Convert SIMD wasm ops to runtime function calls

src/compiler/wasm-compiler.cc

 // Copyright 2015 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/wasm-compiler.h"
 
 #include "src/isolate-inl.h"
 
 #include "src/base/platform/elapsed-timer.h"
 #include "src/base/platform/platform.h"
 
 #include "src/compiler/access-builder.h"
 #include "src/compiler/common-operator.h"
 #include "src/compiler/diamond.h"
 #include "src/compiler/graph-visualizer.h"
 #include "src/compiler/graph.h"
 #include "src/compiler/instruction-selector.h"
 #include "src/compiler/int64-lowering.h"
 #include "src/compiler/js-generic-lowering.h"
 #include "src/compiler/js-graph.h"
 #include "src/compiler/js-operator.h"
 #include "src/compiler/linkage.h"
 #include "src/compiler/machine-operator.h"
 #include "src/compiler/node-matchers.h"
 #include "src/compiler/pipeline.h"
+#include "src/compiler/simd-lowering.h"
 #include "src/compiler/source-position.h"
 #include "src/compiler/zone-pool.h"
 
 #include "src/code-factory.h"
 #include "src/code-stubs.h"
 #include "src/factory.h"
 #include "src/log-inl.h"
 
 #include "src/wasm/ast-decoder.h"
 #include "src/wasm/wasm-module.h"
@@ skipping 568 matching lines @@
     case wasm::kExprF64Max:
       return BuildF64Max(left, right);
     case wasm::kExprF64Pow:
       op = m->Float64Pow();
       break;
     case wasm::kExprF64Atan2:
       op = m->Float64Atan2();
       break;
     case wasm::kExprF64Mod:
       return BuildF64Mod(left, right);
+    case wasm::kExprI32x4ExtractLane:
+      set_has_simd_ops(true);

[titzer, 2016/07/08 14:16:03]

Can you add a helper method called simd() that returns the MachineOperator
builder and also sets the has_simd_ops?

That way this code will be 

graph()->NewNode(simd()->Int32x4ExtractLane(), left, right);

And it will be harder to forget this in the future.
(also we should probably split the SIMD operations into their own
OperatorBuilder).

[gdeepti, 2016/07/11 09:50:34]

Added simd() helper method, a little fuzzy on what you mean when you say
OperatorBuilder. Do you mean an OperatorBuilder like
MachineOperatorBuilder/SimplifiedOperatorBuilder or just a method that builds
Simd ops? Added a SimdOp method that does not piggyback on Unop/Binop for now,
let me know if I've misunderstood. 

+      return graph()->NewNode(jsgraph()->machine()->Int32x4ExtractLane(), left,
+                              right);
     case wasm::kExprI32AsmjsDivS:
       return BuildI32AsmjsDivS(left, right);
     case wasm::kExprI32AsmjsDivU:
       return BuildI32AsmjsDivU(left, right);
     case wasm::kExprI32AsmjsRemS:
       return BuildI32AsmjsRemS(left, right);
     case wasm::kExprI32AsmjsRemU:
       return BuildI32AsmjsRemU(left, right);
     case wasm::kExprI32AsmjsStoreMem8:
       return BuildAsmjsStoreMem(MachineType::Int8(), left, right);
@@ skipping 255 matching lines @@
     case wasm::kExprI64SConvertF32:
       return BuildI64SConvertF32(input, position);
     case wasm::kExprI64SConvertF64:
       return BuildI64SConvertF64(input, position);
     case wasm::kExprI64UConvertF32:
       return BuildI64UConvertF32(input, position);
     case wasm::kExprI64UConvertF64:
       return BuildI64UConvertF64(input, position);
     case wasm::kExprGrowMemory:
       return BuildGrowMemory(input);
+    case wasm::kExprI32x4Splat:
+      set_has_simd_ops(true);
+      return graph()->NewNode(jsgraph()->machine()->CreateInt32x4(), input,
+                              input, input, input);
     case wasm::kExprI32AsmjsLoadMem8S:
       return BuildAsmjsLoadMem(MachineType::Int8(), input);
     case wasm::kExprI32AsmjsLoadMem8U:
       return BuildAsmjsLoadMem(MachineType::Uint8(), input);
     case wasm::kExprI32AsmjsLoadMem16S:
       return BuildAsmjsLoadMem(MachineType::Int16(), input);
     case wasm::kExprI32AsmjsLoadMem16U:
       return BuildAsmjsLoadMem(MachineType::Uint16(), input);
     case wasm::kExprI32AsmjsLoadMem:
       return BuildAsmjsLoadMem(MachineType::Int32(), input);
@@ skipping 1342 matching lines @@
     case wasm::kAstF64:
       return BuildChangeFloat64ToTagged(node);
     case wasm::kAstStmt:
       return jsgraph()->UndefinedConstant();
     default:
       UNREACHABLE();
       return nullptr;
   }
 }
 
+Node* WasmGraphBuilder::BuildChangeTaggedToInt32(Node* value) {
+  value = BuildChangeTaggedToFloat64(value);
+  value = graph()->NewNode(jsgraph()->machine()->ChangeFloat64ToInt32(), value);
+  return value;
+}
+
 Node* WasmGraphBuilder::BuildJavaScriptToNumber(Node* node, Node* context,
                                                 Node* effect, Node* control) {
   Callable callable = CodeFactory::ToNumber(jsgraph()->isolate());
   CallDescriptor* desc = Linkage::GetStubCallDescriptor(
       jsgraph()->isolate(), jsgraph()->zone(), callable.descriptor(), 0,
       CallDescriptor::kNoFlags, Operator::kNoProperties);
   Node* stub_code = jsgraph()->HeapConstant(callable.code());
 
   Node* result = graph()->NewNode(jsgraph()->common()->Call(desc), stub_code,
                                   node, context, effect, control);
@@ skipping 419 matching lines @@
     if (!mem_size_)
       mem_size_ = jsgraph()->RelocatableInt32Constant(
           size, RelocInfo::WASM_MEMORY_SIZE_REFERENCE);
     return mem_size_;
   } else {
     return jsgraph()->RelocatableInt32Constant(
         size + offset, RelocInfo::WASM_MEMORY_SIZE_REFERENCE);
   }
 }
 
+Node* WasmGraphBuilder::DefaultS128Value() {
+  Node* zero = jsgraph()->Int32Constant(0);
+  return graph()->NewNode(jsgraph()->machine()->CreateInt32x4(), zero, zero,
+                          zero, zero);
+}
+
 Node* WasmGraphBuilder::FunctionTable() {
   DCHECK(module_ && module_->instance &&
          !module_->instance->function_table.is_null());
   if (!function_table_) {
     function_table_ = HeapConstant(module_->instance->function_table);
   }
   return function_table_;
 }
 
+Node* WasmGraphBuilder::ChangeToRuntimeCall(Node* node,
+                                            Runtime::FunctionId function_id,
+                                            const Conversion* signature) {
+  SimplifiedOperatorBuilder simplified(jsgraph()->zone());
+  const Runtime::Function* function = Runtime::FunctionForId(function_id);
+  CallDescriptor* desc = Linkage::GetRuntimeCallDescriptor(
+      jsgraph()->zone(), function_id, function->nargs, Operator::kNoProperties,
+      CallDescriptor::kNoFlags);
+  const int kInputSize = 16;
+  const int kDefaultFunctionParams = 6;
+  Node* inputs[kInputSize + kDefaultFunctionParams];
+  DCHECK_LE(function->nargs + kDefaultFunctionParams,
+            static_cast<int>(arraysize(inputs)));
+  // Either there are control + effect or not.

[titzer, 2016/07/08 14:16:03]

When do runtime calls not have effect and control inputs?

[gdeepti, 2016/07/11 09:50:34]

Runtime calls always have effect and control, this check is for input count on
the node. For everything except loads and stores these are the start nodes, but
loads and stores (not implemented yet) will need additional inputs for effect
and control, hence the check. Would you prefer that I defer checking this till
Loads/Stores are added? 

+  DCHECK(node->InputCount() == function->nargs ||
+         node->InputCount() == function->nargs + 2);
+  int index = 0;
+  inputs[index++] = jsgraph()->CEntryStubConstant(function->result_size);
+  for (int i = 0; i < function->nargs; ++i) {
+    Node* arg = node->InputAt(i);
+    switch (signature[i + 1]) {
+      case Conversion::kInt32:
+        arg = BuildChangeInt32ToTagged(arg);
+        break;
+      case Conversion::kFloat32:
+        arg = jsgraph()->graph()->NewNode(
+            jsgraph()->machine()->ChangeFloat32ToFloat64(), arg);
+        arg = BuildChangeFloat64ToTagged(arg);
+        break;
+      case Conversion::kFloat64:
+        arg = BuildChangeFloat64ToTagged(arg);
+        break;
+      default:
+        break;
+    }
+    inputs[index++] = arg;
+  }
+  inputs[index++] = jsgraph()->ExternalConstant(
+      ExternalReference(function_id, jsgraph()->isolate()));
+  inputs[index++] = jsgraph()->Int32Constant(function->nargs);
+  inputs[index++] = jsgraph()->Constant(module_->instance->context);
+  // Loads and stores have control and effect, others do not and use
+  // the start node instead.
+  if (node->InputCount() == function->nargs + 2) {
+    inputs[index++] = node->InputAt(function->nargs + 1);  // effect
+    inputs[index++] = node->InputAt(function->nargs + 2);  // control
+  } else {
+    inputs[index++] = jsgraph()->graph()->start();  // effect
+    inputs[index++] = jsgraph()->graph()->start();  // control
+  }
+  Node* ret = jsgraph()->graph()->NewNode(jsgraph()->common()->Call(desc),
+                                          index, inputs);
+
+  Conversion return_type = signature[0];
+  switch (return_type) {
+    case Conversion::kInt32:
+      ret = BuildChangeTaggedToInt32(ret);
+      break;
+    case Conversion::kFloat32:
+      NodeProperties::SetType(ret, Type::Number());
+      ret = BuildChangeTaggedToFloat64(ret);
+      ret = jsgraph()->graph()->NewNode(
+          jsgraph()->machine()->TruncateInt64ToInt32(), ret);
+      break;
+    case Conversion::kFloat64:
+      ret = BuildChangeTaggedToFloat64(ret);
+      break;
+    default:
+      break;
+  }
+  return ret;
+}
+
 Node* WasmGraphBuilder::LoadGlobal(uint32_t index) {
   MachineType mem_type = module_->GetGlobalType(index);
   Node* addr = jsgraph()->RelocatableIntPtrConstant(
       reinterpret_cast<uintptr_t>(module_->instance->globals_start +
                                   module_->module->globals[index].offset),
       RelocInfo::WASM_GLOBAL_REFERENCE);
   const Operator* op = jsgraph()->machine()->Load(mem_type);
   Node* node = graph()->NewNode(op, addr, jsgraph()->Int32Constant(0), *effect_,
                                 *control_);
   *effect_ = node;
@@ skipping 609 matching lines @@
     }
     return nullptr;
   }
 
   if (machine->Is32()) {
     Int64Lowering r(graph, machine, common, jsgraph_->zone(), function_->sig);
     r.LowerGraph();
   }
 
   int index = static_cast<int>(function_->func_index);
+
+  // Run lowering pass if SIMD ops are present in the function
+  if (builder.has_simd_ops()) {
+    SimdLowering simd(&builder);
+    GraphReducer graph_reducer(jsgraph_->zone(), graph);
+    graph_reducer.AddReducer(&simd);
+    graph_reducer.ReduceGraph();
+
+    if (FLAG_trace_turbo_graph) {  // Simple textual RPO.
+      OFStream os(stdout);
+      os << "-- Graph after simd lowering -- " << std::endl;
+      os << AsRPO(*graph);
+    }
+  }
+
   if (index >= FLAG_trace_wasm_ast_start && index < FLAG_trace_wasm_ast_end) {
     OFStream os(stdout);
     PrintAst(isolate_->allocator(), body, os, nullptr);
   }
   if (FLAG_trace_wasm_decode_time) {
     *decode_ms = decode_timer.Elapsed().InMillisecondsF();
   }
   return source_position_table;
 }
 
@@ skipping 126 matching lines @@
                             function_->code_start_offset),
            compile_ms);
   }
 
   return code;
 }
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8