// Copyright 2014 the V8 project authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "src/compiler/backend/gap-resolver.h" #include "src/base/utils/random-number-generator.h" #include "test/cctest/cctest.h" namespace v8 { namespace internal { namespace compiler { const auto GetRegConfig = RegisterConfiguration::Default; // Fragments the given FP operand into an equivalent set of FP operands to // simplify ParallelMove equivalence testing. void GetCanonicalOperands(const InstructionOperand& op, std::vector* fragments) { CHECK_EQ(kFPAliasing, AliasingKind::kCombine); CHECK(op.IsFPLocationOperand()); const LocationOperand& loc = LocationOperand::cast(op); MachineRepresentation rep = loc.representation(); int base = -1; int aliases = GetRegConfig()->GetAliases( rep, 0, MachineRepresentation::kFloat32, &base); CHECK_LT(0, aliases); CHECK_GE(4, aliases); int index = -1; int step = 1; if (op.IsFPRegister()) { index = loc.register_code() * aliases; } else { index = loc.index(); step = -1; } for (int i = 0; i < aliases; i++) { fragments->push_back(AllocatedOperand(loc.location_kind(), MachineRepresentation::kFloat32, index + i * step)); } } // Fake frame size. The test's stack operand indices should be below this value. // Stack slots above this value correspond to temporaries pushed by the gap // resolver to resolve move cycles, and are ignored when comparing interpreter // states. constexpr int kLastFrameSlotId = 1000; // The state of our move interpreter is the mapping of operands to values. Note // that the actual values don't really matter, all we care about is equality. class InterpreterState { public: void ExecuteInParallel(const ParallelMove* moves) { InterpreterState copy(*this); for (const auto m : *moves) { CHECK(!m->IsRedundant()); const InstructionOperand& src = m->source(); const InstructionOperand& dst = m->destination(); if (kFPAliasing == AliasingKind::kCombine && src.IsFPLocationOperand() && dst.IsFPLocationOperand()) { // Canonicalize FP location-location moves by fragmenting them into // an equivalent sequence of float32 moves, to simplify state // equivalence testing. std::vector src_fragments; GetCanonicalOperands(src, &src_fragments); CHECK(!src_fragments.empty()); std::vector dst_fragments; GetCanonicalOperands(dst, &dst_fragments); CHECK_EQ(src_fragments.size(), dst_fragments.size()); for (size_t i = 0; i < src_fragments.size(); ++i) { write(dst_fragments[i], copy.read(src_fragments[i])); } continue; } // All other moves. write(dst, copy.read(src)); } } void ExecuteMove(Zone* zone, InstructionOperand* source, InstructionOperand* dest) { ParallelMove* moves = zone->New(zone); moves->AddMove(*source, *dest); ExecuteInParallel(moves); } void MoveToTempLocation(InstructionOperand& source) { scratch_ = KeyFor(source); } void MoveFromTempLocation(InstructionOperand& dst) { AllocatedOperand src(scratch_.kind, scratch_.rep, scratch_.index); if (kFPAliasing == AliasingKind::kCombine && src.IsFPLocationOperand() && dst.IsFPLocationOperand()) { // Canonicalize FP location-location moves by fragmenting them into // an equivalent sequence of float32 moves, to simplify state // equivalence testing. std::vector src_fragments; GetCanonicalOperands(src, &src_fragments); CHECK(!src_fragments.empty()); std::vector dst_fragments; GetCanonicalOperands(dst, &dst_fragments); CHECK_EQ(src_fragments.size(), dst_fragments.size()); for (size_t i = 0; i < src_fragments.size(); ++i) { write(dst_fragments[i], KeyFor(src_fragments[i])); } return; } write(dst, scratch_); } bool operator==(const InterpreterState& other) const { return values_ == other.values_; } // Clear stack operands above kLastFrameSlotId. They correspond to temporaries // pushed by the gap resolver to break cycles. void ClearTemps() { auto it = values_.begin(); while (it != values_.end()) { if (it->first.kind == LocationOperand::STACK_SLOT && it->first.index >= kLastFrameSlotId) { it = values_.erase(it); } else { it++; } } } private: // struct for mapping operands to a unique value, that makes it easier to // detect illegal parallel moves, and to evaluate moves for equivalence. This // is a one way transformation. All general register and slot operands are // mapped to the default representation. FP registers and slots are mapped to // float64 except on architectures with non-simple FP register aliasing, where // the actual representation is used. struct Key { bool is_constant; MachineRepresentation rep; LocationOperand::LocationKind kind; int index; bool operator<(const Key& other) const { if (this->is_constant != other.is_constant) { return this->is_constant; } if (this->rep != other.rep) { return this->rep < other.rep; } if (this->kind != other.kind) { return this->kind < other.kind; } return this->index < other.index; } bool operator==(const Key& other) const { return this->is_constant == other.is_constant && this->rep == other.rep && this->kind == other.kind && this->index == other.index; } }; // Internally, the state is a normalized permutation of Value pairs. using Value = Key; using OperandMap = std::map; Value read(const InstructionOperand& op) const { OperandMap::const_iterator it = values_.find(KeyFor(op)); return (it == values_.end()) ? ValueFor(op) : it->second; } void write(const InstructionOperand& dst, Value v) { if (v == ValueFor(dst)) { values_.erase(KeyFor(dst)); } else { values_[KeyFor(dst)] = v; } } static Key KeyFor(const InstructionOperand& op) { bool is_constant = op.IsConstant(); MachineRepresentation rep = v8::internal::compiler::InstructionSequence::DefaultRepresentation(); LocationOperand::LocationKind kind; int index; if (!is_constant) { const LocationOperand& loc_op = LocationOperand::cast(op); // Preserve FP representation when FP register aliasing is complex. // Otherwise, canonicalize to kFloat64. if (IsFloatingPoint(loc_op.representation())) { if (kFPAliasing == AliasingKind::kIndependent) { rep = IsSimd128(loc_op.representation()) ? MachineRepresentation::kSimd128 : MachineRepresentation::kFloat64; } else if (kFPAliasing == AliasingKind::kOverlap) { rep = MachineRepresentation::kFloat64; } else { rep = loc_op.representation(); } } if (loc_op.IsAnyRegister()) { index = loc_op.register_code(); } else { index = loc_op.index(); } kind = loc_op.location_kind(); } else { index = ConstantOperand::cast(op).virtual_register(); kind = LocationOperand::REGISTER; } Key key = {is_constant, rep, kind, index}; return key; } static Value ValueFor(const InstructionOperand& op) { return KeyFor(op); } static InstructionOperand FromKey(Key key) { if (key.is_constant) { return ConstantOperand(key.index); } return AllocatedOperand(key.kind, key.rep, key.index); } friend std::ostream& operator<<(std::ostream& os, const InterpreterState& is) { const char* space = ""; for (auto& value : is.values_) { InstructionOperand source = FromKey(value.second); InstructionOperand destination = FromKey(value.first); os << space << MoveOperands{source, destination}; space = " "; } return os; } OperandMap values_; Key scratch_ = {}; }; // An abstract interpreter for moves, swaps and parallel moves. class MoveInterpreter : public GapResolver::Assembler { public: explicit MoveInterpreter(Zone* zone) : zone_(zone) {} AllocatedOperand Push(InstructionOperand* source) override { auto rep = LocationOperand::cast(source)->representation(); int new_slots = ElementSizeInPointers(rep); AllocatedOperand stack_slot(LocationOperand::STACK_SLOT, rep, kLastFrameSlotId + sp_delta_ + new_slots); ParallelMove* moves = zone_->New(zone_); moves->AddMove(*source, stack_slot); state_.ExecuteMove(zone_, source, &stack_slot); sp_delta_ += new_slots; return stack_slot; } void Pop(InstructionOperand* dest, MachineRepresentation rep) override { int new_slots = ElementSizeInPointers(rep); int temp_slot = kLastFrameSlotId + sp_delta_ + new_slots; AllocatedOperand stack_slot(LocationOperand::STACK_SLOT, rep, temp_slot); state_.ExecuteMove(zone_, &stack_slot, dest); sp_delta_ -= new_slots; } void PopTempStackSlots() override { sp_delta_ = 0; state_.ClearTemps(); } void MoveToTempLocation(InstructionOperand* source, MachineRepresentation rep) final { state_.MoveToTempLocation(*source); } void MoveTempLocationTo(InstructionOperand* dest, MachineRepresentation rep) final { state_.MoveFromTempLocation(*dest); } void SetPendingMove(MoveOperands* move) final {} void AssembleMove(InstructionOperand* source, InstructionOperand* destination) override { ParallelMove* moves = zone_->New(zone_); moves->AddMove(*source, *destination); state_.ExecuteInParallel(moves); } void AssembleSwap(InstructionOperand* source, InstructionOperand* destination) override { ParallelMove* moves = zone_->New(zone_); moves->AddMove(*source, *destination); moves->AddMove(*destination, *source); state_.ExecuteInParallel(moves); } void AssembleParallelMove(const ParallelMove* moves) { state_.ExecuteInParallel(moves); } InterpreterState state() const { return state_; } private: Zone* const zone_; InterpreterState state_; int sp_delta_ = 0; }; class ParallelMoveCreator : public HandleAndZoneScope { public: ParallelMoveCreator() : rng_(CcTest::random_number_generator()) {} // Creates a ParallelMove with 'size' random MoveOperands. Note that illegal // moves will be rejected, so the actual number of MoveOperands may be less. ParallelMove* Create(int size) { ParallelMove* parallel_move = main_zone()->New(main_zone()); // Valid ParallelMoves can't have interfering destination ops. std::set destinations; // Valid ParallelMoves can't have interfering source ops of different reps. std::map sources; for (int i = 0; i < size; ++i) { MachineRepresentation rep = RandomRepresentation(); MoveOperands mo(CreateRandomOperand(true, rep), CreateRandomOperand(false, rep)); if (mo.IsRedundant()) continue; const InstructionOperand& dst = mo.destination(); bool reject = false; // On architectures where FP register aliasing is non-simple, update the // destinations set with the float equivalents of the operand and check // that all destinations are unique and do not alias each other. if (kFPAliasing == AliasingKind::kCombine && mo.destination().IsFPLocationOperand()) { std::vector dst_fragments; GetCanonicalOperands(dst, &dst_fragments); CHECK(!dst_fragments.empty()); for (size_t j = 0; j < dst_fragments.size(); ++j) { if (destinations.find(dst_fragments[j]) == destinations.end()) { destinations.insert(dst_fragments[j]); } else { reject = true; break; } } // Update the sources map, and check that no FP source has multiple // representations. const InstructionOperand& src = mo.source(); if (src.IsFPRegister()) { std::vector src_fragments; MachineRepresentation src_rep = LocationOperand::cast(src).representation(); GetCanonicalOperands(src, &src_fragments); CHECK(!src_fragments.empty()); for (size_t j = 0; j < src_fragments.size(); ++j) { auto find_it = sources.find(src_fragments[j]); if (find_it != sources.end() && find_it->second != src_rep) { reject = true; break; } sources.insert(std::make_pair(src_fragments[j], src_rep)); } } } else { if (destinations.find(dst) == destinations.end()) { destinations.insert(dst); } else { reject = true; } } if (!reject) { parallel_move->AddMove(mo.source(), mo.destination()); } } return parallel_move; } // Creates a ParallelMove from a list of operand pairs. Even operands are // destinations, odd ones are sources. ParallelMove* Create(const std::vector& operand_pairs) { ParallelMove* parallel_move = main_zone()->New(main_zone()); for (size_t i = 0; i < operand_pairs.size(); i += 2) { const InstructionOperand& dst = operand_pairs[i]; const InstructionOperand& src = operand_pairs[i + 1]; parallel_move->AddMove(src, dst); } return parallel_move; } private: MachineRepresentation RandomRepresentation() { int index = rng_->NextInt(6); switch (index) { case 0: return MachineRepresentation::kWord32; case 1: return MachineRepresentation::kWord64; case 2: return MachineRepresentation::kFloat32; case 3: return MachineRepresentation::kFloat64; case 4: return MachineRepresentation::kSimd128; case 5: return MachineRepresentation::kTagged; } UNREACHABLE(); } // min(num_alloctable_general_registers for each arch) == 5 from // assembler-ia32.h const int kMaxIndex = 5; const int kMaxIndices = kMaxIndex + 1; // Non-FP slots shouldn't overlap FP slots. // FP slots with different representations shouldn't overlap. int GetValidSlotIndex(MachineRepresentation rep, int index) { DCHECK_GE(kMaxIndex, index); // The first group of slots are for non-FP values. if (!IsFloatingPoint(rep)) return index; // The next group are for float values. int base = kMaxIndices; if (rep == MachineRepresentation::kFloat32) return base + index; // Double values. base += kMaxIndices; if (rep == MachineRepresentation::kFloat64) return base + index * 2; // SIMD values base += kMaxIndices * 2; CHECK_EQ(MachineRepresentation::kSimd128, rep); return base + index * 4; } InstructionOperand CreateRandomOperand(bool is_source, MachineRepresentation rep) { auto conf = RegisterConfiguration::Default(); auto GetValidRegisterCode = [&conf](MachineRepresentation rep, int index) { switch (rep) { case MachineRepresentation::kFloat32: return conf->RegisterConfiguration::GetAllocatableFloatCode(index); case MachineRepresentation::kFloat64: return conf->RegisterConfiguration::GetAllocatableDoubleCode(index); case MachineRepresentation::kSimd128: return conf->RegisterConfiguration::GetAllocatableSimd128Code(index); default: return conf->RegisterConfiguration::GetAllocatableGeneralCode(index); } UNREACHABLE(); }; int index = rng_->NextInt(kMaxIndex); // destination can't be Constant. switch (rng_->NextInt(is_source ? 3 : 2)) { case 0: return AllocatedOperand(LocationOperand::STACK_SLOT, rep, GetValidSlotIndex(rep, index)); case 1: return AllocatedOperand(LocationOperand::REGISTER, rep, GetValidRegisterCode(rep, index)); case 2: return ConstantOperand(index); } UNREACHABLE(); } private: v8::base::RandomNumberGenerator* rng_; }; void RunTest(ParallelMove* pm, Zone* zone) { // Note: The gap resolver modifies the ParallelMove, so interpret first. MoveInterpreter mi1(zone); mi1.AssembleParallelMove(pm); MoveInterpreter mi2(zone); GapResolver resolver(&mi2); resolver.Resolve(pm); auto mi2_state = mi2.state(); CHECK_EQ(mi1.state(), mi2_state); } TEST(Aliasing) { // On platforms with simple aliasing, these parallel moves are ill-formed. if (kFPAliasing != AliasingKind::kCombine) return; ParallelMoveCreator pmc; Zone* zone = pmc.main_zone(); auto s0 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat32, 0); auto s1 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat32, 1); auto s2 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat32, 2); auto s3 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat32, 3); auto s4 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat32, 4); auto d0 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat64, 0); auto d1 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat64, 1); auto d16 = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kFloat64, 16); // Double slots must be odd to match frame allocation. auto dSlot = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kFloat64, 3); // Cycles involving s- and d-registers. { std::vector moves = { s2, s0, // s2 <- s0 d0, d1 // d0 <- d1 }; RunTest(pmc.Create(moves), zone); } { std::vector moves = { d0, d1, // d0 <- d1 s2, s0 // s2 <- s0 }; RunTest(pmc.Create(moves), zone); } { std::vector moves = { s2, s1, // s2 <- s1 d0, d1 // d0 <- d1 }; RunTest(pmc.Create(moves), zone); } { std::vector moves = { d0, d1, // d0 <- d1 s2, s1 // s2 <- s1 }; RunTest(pmc.Create(moves), zone); } // Two cycles involving a single d-register. { std::vector moves = { d0, d1, // d0 <- d1 s2, s1, // s2 <- s1 s3, s0 // s3 <- s0 }; RunTest(pmc.Create(moves), zone); } // Cycle with a float move that must be deferred until after swaps. { std::vector moves = { d0, d1, // d0 <- d1 s2, s0, // s2 <- s0 s3, s4 // s3 <- s4 must be deferred }; RunTest(pmc.Create(moves), zone); } // Cycles involving s-registers and a non-aliased d-register. { std::vector moves = { d16, d0, // d16 <- d0 s1, s2, // s1 <- s2 d1, d16 // d1 <- d16 }; RunTest(pmc.Create(moves), zone); } { std::vector moves = { s2, s1, // s1 <- s2 d0, d16, // d16 <- d0 d16, d1 // d1 <- d16 }; RunTest(pmc.Create(moves), zone); } { std::vector moves = { d0, d16, // d0 <- d16 d16, d1, // s2 <- s0 s3, s0 // d0 <- d1 }; RunTest(pmc.Create(moves), zone); } // Cycle involving aliasing registers and a slot. { std::vector moves = { dSlot, d0, // dSlot <- d0 d1, dSlot, // d1 <- dSlot s0, s3 // s0 <- s3 }; RunTest(pmc.Create(moves), zone); } } // Test parallel moves that change the frame layout. These typically happen when // preparing tail-calls. TEST(ComplexParallelMoves) { ParallelMoveCreator pmc; Zone* zone = pmc.main_zone(); auto w64_2 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kWord64, 2); auto w64_5 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kWord64, 5); auto s128_1 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kSimd128, 1); auto s128_4 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kSimd128, 4); auto s128_5 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kSimd128, 5); auto s128_2 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kSimd128, 2); auto w64_3 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kWord64, 3); auto w64_0 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kWord64, 0); auto s128_6 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kSimd128, 6); auto w64_6 = AllocatedOperand(LocationOperand::STACK_SLOT, MachineRepresentation::kWord64, 6); auto s128_reg = AllocatedOperand(LocationOperand::REGISTER, MachineRepresentation::kSimd128, 0); { // A parallel move with multiple cycles that requires > 1 temporary // location. std::vector moves = { w64_2, w64_5, // - s128_1, s128_4, // - s128_5, s128_2, // - w64_3, w64_0 // - }; RunTest(pmc.Create(moves), zone); } // Regression test for https://crbug.com/1335537. { std::vector moves = { s128_5, s128_6, // - s128_1, s128_6, // - w64_6, w64_0 // - }; RunTest(pmc.Create(moves), zone); } // A cycle with 2 moves that should not use a swap, because the // interfering operands don't have the same base address. { std::vector moves = { s128_1, s128_reg, // - s128_reg, s128_2 // - }; RunTest(pmc.Create(moves), zone); } // Another cycle with 2 moves that should not use a swap, because the // interfering operands don't have the same representation. { std::vector moves = { s128_2, s128_5, // - w64_2, w64_5 // - }; RunTest(pmc.Create(moves), zone); } } TEST(FuzzResolver) { ParallelMoveCreator pmc; for (int size = 0; size < 80; ++size) { for (int repeat = 0; repeat < 50; ++repeat) { RunTest(pmc.Create(size), pmc.main_zone()); } } } } // namespace compiler } // namespace internal } // namespace v8