// Copyright 2016 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/simd-scalar-lowering.h" #include "src/compiler/diamond.h" #include "src/compiler/linkage.h" #include "src/compiler/node-matchers.h" #include "src/compiler/node-properties.h" #include "src/compiler/node.h" #include "src/objects-inl.h" #include "src/wasm/wasm-module.h" namespace v8 { namespace internal { namespace compiler { SimdScalarLowering::SimdScalarLowering( Graph* graph, MachineOperatorBuilder* machine, CommonOperatorBuilder* common, Zone* zone, Signature<MachineRepresentation>* signature) : zone_(zone), graph_(graph), machine_(machine), common_(common), state_(graph, 3), stack_(zone), replacements_(nullptr), signature_(signature), placeholder_( graph->NewNode(common->Parameter(-2, "placeholder"), graph->start())), parameter_count_after_lowering_(-1) { DCHECK_NOT_NULL(graph); DCHECK_NOT_NULL(graph->end()); replacements_ = zone->NewArray<Replacement>(graph->NodeCount()); memset(replacements_, 0, sizeof(Replacement) * graph->NodeCount()); } void SimdScalarLowering::LowerGraph() { stack_.push_back({graph()->end(), 0}); state_.Set(graph()->end(), State::kOnStack); replacements_[graph()->end()->id()].type = SimdType::kInt32; while (!stack_.empty()) { NodeState& top = stack_.back(); if (top.input_index == top.node->InputCount()) { // All inputs of top have already been lowered, now lower top. stack_.pop_back(); state_.Set(top.node, State::kVisited); LowerNode(top.node); } else { // Push the next input onto the stack. Node* input = top.node->InputAt(top.input_index++); if (state_.Get(input) == State::kUnvisited) { SetLoweredType(input, top.node); if (input->opcode() == IrOpcode::kPhi) { // To break cycles with phi nodes we push phis on a separate stack so // that they are processed after all other nodes. PreparePhiReplacement(input); stack_.push_front({input, 0}); } else if (input->opcode() == IrOpcode::kEffectPhi || input->opcode() == IrOpcode::kLoop) { stack_.push_front({input, 0}); } else { stack_.push_back({input, 0}); } state_.Set(input, State::kOnStack); } } } } #define FOREACH_INT32X4_OPCODE(V) \ V(Int32x4Add) \ V(Int32x4ExtractLane) \ V(CreateInt32x4) \ V(Int32x4ReplaceLane) #define FOREACH_FLOAT32X4_OPCODE(V) \ V(Float32x4Add) \ V(Float32x4ExtractLane) \ V(CreateFloat32x4) \ V(Float32x4ReplaceLane) void SimdScalarLowering::SetLoweredType(Node* node, Node* output) { switch (node->opcode()) { #define CASE_STMT(name) case IrOpcode::k##name: FOREACH_INT32X4_OPCODE(CASE_STMT) case IrOpcode::kReturn: case IrOpcode::kParameter: case IrOpcode::kCall: { replacements_[node->id()].type = SimdType::kInt32; break; } FOREACH_FLOAT32X4_OPCODE(CASE_STMT) { replacements_[node->id()].type = SimdType::kFloat32; break; } #undef CASE_STMT default: replacements_[node->id()].type = replacements_[output->id()].type; } } static int GetParameterIndexAfterLowering( Signature<MachineRepresentation>* signature, int old_index) { // In function calls, the simd128 types are passed as 4 Int32 types. The // parameters are typecast to the types as needed for various operations. int result = old_index; for (int i = 0; i < old_index; ++i) { if (signature->GetParam(i) == MachineRepresentation::kSimd128) { result += 3; } } return result; } int SimdScalarLowering::GetParameterCountAfterLowering() { if (parameter_count_after_lowering_ == -1) { // GetParameterIndexAfterLowering(parameter_count) returns the parameter // count after lowering. parameter_count_after_lowering_ = GetParameterIndexAfterLowering( signature(), static_cast<int>(signature()->parameter_count())); } return parameter_count_after_lowering_; } static int GetReturnCountAfterLowering( Signature<MachineRepresentation>* signature) { int result = static_cast<int>(signature->return_count()); for (int i = 0; i < static_cast<int>(signature->return_count()); ++i) { if (signature->GetReturn(i) == MachineRepresentation::kSimd128) { result += 3; } } return result; } void SimdScalarLowering::GetIndexNodes(Node* index, Node** new_indices) { new_indices[0] = index; for (size_t i = 1; i < kMaxLanes; ++i) { new_indices[i] = graph()->NewNode(machine()->Int32Add(), index, graph()->NewNode(common()->Int32Constant( static_cast<int>(i) * kLaneWidth))); } } void SimdScalarLowering::LowerLoadOp(MachineRepresentation rep, Node* node, const Operator* load_op) { if (rep == MachineRepresentation::kSimd128) { Node* base = node->InputAt(0); Node* index = node->InputAt(1); Node* indices[kMaxLanes]; GetIndexNodes(index, indices); Node* rep_nodes[kMaxLanes]; rep_nodes[0] = node; NodeProperties::ChangeOp(rep_nodes[0], load_op); if (node->InputCount() > 2) { DCHECK(node->InputCount() > 3); Node* effect_input = node->InputAt(2); Node* control_input = node->InputAt(3); rep_nodes[3] = graph()->NewNode(load_op, base, indices[3], effect_input, control_input); rep_nodes[2] = graph()->NewNode(load_op, base, indices[2], rep_nodes[3], control_input); rep_nodes[1] = graph()->NewNode(load_op, base, indices[1], rep_nodes[2], control_input); rep_nodes[0]->ReplaceInput(2, rep_nodes[1]); } else { for (size_t i = 1; i < kMaxLanes; ++i) { rep_nodes[i] = graph()->NewNode(load_op, base, indices[i]); } } ReplaceNode(node, rep_nodes); } else { DefaultLowering(node); } } void SimdScalarLowering::LowerStoreOp(MachineRepresentation rep, Node* node, const Operator* store_op, SimdType rep_type) { if (rep == MachineRepresentation::kSimd128) { Node* base = node->InputAt(0); Node* index = node->InputAt(1); Node* indices[kMaxLanes]; GetIndexNodes(index, indices); DCHECK(node->InputCount() > 2); Node* value = node->InputAt(2); DCHECK(HasReplacement(1, value)); Node* rep_nodes[kMaxLanes]; rep_nodes[0] = node; Node** rep_inputs = GetReplacementsWithType(value, rep_type); rep_nodes[0]->ReplaceInput(2, rep_inputs[0]); NodeProperties::ChangeOp(node, store_op); if (node->InputCount() > 3) { DCHECK(node->InputCount() > 4); Node* effect_input = node->InputAt(3); Node* control_input = node->InputAt(4); rep_nodes[3] = graph()->NewNode(store_op, base, indices[3], rep_inputs[3], effect_input, control_input); rep_nodes[2] = graph()->NewNode(store_op, base, indices[2], rep_inputs[2], rep_nodes[3], control_input); rep_nodes[1] = graph()->NewNode(store_op, base, indices[1], rep_inputs[1], rep_nodes[2], control_input); rep_nodes[0]->ReplaceInput(3, rep_nodes[1]); } else { for (size_t i = 1; i < kMaxLanes; ++i) { rep_nodes[i] = graph()->NewNode(store_op, base, indices[i], rep_inputs[i]); } } ReplaceNode(node, rep_nodes); } else { DefaultLowering(node); } } void SimdScalarLowering::LowerBinaryOp(Node* node, SimdType rep_type, const Operator* op) { DCHECK(node->InputCount() == 2); Node** rep_left = GetReplacementsWithType(node->InputAt(0), rep_type); Node** rep_right = GetReplacementsWithType(node->InputAt(1), rep_type); Node* rep_node[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { rep_node[i] = graph()->NewNode(op, rep_left[i], rep_right[i]); } ReplaceNode(node, rep_node); } void SimdScalarLowering::LowerNode(Node* node) { SimdType rep_type = ReplacementType(node); switch (node->opcode()) { case IrOpcode::kStart: { int parameter_count = GetParameterCountAfterLowering(); // Only exchange the node if the parameter count actually changed. if (parameter_count != static_cast<int>(signature()->parameter_count())) { int delta = parameter_count - static_cast<int>(signature()->parameter_count()); int new_output_count = node->op()->ValueOutputCount() + delta; NodeProperties::ChangeOp(node, common()->Start(new_output_count)); } break; } case IrOpcode::kParameter: { DCHECK(node->InputCount() == 1); // Only exchange the node if the parameter count actually changed. We do // not even have to do the default lowering because the the start node, // the only input of a parameter node, only changes if the parameter count // changes. if (GetParameterCountAfterLowering() != static_cast<int>(signature()->parameter_count())) { int old_index = ParameterIndexOf(node->op()); int new_index = GetParameterIndexAfterLowering(signature(), old_index); if (old_index == new_index) { NodeProperties::ChangeOp(node, common()->Parameter(new_index)); Node* new_node[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { new_node[i] = nullptr; } new_node[0] = node; if (signature()->GetParam(old_index) == MachineRepresentation::kSimd128) { for (int i = 1; i < kMaxLanes; ++i) { new_node[i] = graph()->NewNode(common()->Parameter(new_index + i), graph()->start()); } } ReplaceNode(node, new_node); } } break; } case IrOpcode::kLoad: { MachineRepresentation rep = LoadRepresentationOf(node->op()).representation(); const Operator* load_op; if (rep_type == SimdType::kInt32) { load_op = machine()->Load(MachineType::Int32()); } else if (rep_type == SimdType::kFloat32) { load_op = machine()->Load(MachineType::Float32()); } LowerLoadOp(rep, node, load_op); break; } case IrOpcode::kUnalignedLoad: { MachineRepresentation rep = UnalignedLoadRepresentationOf(node->op()).representation(); const Operator* load_op; if (rep_type == SimdType::kInt32) { load_op = machine()->UnalignedLoad(MachineType::Int32()); } else if (rep_type == SimdType::kFloat32) { load_op = machine()->UnalignedLoad(MachineType::Float32()); } LowerLoadOp(rep, node, load_op); break; } case IrOpcode::kStore: { MachineRepresentation rep = StoreRepresentationOf(node->op()).representation(); WriteBarrierKind write_barrier_kind = StoreRepresentationOf(node->op()).write_barrier_kind(); const Operator* store_op; if (rep_type == SimdType::kInt32) { store_op = machine()->Store(StoreRepresentation( MachineRepresentation::kWord32, write_barrier_kind)); } else { store_op = machine()->Store(StoreRepresentation( MachineRepresentation::kFloat32, write_barrier_kind)); } LowerStoreOp(rep, node, store_op, rep_type); break; } case IrOpcode::kUnalignedStore: { MachineRepresentation rep = UnalignedStoreRepresentationOf(node->op()); const Operator* store_op; if (rep_type == SimdType::kInt32) { store_op = machine()->UnalignedStore(MachineRepresentation::kWord32); } else { store_op = machine()->UnalignedStore(MachineRepresentation::kFloat32); } LowerStoreOp(rep, node, store_op, rep_type); break; } case IrOpcode::kReturn: { DefaultLowering(node); int new_return_count = GetReturnCountAfterLowering(signature()); if (static_cast<int>(signature()->return_count()) != new_return_count) { NodeProperties::ChangeOp(node, common()->Return(new_return_count)); } break; } case IrOpcode::kCall: { // TODO(turbofan): Make WASM code const-correct wrt. CallDescriptor. CallDescriptor* descriptor = const_cast<CallDescriptor*>(CallDescriptorOf(node->op())); if (DefaultLowering(node) || (descriptor->ReturnCount() == 1 && descriptor->GetReturnType(0) == MachineType::Simd128())) { // We have to adjust the call descriptor. const Operator* op = common()->Call(wasm::ModuleEnv::GetI32WasmCallDescriptorForSimd( zone(), descriptor)); NodeProperties::ChangeOp(node, op); } if (descriptor->ReturnCount() == 1 && descriptor->GetReturnType(0) == MachineType::Simd128()) { // We access the additional return values through projections. Node* rep_node[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { rep_node[i] = graph()->NewNode(common()->Projection(i), node, graph()->start()); } ReplaceNode(node, rep_node); } break; } case IrOpcode::kPhi: { MachineRepresentation rep = PhiRepresentationOf(node->op()); if (rep == MachineRepresentation::kSimd128) { // The replacement nodes have already been created, we only have to // replace placeholder nodes. Node** rep_node = GetReplacements(node); for (int i = 0; i < node->op()->ValueInputCount(); ++i) { Node** rep_input = GetReplacementsWithType(node->InputAt(i), rep_type); for (int j = 0; j < kMaxLanes; j++) { rep_node[j]->ReplaceInput(i, rep_input[j]); } } } else { DefaultLowering(node); } break; } case IrOpcode::kInt32x4Add: { LowerBinaryOp(node, rep_type, machine()->Int32Add()); break; } case IrOpcode::kFloat32x4Add: { LowerBinaryOp(node, rep_type, machine()->Float32Add()); break; } case IrOpcode::kCreateInt32x4: case IrOpcode::kCreateFloat32x4: { Node* rep_node[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { if (HasReplacement(0, node->InputAt(i))) { rep_node[i] = GetReplacements(node->InputAt(i))[0]; } else { rep_node[i] = node->InputAt(i); } } ReplaceNode(node, rep_node); break; } case IrOpcode::kInt32x4ExtractLane: case IrOpcode::kFloat32x4ExtractLane: { int32_t lane = OpParameter<int32_t>(node); Node* rep_node[kMaxLanes] = { GetReplacementsWithType(node->InputAt(0), rep_type)[lane], nullptr, nullptr, nullptr}; ReplaceNode(node, rep_node); break; } case IrOpcode::kInt32x4ReplaceLane: case IrOpcode::kFloat32x4ReplaceLane: { DCHECK_EQ(2, node->InputCount()); Node* repNode = node->InputAt(1); int32_t lane = OpParameter<int32_t>(node); DCHECK(lane >= 0 && lane <= 3); Node** rep_node = GetReplacementsWithType(node->InputAt(0), rep_type); if (HasReplacement(0, repNode)) { rep_node[lane] = GetReplacements(repNode)[0]; } else { rep_node[lane] = repNode; } ReplaceNode(node, rep_node); break; } default: { DefaultLowering(node); } } } bool SimdScalarLowering::DefaultLowering(Node* node) { bool something_changed = false; for (int i = NodeProperties::PastValueIndex(node) - 1; i >= 0; i--) { Node* input = node->InputAt(i); if (HasReplacement(0, input)) { something_changed = true; node->ReplaceInput(i, GetReplacements(input)[0]); } if (HasReplacement(1, input)) { something_changed = true; for (int j = 1; j < kMaxLanes; j++) { node->InsertInput(zone(), i + j, GetReplacements(input)[j]); } } } return something_changed; } void SimdScalarLowering::ReplaceNode(Node* old, Node** new_node) { // if new_low == nullptr, then also new_high == nullptr. DCHECK(new_node[0] != nullptr || (new_node[1] == nullptr && new_node[2] == nullptr && new_node[3] == nullptr)); for (int i = 0; i < kMaxLanes; ++i) { replacements_[old->id()].node[i] = new_node[i]; } } bool SimdScalarLowering::HasReplacement(size_t index, Node* node) { return replacements_[node->id()].node[index] != nullptr; } SimdScalarLowering::SimdType SimdScalarLowering::ReplacementType(Node* node) { return replacements_[node->id()].type; } Node** SimdScalarLowering::GetReplacements(Node* node) { Node** result = replacements_[node->id()].node; DCHECK(result); return result; } Node** SimdScalarLowering::GetReplacementsWithType(Node* node, SimdType type) { Node** replacements = GetReplacements(node); if (ReplacementType(node) == type) { return GetReplacements(node); } Node** result = zone()->NewArray<Node*>(kMaxLanes); if (ReplacementType(node) == SimdType::kInt32 && type == SimdType::kFloat32) { for (int i = 0; i < kMaxLanes; ++i) { if (replacements[i] != nullptr) { result[i] = graph()->NewNode(machine()->BitcastInt32ToFloat32(), replacements[i]); } else { result[i] = nullptr; } } } else { for (int i = 0; i < kMaxLanes; ++i) { if (replacements[i] != nullptr) { result[i] = graph()->NewNode(machine()->BitcastFloat32ToInt32(), replacements[i]); } else { result[i] = nullptr; } } } return result; } void SimdScalarLowering::PreparePhiReplacement(Node* phi) { MachineRepresentation rep = PhiRepresentationOf(phi->op()); if (rep == MachineRepresentation::kSimd128) { // We have to create the replacements for a phi node before we actually // lower the phi to break potential cycles in the graph. The replacements of // input nodes do not exist yet, so we use a placeholder node to pass the // graph verifier. int value_count = phi->op()->ValueInputCount(); SimdType type = ReplacementType(phi); Node** inputs_rep[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { inputs_rep[i] = zone()->NewArray<Node*>(value_count + 1); inputs_rep[i][value_count] = NodeProperties::GetControlInput(phi, 0); } for (int i = 0; i < value_count; ++i) { for (int j = 0; j < kMaxLanes; j++) { inputs_rep[j][i] = placeholder_; } } Node* rep_nodes[kMaxLanes]; for (int i = 0; i < kMaxLanes; ++i) { if (type == SimdType::kInt32) { rep_nodes[i] = graph()->NewNode( common()->Phi(MachineRepresentation::kWord32, value_count), value_count + 1, inputs_rep[i], false); } else if (type == SimdType::kFloat32) { rep_nodes[i] = graph()->NewNode( common()->Phi(MachineRepresentation::kFloat32, value_count), value_count + 1, inputs_rep[i], false); } else { UNREACHABLE(); } } ReplaceNode(phi, rep_nodes); } } } // namespace compiler } // namespace internal } // namespace v8