers) {
+ // Note: Need to print these in a stable order since this is checked in
+ // some tests.
errs() << "Base Pairs (w/o Relocation):\n";
+ SmallVector<Value *, 64> Temp;
+ Temp.reserve(PointerToBase.size());
for (auto Pair : PointerToBase) {
- errs() << " derived %" << Pair.first->getName() << " base %"
- << Pair.second->getName() << "\n";
+ Temp.push_back(Pair.first);
}
- }
-
- result.PointerToBase = PointerToBase;
- result.NewInsertedDefs = NewInsertedDefs;
-}
-
-/// Check for liveness of items in the insert defs and add them to the live
-/// and base pointer sets
-static void fixupLiveness(DominatorTree &DT, const CallSite &CS,
- const std::set<Value *> &allInsertedDefs,
- PartiallyConstructedSafepointRecord &result) {
- Instruction *inst = CS.getInstruction();
-
- auto liveset = result.liveset;
- auto PointerToBase = result.PointerToBase;
-
- auto is_live_gc_reference =
- [&](Value &V) { return isLiveGCReferenceAt(V, inst, DT, nullptr); };
-
- // For each new definition, check to see if a) the definition dominates the
- // instruction we're interested in, and b) one of the uses of that definition
- // is edge-reachable from the instruction we're interested in. This is the
- // same definition of liveness we used in the intial liveness analysis
- for (Value *newDef : allInsertedDefs) {
- if (liveset.count(newDef)) {
- // already live, no action needed
- continue;
- }
-
- // PERF: Use DT to check instruction domination might not be good for
- // compilation time, and we could change to optimal solution if this
- // turn to be a issue
- if (!DT.dominates(cast<Instruction>(newDef), inst)) {
- // can't possibly be live at inst
- continue;
- }
-
- if (is_live_gc_reference(*newDef)) {
- // Add the live new defs into liveset and PointerToBase
- liveset.insert(newDef);
- PointerToBase[newDef] = newDef;
+ std::sort(Temp.begin(), Temp.end(), order_by_name);
+ for (Value *Ptr : Temp) {
+ Value *Base = PointerToBase[Ptr];
+ errs() << " derived %" << Ptr->getName() << " base %" << Base->getName()
+ << "\n";
}
}
- result.liveset = liveset;
result.PointerToBase = PointerToBase;
}
-static void fixupLiveReferences(
- Function &F, DominatorTree &DT, Pass *P,
- const std::set<llvm::Value *> &allInsertedDefs,
- std::vector<CallSite> &toUpdate,
- std::vector<struct PartiallyConstructedSafepointRecord> &records) {
+/// Given an updated version of the dataflow liveness results, update the
+/// liveset and base pointer maps for the call site CS.
+static void recomputeLiveInValues(GCPtrLivenessData &RevisedLivenessData,
+ const CallSite &CS,
+ PartiallyConstructedSafepointRecord &result);
+
+static void recomputeLiveInValues(
+ 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.
+ GCPtrLivenessData RevisedLivenessData;
+ computeLiveInValues(DT, F, RevisedLivenessData);
for (size_t i = 0; i < records.size(); i++) {
struct PartiallyConstructedSafepointRecord &info = records[i];
- CallSite &CS = toUpdate[i];
- fixupLiveness(DT, CS, allInsertedDefs, info);
+ const CallSite &CS = toUpdate[i];
+ recomputeLiveInValues(RevisedLivenessData, CS, info);
}
}
-// Normalize basic block to make it ready to be target of invoke statepoint.
-// It means spliting it to have single predecessor. Return newly created BB
-// ready to be successor of invoke statepoint.
-static BasicBlock *normalizeBBForInvokeSafepoint(BasicBlock *BB,
- BasicBlock *InvokeParent,
- Pass *P) {
- BasicBlock *ret = BB;
-
+// When inserting gc.relocate calls, we need to ensure there are no uses
+// of the original value between the gc.statepoint and the gc.relocate call.
+// One case which can arise is a phi node starting one of the successor blocks.
+// We also need to be able to insert the gc.relocates only on the path which
+// goes through the statepoint. We might need to split an edge to make this
+// possible.
+static BasicBlock *
+normalizeForInvokeSafepoint(BasicBlock *BB, BasicBlock *InvokeParent,
+ DominatorTree &DT) {
+ BasicBlock *Ret = BB;
if (!BB->getUniquePredecessor()) {
- ret = SplitBlockPredecessors(BB, InvokeParent, "");
+ Ret = SplitBlockPredecessors(BB, InvokeParent, "", nullptr, &DT);
}
- // Another requirement for such basic blocks is to not have any phi nodes.
- // Since we just ensured that new BB will have single predecessor,
- // all phi nodes in it will have one value. Here it would be naturall place
- // to
- // remove them all. But we can not do this because we are risking to remove
- // one of the values stored in liveset of another statepoint. We will do it
- // later after placing all safepoints.
+ // Now that 'ret' has unique predecessor we can safely remove all phi nodes
+ // from it
+ FoldSingleEntryPHINodes(Ret);
+ assert(!isa<PHINode>(Ret->begin()));
- return ret;
+ // At this point, we can safely insert a gc.relocate as the first instruction
+ // in Ret if needed.
+ return Ret;
}
-static void
-VerifySafepointBounds(const std::pair<Instruction *, Instruction *> &bounds) {
- assert(bounds.first->getParent() && bounds.second->getParent() &&
- "both must belong to basic blocks");
- if (bounds.first->getParent() == bounds.second->getParent()) {
- // This is a call safepoint
- // TODO: scan the range to find the statepoint
- // TODO: check that the following instruction is not a gc_relocate or
- // gc_result
- } else {
- // This is an invoke safepoint
- InvokeInst *invoke = dyn_cast<InvokeInst>(bounds.first);
- (void)invoke;
- assert(invoke && "only continues over invokes!");
- assert(invoke->getNormalDest() == bounds.second->getParent() &&
- "safepoint should continue into normal exit block");
- }
-}
-
-static int find_index(const SmallVectorImpl<Value *> &livevec, Value *val) {
+static int find_index(ArrayRef<Value *> livevec, Value *val) {
auto itr = std::find(livevec.begin(), livevec.end(), val);
assert(livevec.end() != itr);
size_t index = std::distance(livevec.begin(), itr);
/// statepointToken - statepoint instruction to which relocates should be
/// bound.
/// Builder - Llvm IR builder to be used to construct new calls.
-/// Returns array with newly created relocates.
-static std::vector<llvm::Instruction *>
-CreateGCRelocates(const SmallVectorImpl<llvm::Value *> &liveVariables,
- const int liveStart,
- const SmallVectorImpl<llvm::Value *> &basePtrs,
- Instruction *statepointToken, IRBuilder<> Builder) {
-
- std::vector<llvm::Instruction *> newDefs;
+static void CreateGCRelocates(ArrayRef<llvm::Value *> LiveVariables,
+ const int LiveStart,
+ ArrayRef<llvm::Value *> BasePtrs,
+ Instruction *StatepointToken,
+ IRBuilder<> Builder) {
+ SmallVector<Instruction *, 64> NewDefs;
+ NewDefs.reserve(LiveVariables.size());
- Module *M = statepointToken->getParent()->getParent()->getParent();
+ Module *M = StatepointToken->getParent()->getParent()->getParent();
- for (unsigned i = 0; i < liveVariables.size(); i++) {
+ 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.
- std::vector<Type *> types; // one per 'any' type
- types.push_back(liveVariables[i]->getType()); // result type
- Value *gc_relocate_decl = Intrinsic::getDeclaration(
- M, Intrinsic::experimental_gc_relocate, types);
+ 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 =
+ Value *BaseIdx =
ConstantInt::get(Type::getInt32Ty(M->getContext()),
- liveStart + find_index(liveVariables, basePtrs[i]));
- Value *liveIdx = ConstantInt::get(
+ LiveStart + find_index(LiveVariables, BasePtrs[i]));
+ Value *LiveIdx = ConstantInt::get(
Type::getInt32Ty(M->getContext()),
- liveStart + find_index(liveVariables, liveVariables[i]));
+ LiveStart + find_index(LiveVariables, LiveVariables[i]));
// only specify a debug name if we can give a useful one
- Value *reloc = Builder.CreateCall3(
- gc_relocate_decl, statepointToken, baseIdx, liveIdx,
- liveVariables[i]->hasName() ? liveVariables[i]->getName() + ".relocated"
+ Value *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);
+ cast<CallInst>(Reloc)->setCallingConv(CallingConv::Cold);
- newDefs.push_back(cast<Instruction>(reloc));
+ NewDefs.push_back(cast<Instruction>(Reloc));
}
- assert(newDefs.size() == liveVariables.size() &&
+ assert(NewDefs.size() == LiveVariables.size() &&
"missing or extra redefinition at safepoint");
-
- return newDefs;
}
static void
IRBuilder<> Builder(insertBefore);
// Copy all of the arguments from the original statepoint - this includes the
// target, call args, and deopt args
- std::vector<llvm::Value *> args;
+ SmallVector<llvm::Value *, 64> args;
args.insert(args.end(), CS.arg_begin(), CS.arg_end());
// TODO: Clear the 'needs rewrite' flag
Builder.SetInsertPoint(IP);
Builder.SetCurrentDebugLocation(IP->getDebugLoc());
- } else if (CS.isInvoke()) {
+ } else {
InvokeInst *toReplace = cast<InvokeInst>(CS.getInstruction());
// Insert the new invoke into the old block. We'll remove the old one in a
token = invoke;
// Generate gc relocates in exceptional path
- BasicBlock *unwindBlock = normalizeBBForInvokeSafepoint(
- toReplace->getUnwindDest(), invoke->getParent(), P);
+ BasicBlock *unwindBlock = toReplace->getUnwindDest();
+ assert(!isa<PHINode>(unwindBlock->begin()) &&
+ unwindBlock->getUniquePredecessor() &&
+ "can't safely insert in this block!");
Instruction *IP = &*(unwindBlock->getFirstInsertionPt());
Builder.SetInsertPoint(IP);
exceptional_token, Builder);
// Generate gc relocates and returns for normal block
- BasicBlock *normalDest = normalizeBBForInvokeSafepoint(
- toReplace->getNormalDest(), invoke->getParent(), P);
+ BasicBlock *normalDest = toReplace->getNormalDest();
+ assert(!isa<PHINode>(normalDest->begin()) &&
+ normalDest->getUniquePredecessor() &&
+ "can't safely insert in this block!");
IP = &*(normalDest->getFirstInsertionPt());
Builder.SetInsertPoint(IP);
// gc relocates will be generated later as if it were regular call
// statepoint
- } else {
- llvm_unreachable("unexpect type of CallSite");
}
assert(token);
// Take the name of the original value call if it had one.
token->takeName(CS.getInstruction());
- // The GCResult is already inserted, we just need to find it
- Instruction *gc_result = nullptr;
- /* scope */ {
- Instruction *toReplace = CS.getInstruction();
- assert((toReplace->hasNUses(0) || toReplace->hasNUses(1)) &&
- "only valid use before rewrite is gc.result");
- if (toReplace->hasOneUse()) {
- Instruction *GCResult = cast<Instruction>(*toReplace->user_begin());
- assert(isGCResult(GCResult));
- gc_result = GCResult;
- }
- }
+// The GCResult is already inserted, we just need to find it
+#ifndef NDEBUG
+ Instruction *toReplace = CS.getInstruction();
+ assert((toReplace->hasNUses(0) || toReplace->hasNUses(1)) &&
+ "only valid use before rewrite is gc.result");
+ assert(!toReplace->hasOneUse() ||
+ isGCResult(cast<Instruction>(*toReplace->user_begin())));
+#endif
// Update the gc.result of the original statepoint (if any) to use the newly
// inserted statepoint. This is safe to do here since the token can't be
// considered a live reference.
CS.getInstruction()->replaceAllUsesWith(token);
- // Second, create a gc.relocate for every live variable
- std::vector<llvm::Instruction *> newDefs =
- CreateGCRelocates(liveVariables, live_start, basePtrs, token, Builder);
-
- // Need to pass through the last part of the safepoint block so that we
- // don't accidentally update uses in a following gc.relocate which is
- // still conceptually part of the same safepoint. Gah.
- Instruction *last = nullptr;
- if (!newDefs.empty()) {
- last = newDefs.back();
- } else if (gc_result) {
- last = gc_result;
- } else {
- last = token;
- }
- assert(last && "can't be null");
- const auto bounds = std::make_pair(token, last);
-
- // Sanity check our results - this is slightly non-trivial due to invokes
- VerifySafepointBounds(bounds);
+ result.StatepointToken = token;
- result.SafepointBounds = bounds;
+ // Second, create a gc.relocate for every live variable
+ CreateGCRelocates(liveVariables, live_start, basePtrs, token, Builder);
}
namespace {
SmallVectorImpl<Value *> &livevec) {
assert(basevec.size() == livevec.size());
- std::vector<name_ordering> temp;
+ SmallVector<name_ordering, 64> temp;
for (size_t i = 0; i < basevec.size(); i++) {
name_ordering v;
v.base = basevec[i];
// Replace an existing gc.statepoint with a new one and a set of gc.relocates
// which make the relocations happening at this safepoint explicit.
-//
+//
// WARNING: Does not do any fixup to adjust users of the original live
// values. That's the callers responsibility.
static void
// Add visited values into the visitedLiveValues set we will later use them
// for sanity check.
static void
-insertRelocationStores(iterator_range<Value::user_iterator> gcRelocs,
- DenseMap<Value *, Value *> &allocaMap,
- DenseSet<Value *> &visitedLiveValues) {
+insertRelocationStores(iterator_range<Value::user_iterator> GCRelocs,
+ DenseMap<Value *, Value *> &AllocaMap,
+ DenseSet<Value *> &VisitedLiveValues) {
- for (User *U : gcRelocs) {
+ for (User *U : GCRelocs) {
if (!isa<IntrinsicInst>(U))
continue;
- IntrinsicInst *relocatedValue = cast<IntrinsicInst>(U);
+ IntrinsicInst *RelocatedValue = cast<IntrinsicInst>(U);
// We only care about relocates
- if (relocatedValue->getIntrinsicID() !=
+ if (RelocatedValue->getIntrinsicID() !=
Intrinsic::experimental_gc_relocate) {
continue;
}
- GCRelocateOperands relocateOperands(relocatedValue);
- Value *originalValue = const_cast<Value *>(relocateOperands.derivedPtr());
- assert(allocaMap.count(originalValue));
- Value *alloca = allocaMap[originalValue];
+ GCRelocateOperands RelocateOperands(RelocatedValue);
+ Value *OriginalValue =
+ const_cast<Value *>(RelocateOperands.getDerivedPtr());
+ assert(AllocaMap.count(OriginalValue));
+ Value *Alloca = AllocaMap[OriginalValue];
// Emit store into the related alloca
- StoreInst *store = new StoreInst(relocatedValue, alloca);
- store->insertAfter(relocatedValue);
+ // All gc_relocate are i8 addrspace(1)* typed, and it must be bitcasted to
+ // the correct type according to alloca.
+ assert(RelocatedValue->getNextNode() && "Should always have one since it's not a terminator");
+ IRBuilder<> Builder(RelocatedValue->getNextNode());
+ Value *CastedRelocatedValue =
+ Builder.CreateBitCast(RelocatedValue, cast<AllocaInst>(Alloca)->getAllocatedType(),
+ RelocatedValue->hasName() ? RelocatedValue->getName() + ".casted" : "");
+
+ StoreInst *Store = new StoreInst(CastedRelocatedValue, Alloca);
+ Store->insertAfter(cast<Instruction>(CastedRelocatedValue));
#ifndef NDEBUG
- visitedLiveValues.insert(originalValue);
+ VisitedLiveValues.insert(OriginalValue);
+#endif
+ }
+}
+
+// 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, const std::vector<Value *> &live,
- const std::vector<struct PartiallyConstructedSafepointRecord> &records) {
+ Function &F, DominatorTree &DT, ArrayRef<Value *> Live,
+ ArrayRef<struct PartiallyConstructedSafepointRecord> Records) {
#ifndef NDEBUG
- int initialAllocaNum = 0;
-
- // record initial number of allocas
- for (inst_iterator itr = inst_begin(F), end = inst_end(F); itr != end;
- itr++) {
- if (isa<AllocaInst>(*itr))
- initialAllocaNum++;
- }
+ // record initial number of (static) allocas; we'll check we have the same
+ // number when we get done.
+ int InitialAllocaNum = 0;
+ for (auto I = F.getEntryBlock().begin(), E = F.getEntryBlock().end(); I != E;
+ I++)
+ if (isa<AllocaInst>(*I))
+ InitialAllocaNum++;
#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.SafepointBounds.first;
+ 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);
+ if (isa<InvokeInst>(Statepoint)) {
+ insertRelocationStores(Info.UnwindToken->users(), AllocaMap,
+ VisitedLiveValues);
}
-#ifndef NDEBUG
- // As a debuging aid, pretend that an unrelocated pointer becomes null at
- // the gc.statepoint. This will turn some subtle GC problems into slightly
- // easy to debug SEGVs
- for (auto Pair : allocaMap) {
- Value *def = Pair.first;
- Value *alloca = Pair.second;
-
- // This value was relocated
- if (visitedLiveValues.count(def)) {
- continue;
+ // 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
+ // 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) {
+ Value *Def = Pair.first;
+ AllocaInst *Alloca = cast<AllocaInst>(Pair.second);
+
+ // This value was relocated
+ if (VisitedLiveValues.count(Def)) {
+ continue;
+ }
+ ToClobber.push_back(Alloca);
}
- auto PT = cast<PointerType>(def->getType());
- Constant *CPN = ConstantPointerNull::get(PT);
- StoreInst *store = new StoreInst(CPN, alloca);
- store->insertBefore(info.SafepointBounds.second);
+ auto InsertClobbersAt = [&](Instruction *IP) {
+ for (auto *AI : ToClobber) {
+ 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);
+ }
+ };
+
+ // Insert the clobbering stores. These may get intermixed with the
+ // gc.results and gc.relocates, but that's fine.
+ if (auto II = dyn_cast<InvokeInst>(Statepoint)) {
+ InsertClobbersAt(II->getNormalDest()->getFirstInsertionPt());
+ InsertClobbersAt(II->getUnwindDest()->getFirstInsertionPt());
+ } else {
+ BasicBlock::iterator Next(cast<CallInst>(Statepoint));
+ Next++;
+ InsertClobbersAt(Next);
+ }
}
-#endif
}
// 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 (isa<Instruction>(def)) {
- store->insertAfter(cast<Instruction>(def));
+ 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());
+ } else {
+ assert(!Inst->isTerminator() &&
+ "The only TerminatorInst that can produce a value is "
+ "InvokeInst which is handled above.");
+ Store->insertAfter(Inst);
+ }
} else {
- assert((isa<Argument>(def) || isa<GlobalVariable>(def) ||
- (isa<Constant>(def) && cast<Constant>(def)->isNullValue())) &&
- "Must be argument or global");
- 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
}
#ifndef NDEBUG
- for (inst_iterator itr = inst_begin(F), end = inst_end(F); itr != end;
- itr++) {
- if (isa<AllocaInst>(*itr))
- initialAllocaNum--;
- }
- assert(initialAllocaNum == 0 && "We must not introduce any extra allocas");
+ for (auto I = F.getEntryBlock().begin(), E = F.getEntryBlock().end(); I != E;
+ I++)
+ if (isa<AllocaInst>(*I))
+ InitialAllocaNum--;
+ assert(InitialAllocaNum == 0 && "We must not introduce any extra allocas");
#endif
}
/// Implement a unique function which doesn't require we sort the input
/// 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(std::vector<T> &vec) {
- DenseSet<T> seen;
- std::vector<T> tmp;
- vec.reserve(vec.size());
- std::swap(tmp, vec);
- for (auto V : tmp) {
- if (seen.insert(V).second) {
- vec.push_back(V);
- }
- }
-}
-
-static Function *getUseHolder(Module &M) {
- FunctionType *ftype =
- FunctionType::get(Type::getVoidTy(M.getContext()), true);
- Function *Func = cast<Function>(M.getOrInsertFunction("__tmp_use", ftype));
- return Func;
+template <typename T> static void unique_unsorted(SmallVectorImpl<T> &Vec) {
+ 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
-/// liftetime of the call.
+/// lifetime of the call.
static void insertUseHolderAfter(CallSite &CS, const ArrayRef<Value *> Values,
- std::vector<CallInst *> &holders) {
+ SmallVectorImpl<CallInst *> &Holders) {
+ if (Values.empty())
+ // No values to hold live, might as well not insert the empty holder
+ return;
+
Module *M = CS.getInstruction()->getParent()->getParent()->getParent();
- Function *Func = getUseHolder(*M);
+ // Use a dummy vararg function to actually hold the values live
+ Function *Func = cast<Function>(M->getOrInsertFunction(
+ "__tmp_use", FunctionType::get(Type::getVoidTy(M->getContext()), true)));
if (CS.isCall()) {
// For call safepoints insert dummy calls right after safepoint
- BasicBlock::iterator next(CS.getInstruction());
- next++;
- CallInst *base_holder = CallInst::Create(Func, Values, "", next);
- holders.push_back(base_holder);
- } else if (CS.isInvoke()) {
- // For invoke safepooints insert dummy calls both in normal and
- // exceptional destination blocks
- InvokeInst *invoke = cast<InvokeInst>(CS.getInstruction());
- CallInst *normal_holder = CallInst::Create(
- Func, Values, "", invoke->getNormalDest()->getFirstInsertionPt());
- CallInst *unwind_holder = CallInst::Create(
- Func, Values, "", invoke->getUnwindDest()->getFirstInsertionPt());
- holders.push_back(normal_holder);
- holders.push_back(unwind_holder);
- } else {
- assert(false && "Unsupported");
- }
+ BasicBlock::iterator Next(CS.getInstruction());
+ Next++;
+ Holders.push_back(CallInst::Create(Func, Values, "", Next));
+ return;
+ }
+ // For invoke safepooints insert dummy calls both in normal and
+ // exceptional destination blocks
+ auto *II = cast<InvokeInst>(CS.getInstruction());
+ Holders.push_back(CallInst::Create(
+ Func, Values, "", II->getNormalDest()->getFirstInsertionPt()));
+ Holders.push_back(CallInst::Create(
+ Func, Values, "", II->getUnwindDest()->getFirstInsertionPt()));
}
static void findLiveReferences(
- Function &F, DominatorTree &DT, Pass *P, std::vector<CallSite> &toUpdate,
- std::vector<struct PartiallyConstructedSafepointRecord> &records) {
+ Function &F, DominatorTree &DT, Pass *P, ArrayRef<CallSite> toUpdate,
+ MutableArrayRef<struct PartiallyConstructedSafepointRecord> records) {
+ GCPtrLivenessData OriginalLivenessData;
+ computeLiveInValues(DT, F, OriginalLivenessData);
for (size_t i = 0; i < records.size(); i++) {
struct PartiallyConstructedSafepointRecord &info = records[i];
- CallSite &CS = toUpdate[i];
- analyzeParsePointLiveness(DT, CS, info);
+ const CallSite &CS = toUpdate[i];
+ analyzeParsePointLiveness(DT, OriginalLivenessData, CS, info);
}
}
-static void addBasesAsLiveValues(std::set<Value *> &liveset,
- DenseMap<Value *, Value *> &PointerToBase) {
- // Identify any base pointers which are used in this safepoint, but not
- // themselves relocated. We need to relocate them so that later inserted
- // safepoints can get the properly relocated base register.
- DenseSet<Value *> missing;
- for (Value *L : liveset) {
- assert(PointerToBase.find(L) != PointerToBase.end());
- Value *base = PointerToBase[L];
- assert(base);
- if (liveset.find(base) == liveset.end()) {
- assert(PointerToBase.find(base) == PointerToBase.end());
- // uniqued by set insert
- missing.insert(base);
+/// 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
+/// 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.
+static void splitVectorValues(Instruction *StatepointInst,
+ StatepointLiveSetTy &LiveSet, DominatorTree &DT) {
+ SmallVector<Value *, 16> ToSplit;
+ for (Value *V : LiveSet)
+ if (isa<VectorType>(V->getType()))
+ ToSplit.push_back(V);
+
+ if (ToSplit.empty())
+ return;
+
+ 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;
+
+ VectorType *VT = cast<VectorType>(V->getType());
+ IRBuilder<> Builder(StatepointInst);
+ 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());
+
+ auto InsertVectorReform = [&](Instruction *IP) {
+ Builder.SetInsertPoint(IP);
+ Builder.SetCurrentDebugLocation(IP->getDebugLoc());
+ Value *ResultVec = UndefValue::get(VT);
+ for (unsigned i = 0; i < VT->getNumElements(); i++)
+ ResultVec = Builder.CreateInsertElement(ResultVec, Elements[i],
+ Builder.getInt32(i));
+ return ResultVec;
+ };
+
+ if (isa<CallInst>(StatepointInst)) {
+ BasicBlock::iterator Next(StatepointInst);
+ Next++;
+ Instruction *IP = &*(Next);
+ Replacements[V].first = InsertVectorReform(IP);
+ Replacements[V].second = nullptr;
+ } else {
+ InvokeInst *Invoke = cast<InvokeInst>(StatepointInst);
+ // We've already normalized - check that we don't have shared destination
+ // blocks
+ BasicBlock *NormalDest = Invoke->getNormalDest();
+ assert(!isa<PHINode>(NormalDest->begin()));
+ BasicBlock *UnwindDest = Invoke->getUnwindDest();
+ assert(!isa<PHINode>(UnwindDest->begin()));
+ // Insert insert element sequences in both successors
+ Instruction *IP = &*(NormalDest->getFirstInsertionPt());
+ Replacements[V].first = InsertVectorReform(IP);
+ IP = &*(UnwindDest->getFirstInsertionPt());
+ Replacements[V].second = InsertVectorReform(IP);
+ }
+ }
+ for (Value *V : ToSplit) {
+ AllocaInst *Alloca = AllocaMap[V];
+
+ // Capture all users before we start mutating use lists
+ SmallVector<Instruction *, 16> Users;
+ for (User *U : V->users())
+ Users.push_back(cast<Instruction>(U));
+
+ for (Instruction *I : Users) {
+ if (auto Phi = dyn_cast<PHINode>(I)) {
+ for (unsigned i = 0; i < Phi->getNumIncomingValues(); i++)
+ if (V == Phi->getIncomingValue(i)) {
+ LoadInst *Load = new LoadInst(
+ Alloca, "", Phi->getIncomingBlock(i)->getTerminator());
+ Phi->setIncomingValue(i, Load);
+ }
+ } else {
+ LoadInst *Load = new LoadInst(Alloca, "", I);
+ I->replaceUsesOfWith(V, Load);
+ }
+ }
+
+ // Store the original value and the replacement value into the alloca
+ StoreInst *Store = new StoreInst(V, Alloca);
+ if (auto I = dyn_cast<Instruction>(V))
+ Store->insertAfter(I);
+ else
+ Store->insertAfter(Alloca);
+
+ // Normal return for invoke, or call return
+ Instruction *Replacement = cast<Instruction>(Replacements[V].first);
+ (new StoreInst(Replacement, Alloca))->insertAfter(Replacement);
+ // Unwind return for invoke only
+ Replacement = cast_or_null<Instruction>(Replacements[V].second);
+ if (Replacement)
+ (new StoreInst(Replacement, Alloca))->insertAfter(Replacement);
+ }
+
+ // apply mem2reg to promote alloca to SSA
+ SmallVector<AllocaInst *, 16> Allocas;
+ for (Value *V : ToSplit)
+ Allocas.push_back(AllocaMap[V]);
+ PromoteMemToReg(Allocas, DT);
+}
+
+// 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.
+// 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.find(LiveValue) != Info.PointerToBase.end());
+ 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;
}
}
- // Note that we want these at the end of the list, otherwise
- // register placement gets screwed up once we lower to STATEPOINT
- // instructions. This is an utter hack, but there doesn't seem to be a
- // better one.
- for (Value *base : missing) {
- assert(base);
- liveset.insert(base);
- PointerToBase[base] = base;
+ // Remove rematerializaed values from the live set
+ for (auto LiveValue: LiveValuesToBeDeleted) {
+ Info.liveset.erase(LiveValue);
}
- assert(liveset.size() == PointerToBase.size());
}
static bool insertParsePoints(Function &F, DominatorTree &DT, Pass *P,
- std::vector<CallSite> &toUpdate) {
+ SmallVectorImpl<CallSite> &toUpdate) {
#ifndef NDEBUG
// sanity check the input
std::set<CallSite> uniqued;
}
#endif
+ // When inserting gc.relocates for invokes, we need to be able to insert at
+ // the top of the successor blocks. See the comment on
+ // normalForInvokeSafepoint on exactly what is needed. Note that this step
+ // may restructure the CFG.
+ for (CallSite CS : toUpdate) {
+ if (!CS.isInvoke())
+ continue;
+ InvokeInst *invoke = cast<InvokeInst>(CS.getInstruction());
+ normalizeForInvokeSafepoint(invoke->getNormalDest(), invoke->getParent(),
+ DT);
+ normalizeForInvokeSafepoint(invoke->getUnwindDest(), invoke->getParent(),
+ DT);
+ }
+
// A list of dummy calls added to the IR to keep various values obviously
// live in the IR. We'll remove all of these when done.
- std::vector<CallInst *> holders;
+ SmallVector<CallInst *, 64> holders;
// Insert a dummy call with all of the arguments to the vm_state we'll need
// for the actual safepoint insertion. This ensures reference arguments in
SmallVector<Value *, 64> DeoptValues;
for (Use &U : StatepointCS.vm_state_args()) {
Value *Arg = cast<Value>(&U);
- if (isGCPointerType(Arg->getType()))
+ assert(!isUnhandledGCPointerType(Arg->getType()) &&
+ "support for FCA unimplemented");
+ if (isHandledGCPointerType(Arg->getType()))
DeoptValues.push_back(Arg);
}
insertUseHolderAfter(CS, DeoptValues, holders);
}
- std::vector<struct PartiallyConstructedSafepointRecord> records;
+ SmallVector<struct PartiallyConstructedSafepointRecord, 64> records;
records.reserve(toUpdate.size());
for (size_t i = 0; i < toUpdate.size(); i++) {
struct PartiallyConstructedSafepointRecord info;
// 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
// gep a + 1
// safepoint 2
// br loop
- std::set<llvm::Value *> allInsertedDefs;
- for (size_t i = 0; i < records.size(); i++) {
- struct PartiallyConstructedSafepointRecord &info = records[i];
- allInsertedDefs.insert(info.NewInsertedDefs.begin(),
- info.NewInsertedDefs.end());
- }
-
// We insert some dummy calls after each safepoint to definitely hold live
// the base pointers which were identified for that safepoint. We'll then
// ask liveness for _every_ base inserted to see what is now live. Then we
insertUseHolderAfter(CS, Bases, holders);
}
- // Add the bases explicitly to the live vector set. This may result in a few
- // extra relocations, but the base has to be available whenever a pointer
- // derived from it is used. Thus, we need it to be part of the statepoint's
- // gc arguments list. TODO: Introduce an explicit notion (in the following
- // code) of the GC argument list as seperate from the live Values at a
- // given statepoint.
- for (size_t i = 0; i < records.size(); i++) {
- struct PartiallyConstructedSafepointRecord &info = records[i];
- addBasesAsLiveValues(info.liveset, info.PointerToBase);
- }
+ // By selecting base pointers, we've effectively inserted new uses. Thus, we
+ // need to rerun liveness. We may *also* have inserted new defs, but that's
+ // not the key issue.
+ recomputeLiveInValues(F, DT, P, toUpdate, records);
- // If we inserted any new values, we need to adjust our notion of what is
- // live at a particular safepoint.
- if (!allInsertedDefs.empty()) {
- fixupLiveReferences(F, DT, P, allInsertedDefs, toUpdate, records);
- }
if (PrintBasePointers) {
for (size_t i = 0; i < records.size(); i++) {
struct PartiallyConstructedSafepointRecord &info = records[i];
}
holders.clear();
+ // 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
+ // 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
}
toUpdate.clear(); // prevent accident use of invalid CallSites
- // In case if we inserted relocates in a different basic block than the
- // original safepoint (this can happen for invokes). We need to be sure that
- // original values were not used in any of the phi nodes at the
- // beginning of basic block containing them. Because we know that all such
- // blocks will have single predecessor we can safely assume that all phi
- // nodes have single entry (because of normalizeBBForInvokeSafepoint).
- // Just remove them all here.
- for (size_t i = 0; i < records.size(); i++) {
- Instruction *I = records[i].SafepointBounds.first;
-
- if (InvokeInst *invoke = dyn_cast<InvokeInst>(I)) {
- FoldSingleEntryPHINodes(invoke->getNormalDest());
- assert(!isa<PHINode>(invoke->getNormalDest()->begin()));
-
- FoldSingleEntryPHINodes(invoke->getUnwindDest());
- assert(!isa<PHINode>(invoke->getUnwindDest()->begin()));
- }
- }
-
// Do all the fixups of the original live variables to their relocated selves
- std::vector<Value *> live;
+ SmallVector<Value *, 128> live;
for (size_t i = 0; i < records.size(); i++) {
struct PartiallyConstructedSafepointRecord &info = records[i];
// We can't simply save the live set from the original insertion. One of
// That Value* no longer exists and we need to use the new gc_result.
// Thankfully, the liveset is embedded in the statepoint (and updated), so
// we just grab that.
- Statepoint statepoint(info.SafepointBounds.first);
+ Statepoint statepoint(info.StatepointToken);
live.insert(live.end(), statepoint.gc_args_begin(),
statepoint.gc_args_end());
+#ifndef NDEBUG
+ // Do some basic sanity checks on our liveness results before performing
+ // relocation. Relocation can and will turn mistakes in liveness results
+ // into non-sensical code which is must harder to debug.
+ // TODO: It would be nice to test consistency as well
+ assert(DT.isReachableFromEntry(info.StatepointToken->getParent()) &&
+ "statepoint must be reachable or liveness is meaningless");
+ for (Value *V : statepoint.gc_args()) {
+ if (!isa<Instruction>(V))
+ // Non-instruction values trivial dominate all possible uses
+ continue;
+ auto LiveInst = cast<Instruction>(V);
+ assert(DT.isReachableFromEntry(LiveInst->getParent()) &&
+ "unreachable values should never be live");
+ assert(DT.dominates(LiveInst, info.StatepointToken) &&
+ "basic SSA liveness expectation violated by liveness analysis");
+ }
+#endif
}
unique_unsorted(live);
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 : inst_range(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;
- // Gather all the statepoints which need rewritten.
- std::vector<CallSite> ParsePointNeeded;
- for (inst_iterator itr = inst_begin(F), end = inst_end(F); itr != end;
- itr++) {
+ 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)) {
// TODO: only the ones with the flag set!
- if (isStatepoint(*itr))
- ParsePointNeeded.push_back(CallSite(&*itr));
+ if (isStatepoint(I)) {
+ if (DT.isReachableFromEntry(I.getParent()))
+ ParsePointNeeded.push_back(CallSite(&I));
+ else
+ HasUnreachableStatepoint = true;
+ }
}
+ bool MadeChange = false;
+
+ // Delete any unreachable statepoints so that we don't have unrewritten
+ // statepoints surviving this pass. This makes testing easier and the
+ // resulting IR less confusing to human readers. Rather than be fancy, we
+ // just reuse a utility function which removes the unreachable blocks.
+ if (HasUnreachableStatepoint)
+ MadeChange |= removeUnreachableBlocks(F);
+
// Return early if no work to do.
if (ParsePointNeeded.empty())
- return false;
+ return MadeChange;
+
+ // As a prepass, go ahead and aggressively destroy single entry phi nodes.
+ // These are created by LCSSA. They have the effect of increasing the size
+ // of liveness sets for no good reason. It may be harder to do this post
+ // insertion since relocations and base phis can confuse things.
+ for (BasicBlock &BB : F)
+ if (BB.getUniquePredecessor()) {
+ MadeChange = true;
+ FoldSingleEntryPHINodes(&BB);
+ }
+
+ MadeChange |= insertParsePoints(F, DT, this, ParsePointNeeded);
+ return MadeChange;
+}
+
+// liveness computation via standard dataflow
+// -------------------------------------------------------------------
+
+// TODO: Consider using bitvectors for liveness, the set of potentially
+// interesting values should be small and easy to pre-compute.
+
+/// Compute the live-in set for the location rbegin starting from
+/// the live-out set of the basic block
+static void computeLiveInValues(BasicBlock::reverse_iterator rbegin,
+ BasicBlock::reverse_iterator rend,
+ DenseSet<Value *> &LiveTmp) {
+
+ for (BasicBlock::reverse_iterator ritr = rbegin; ritr != rend; ritr++) {
+ Instruction *I = &*ritr;
+
+ // KILL/Def - Remove this definition from LiveIn
+ LiveTmp.erase(I);
+
+ // Don't consider *uses* in PHI nodes, we handle their contribution to
+ // predecessor blocks when we seed the LiveOut sets
+ if (isa<PHINode>(I))
+ continue;
+
+ // USE - Add to the LiveIn set for this instruction
+ for (Value *V : I->operands()) {
+ assert(!isUnhandledGCPointerType(V->getType()) &&
+ "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:
+ // - 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.
+ // - Second, we can't disallow arbitrary inttoptr constants even
+ // if the language frontend does. Optimization passes are free to
+ // locally exploit facts without respect to global reachability. This
+ // can create sections of code which are dynamically unreachable and
+ // contain just about anything. (see constants.ll in tests)
+ LiveTmp.insert(V);
+ }
+ }
+ }
+}
+
+static void computeLiveOutSeed(BasicBlock *BB, DenseSet<Value *> &LiveTmp) {
+
+ for (BasicBlock *Succ : successors(BB)) {
+ const BasicBlock::iterator E(Succ->getFirstNonPHI());
+ for (BasicBlock::iterator I = Succ->begin(); I != E; I++) {
+ PHINode *Phi = cast<PHINode>(&*I);
+ Value *V = Phi->getIncomingValueForBlock(BB);
+ assert(!isUnhandledGCPointerType(V->getType()) &&
+ "support for FCA unimplemented");
+ if (isHandledGCPointerType(V->getType()) && !isa<Constant>(V)) {
+ LiveTmp.insert(V);
+ }
+ }
+ }
+}
+
+static DenseSet<Value *> computeKillSet(BasicBlock *BB) {
+ DenseSet<Value *> KillSet;
+ for (Instruction &I : *BB)
+ if (isHandledGCPointerType(I.getType()))
+ KillSet.insert(&I);
+ return KillSet;
+}
+
+#ifndef NDEBUG
+/// Check that the items in 'Live' dominate 'TI'. This is used as a basic
+/// sanity check for the liveness computation.
+static void checkBasicSSA(DominatorTree &DT, DenseSet<Value *> &Live,
+ TerminatorInst *TI, bool TermOkay = false) {
+ for (Value *V : Live) {
+ if (auto *I = dyn_cast<Instruction>(V)) {
+ // The terminator can be a member of the LiveOut set. LLVM's definition
+ // of instruction dominance states that V does not dominate itself. As
+ // such, we need to special case this to allow it.
+ if (TermOkay && TI == I)
+ continue;
+ assert(DT.dominates(I, TI) &&
+ "basic SSA liveness expectation violated by liveness analysis");
+ }
+ }
+}
+
+/// Check that all the liveness sets used during the computation of liveness
+/// obey basic SSA properties. This is useful for finding cases where we miss
+/// a def.
+static void checkBasicSSA(DominatorTree &DT, GCPtrLivenessData &Data,
+ BasicBlock &BB) {
+ checkBasicSSA(DT, Data.LiveSet[&BB], BB.getTerminator());
+ checkBasicSSA(DT, Data.LiveOut[&BB], BB.getTerminator(), true);
+ checkBasicSSA(DT, Data.LiveIn[&BB], BB.getTerminator());
+}
+#endif
+
+static void computeLiveInValues(DominatorTree &DT, Function &F,
+ GCPtrLivenessData &Data) {
+
+ SmallSetVector<BasicBlock *, 200> Worklist;
+ auto AddPredsToWorklist = [&](BasicBlock *BB) {
+ // We use a SetVector so that we don't have duplicates in the worklist.
+ Worklist.insert(pred_begin(BB), pred_end(BB));
+ };
+ auto NextItem = [&]() {
+ BasicBlock *BB = Worklist.back();
+ Worklist.pop_back();
+ return BB;
+ };
+
+ // Seed the liveness for each individual block
+ for (BasicBlock &BB : F) {
+ Data.KillSet[&BB] = computeKillSet(&BB);
+ Data.LiveSet[&BB].clear();
+ computeLiveInValues(BB.rbegin(), BB.rend(), Data.LiveSet[&BB]);
+
+#ifndef NDEBUG
+ for (Value *Kill : Data.KillSet[&BB])
+ assert(!Data.LiveSet[&BB].count(Kill) && "live set contains kill");
+#endif
+
+ Data.LiveOut[&BB] = DenseSet<Value *>();
+ computeLiveOutSeed(&BB, Data.LiveOut[&BB]);
+ Data.LiveIn[&BB] = Data.LiveSet[&BB];
+ set_union(Data.LiveIn[&BB], Data.LiveOut[&BB]);
+ set_subtract(Data.LiveIn[&BB], Data.KillSet[&BB]);
+ if (!Data.LiveIn[&BB].empty())
+ AddPredsToWorklist(&BB);
+ }
+
+ // Propagate that liveness until stable
+ while (!Worklist.empty()) {
+ BasicBlock *BB = NextItem();
+
+ // Compute our new liveout set, then exit early if it hasn't changed
+ // despite the contribution of our successor.
+ DenseSet<Value *> LiveOut = Data.LiveOut[BB];
+ const auto OldLiveOutSize = LiveOut.size();
+ for (BasicBlock *Succ : successors(BB)) {
+ assert(Data.LiveIn.count(Succ));
+ set_union(LiveOut, Data.LiveIn[Succ]);
+ }
+ // assert OutLiveOut is a subset of LiveOut
+ if (OldLiveOutSize == LiveOut.size()) {
+ // If the sets are the same size, then we didn't actually add anything
+ // when unioning our successors LiveIn Thus, the LiveIn of this block
+ // hasn't changed.
+ continue;
+ }
+ Data.LiveOut[BB] = LiveOut;
+
+ // Apply the effects of this basic block
+ DenseSet<Value *> LiveTmp = LiveOut;
+ set_union(LiveTmp, Data.LiveSet[BB]);
+ set_subtract(LiveTmp, Data.KillSet[BB]);
+
+ assert(Data.LiveIn.count(BB));
+ const DenseSet<Value *> &OldLiveIn = Data.LiveIn[BB];
+ // assert: OldLiveIn is a subset of LiveTmp
+ if (OldLiveIn.size() != LiveTmp.size()) {
+ Data.LiveIn[BB] = LiveTmp;
+ AddPredsToWorklist(BB);
+ }
+ } // while( !worklist.empty() )
+
+#ifndef NDEBUG
+ // Sanity check our ouput against SSA properties. This helps catch any
+ // missing kills during the above iteration.
+ for (BasicBlock &BB : F) {
+ checkBasicSSA(DT, Data, BB);
+ }
+#endif
+}
+
+static void findLiveSetAtInst(Instruction *Inst, GCPtrLivenessData &Data,
+ StatepointLiveSetTy &Out) {
+
+ BasicBlock *BB = Inst->getParent();
+
+ // Note: The copy is intentional and required
+ assert(Data.LiveOut.count(BB));
+ DenseSet<Value *> LiveOut = Data.LiveOut[BB];
+
+ // We want to handle the statepoint itself oddly. It's
+ // call result is not live (normal), nor are it's arguments
+ // (unless they're used again later). This adjustment is
+ // specifically what we need to relocate
+ BasicBlock::reverse_iterator rend(Inst);
+ computeLiveInValues(BB->rbegin(), rend, LiveOut);
+ LiveOut.erase(Inst);
+ Out.insert(LiveOut.begin(), LiveOut.end());
+}
+
+static void recomputeLiveInValues(GCPtrLivenessData &RevisedLivenessData,
+ const CallSite &CS,
+ PartiallyConstructedSafepointRecord &Info) {
+ Instruction *Inst = CS.getInstruction();
+ StatepointLiveSetTy Updated;
+ findLiveSetAtInst(Inst, RevisedLivenessData, Updated);
+
+#ifndef NDEBUG
+ DenseSet<Value *> Bases;
+ for (auto KVPair : Info.PointerToBase) {
+ Bases.insert(KVPair.second);
+ }
+#endif
+ // We may have base pointers which are now live that weren't before. We need
+ // to update the PointerToBase structure to reflect this.
+ for (auto V : Updated)
+ if (!Info.PointerToBase.count(V)) {
+ assert(Bases.count(V) && "can't find base for unexpected live value");
+ Info.PointerToBase[V] = V;
+ continue;
+ }
+
+#ifndef NDEBUG
+ for (auto V : Updated) {
+ assert(Info.PointerToBase.count(V) &&
+ "must be able to find base for live value");
+ }
+#endif
+
+ // Remove any stale base mappings - this can happen since our liveness is
+ // more precise then the one inherent in the base pointer analysis
+ DenseSet<Value *> ToErase;
+ for (auto KVPair : Info.PointerToBase)
+ if (!Updated.count(KVPair.first))
+ ToErase.insert(KVPair.first);
+ for (auto V : ToErase)
+ Info.PointerToBase.erase(V);
+
+#ifndef NDEBUG
+ for (auto KVPair : Info.PointerToBase)
+ assert(Updated.count(KVPair.first) && "record for non-live value");
+#endif
- DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
- return insertParsePoints(F, DT, this, ParsePointNeeded);
+ Info.liveset = Updated;
}