MIPS: Avoid embedding x86 NaN constant in code when snapshot is created.

Due to a different NaN representation on MIPS, the canonical nan
constant embedded in generated code that comes from snapshot will differ from
the one used on MIPS hardware.

TEST=mozilla/ecma/Array/15.4.4.2.js
BUG=
R=plind44@gmail.com

Review URL: https://codereview.chromium.org/150663009

Patch from Dusan Milosavljevic <Dusan.Milosavljevic@rt-rk.com>.

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@19118 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/mips/lithium-codegen-mips.cc

@@ -4222,8 +4222,8 @@ void LCodeGen::DoStoreKeyedFixedDoubleArray(LStoreKeyed* instr) {
 
     // Only load canonical NaN if the comparison above set the overflow.
     __ bind(&is_nan);
-    __ Move(double_scratch,
-            FixedDoubleArray::canonical_not_the_hole_nan_as_double());
+    __ LoadRoot(at, Heap::kNanValueRootIndex);
+    __ ldc1(double_scratch, FieldMemOperand(at, HeapNumber::kValueOffset));
     __ sdc1(double_scratch, MemOperand(scratch, instr->additional_index() <<
         element_size_shift));
     __ Branch(&done);