[turbofan]: Add a context relaxation Reducer

src/compiler/context-relaxation.cc

+// 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/context-relaxation.h"
+#include "src/compiler/js-operator.h"
+#include "src/compiler/node.h"
+#include "src/compiler/node-properties.h"
+
+namespace v8 {
+namespace internal {
+namespace compiler {
+
+Reduction ContextRelaxation::Reduce(Node* node) {

[Michael Starzinger, 2015/07/02 10:29:26]

I think it would make sense to rename this class to JSContextRelaxation and the
file to js-context-relaxation.cc, WDYT?

+  switch (node->opcode()) {
+    case IrOpcode::kJSCallRuntime:
+      switch (OpParameter<CallRuntimeParameters>(node->op()).id()) {
+        case Runtime::kInlineMathClz32:
+        case Runtime::kInlineMathFloor:
+        case Runtime::kInlineMathSqrt:
+          break;
+        default:
+          return NoChange();
+      }
+    // Fall through...
+    case IrOpcode::kJSCallFunction:
+    case IrOpcode::kJSToNumber:
+      if (!context_is_wired_) {

[danno, 2015/07/01 09:11:48]

This is a bit weird, any better ideas?

[Michael Starzinger, 2015/07/02 10:29:26]

As discussed offline: We could move the parameter into the JSGraph, thereby
graph trimming would not kill it. Beut we can do that in a follow-up CL.

+        if (relaxed_context_->InputAt(0) == nullptr) {

[Michael Starzinger, 2015/07/02 10:29:26]

Can we please use relaxed_context_->IsDead() instead?

+          relaxed_context_->ReplaceInput(0, start_);
+        }
+        context_is_wired_ = true;
+      }
+      NodeProperties::ReplaceContextInput(node, relaxed_context_);
+      return Changed(node);
+      break;
+    default:
+      break;
+  }
+  return NoChange();
+}
+
+}  // namespace compiler
+}  // namespace internal
+}  // namespace v8