VM: Reland Inline instance object hash code into object header on 64bit.

runtime/vm/intrinsifier.cc

Index: runtime/vm/intrinsifier.cc
diff --git a/runtime/vm/intrinsifier.cc b/runtime/vm/intrinsifier.cc
index 951fd67cf1abe9b873a44995e5aba984a23b43bf..51d0e29ba441f56b28610f1a815e059a6f298b99 100644
--- a/runtime/vm/intrinsifier.cc
+++ b/runtime/vm/intrinsifier.cc
@@ -228,6 +228,17 @@ bool Intrinsifier::Intrinsify(const ParsedFunction& parsed_function,
     return compiler->intrinsic_slow_path_label()->IsUnused();
   }
 
+#if !defined(HASH_IN_OBJECT_HEADER)

[Vyacheslav Egorov (Google), 2017/06/30 16:47:26]

I think it is just okay to have an empty intrinsic on these platforms. Does
having empty intrinsic break anything?

[erikcorry, 2017/06/30 20:00:09]

I assumed it would break things.  The implementation on 32 bit is a native,
whereas on 64 bit it's intrinsified.

I guess you can't fall through from an intrinsified version to a native version.

+  // These two are more complicated on 32 bit platforms, where the
+  // identity hash is not stored in the header of the object.  We
+  // therefore don't intrinsify them, falling back on the native C++
+  // implementations.
+  if (function.recognized_kind() == MethodRecognizer::kObject_getHash ||
+      function.recognized_kind() == MethodRecognizer::kObject_setHash) {
+    return false;
+  }
+#endif
+
 #define EMIT_CASE(class_name, function_name, enum_name, type, fp)              \
   case MethodRecognizer::k##enum_name:                                         \
     compiler->assembler()->Comment("Intrinsic");                               \