Use range information for optimizing integer boxing and fix bug in range analysis.

runtime/vm/object.h

Index: runtime/vm/object.h
===================================================================
--- runtime/vm/object.h	(revision 37592)
+++ runtime/vm/object.h	(working copy)
@@ -5026,14 +5026,20 @@
     return reinterpret_cast<intptr_t>(New(value));
   }
 
-  static bool IsValid(intptr_t value) {
-    return (value >= kMinValue) && (value <= kMaxValue);
+  template <typename T>
+  static bool IsValid(T value) {
+    COMPILE_ASSERT(sizeof(kMinValue) == sizeof(kMaxValue));
+    if (sizeof(value) < sizeof(kMinValue)) {

[Vyacheslav Egorov (Google), 2014/06/24 12:18:12]

We also need to check that it is integral type because on x64 sizeof(float) <
sizeof(intptr_t).

std::numeric_limits<T>::is_integer

[Florian Schneider, 2014/06/24 12:35:06]

Done. Added a COMPILE_ASSERT because we should never invoke it with a
non-integral type.

+      return true;
+    }
+    if (static_cast<T>(-1) < 0) {

[Vyacheslav Egorov (Google), 2014/06/24 12:18:12]

I would really prefer we write C++ here:

std::numeric_limits<T>::is_signed

[Florian Schneider, 2014/06/24 12:35:06]

Done.

+      return (value >= static_cast<T>(kMinValue))
+          && (value <= static_cast<T>(kMaxValue));
+    } else {
+      return value <= static_cast<T>(kMaxValue);
+    }
   }
 
-  static bool IsValid64(int64_t value) {
-    return (value >= kMinValue) && (value <= kMaxValue);
-  }
-
   RawInteger* ShiftOp(Token::Kind kind,
                       const Smi& other,
                       const bool silent = false) const;