[optimizing] Fold HTypeConversion of constants

While looking into optimizing long shifts on x86, I found that the
compiler wasn't folding HTypeConversion of constants.  Add simple
conversions of constants, taking care of float/double values
with NaNs and small/large values, ensuring Java conversion semantics.

Add checker cases to see that constant folding of HTypeConversion is
done.

Ensure 422-type-conversion type conversion routiness don't get
inlined to avoid compile time folding.

Change-Id: I5a4eb376b64bc4e41bf908af5875bed312efb228
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index bdbd571..fcfedab 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -34,10 +34,15 @@
 // Binary encoding of 2^31 for type double.
 static int64_t constexpr k2Pow31EncodingForDouble = INT64_C(0x41E0000000000000);
 
+// Minimum value for a primitive integer.
+static int32_t constexpr kPrimIntMin = 0x80000000;
+// Minimum value for a primitive long.
+static int64_t constexpr kPrimLongMin = INT64_C(0x8000000000000000);
+
 // Maximum value for a primitive integer.
 static int32_t constexpr kPrimIntMax = 0x7fffffff;
 // Maximum value for a primitive long.
-static int64_t constexpr kPrimLongMax = 0x7fffffffffffffff;
+static int64_t constexpr kPrimLongMax = INT64_C(0x7fffffffffffffff);
 
 class Assembler;
 class CodeGenerator;