Handle breaks on keyed IC loads which can have an inlined version

src/ic-ia32.cc

Index: src/ic-ia32.cc
===================================================================
--- src/ic-ia32.cc	(revision 1755)
+++ src/ic-ia32.cc	(working copy)
@@ -729,8 +729,24 @@
 }
 
 
+// One byte opcode for test eax,0xXXXXXXXX.
+static const byte kTestEaxByte = 0xA9;
+
+
+bool KeyedLoadIC::HasInlinedVersion(Address address) {
+  Address test_instruction_address = address + 4;  // 4 = stub address
+  return *test_instruction_address == kTestEaxByte;
+}
+
+
+void KeyedLoadIC::ClearInlinedVersion(Address address) {
+  // Insert null as the map to check for to make sure the map check fails
+  // sending control flow to the IC instead of the inlined version.
+  PatchInlinedMapCheck(address, Heap::null_value());
+}
+
+
 void KeyedLoadIC::PatchInlinedMapCheck(Address address, Object* value) {
-  static const byte kTestEaxByte = 0xA9;
   Address test_instruction_address = address + 4;  // 4 = stub address
   // The keyed load has a fast inlined case if the IC call instruction
   // is immediately followed by a test instruction.
@@ -744,7 +760,7 @@
     // bytes of the 7-byte operand-immediate compare instruction, so
     // we add 3 to the offset to get the map address.
     Address map_address = test_instruction_address + offset_value + 3;
-    // patch the map check.
+    // Patch the map check.
     (*(reinterpret_cast<Object**>(map_address))) = value;
   }
 }