Suppress inlining of complex Bigint intrinsics.

runtime/vm/method_recognizer.h

Index: runtime/vm/method_recognizer.h
diff --git a/runtime/vm/method_recognizer.h b/runtime/vm/method_recognizer.h
index 0c4a9ec8169cf179964d20a80617ae5bb7e358e3..0379a71cc9d4c1b4653fb2b1ce60ba40db44a078 100644
--- a/runtime/vm/method_recognizer.h
+++ b/runtime/vm/method_recognizer.h
@@ -378,6 +378,14 @@ namespace dart {
   V(_Bigint, get:_digits, Bigint_getDigits, 1408062672)                        \
   V(_Bigint, set:_digits, Bigint_setDigits, 1135754410)                        \
 
+// A list of core function that should never be inlined.
+#define INLINE_BLACK_LIST(V)                                                   \
+  V(_Bigint, _absAdd, Bigint_absAdd, 233965936)                                \
+  V(_Bigint, _absSub, Bigint_absSub, 1401148862)                               \
+  V(_Bigint, _mulAdd, Bigint_mulAdd, 1283124653)                               \
+  V(_Bigint, _sqrAdd, Bigint_sqrAdd, 1665155090)                               \
+  V(_Bigint, _estQuotientDigit, Bigint_estQuotientDigit, 643982609)            \
+

[regis, 2014/11/06 17:25:19]

V(_Montgomery, _mulMod, Montgomery_mulMod, 149127024)
should be added as well. It is technically part of Bigint (used for modPow, e.g.
in RSA benchmark).

I'll add it in my next cl.

Little twist: no intrinsic for _estQuotientDigit is provided on arm or mips,
because no 64-bit by 32-bit division instruction exists on these platforms.
Prohibiting inlining is therefore not optimal in that particular case. Since a
runtime call will be used for the division, it probably does not matter.  

 // A list of core functions that internally dispatch based on received id.
 #define POLYMORPHIC_TARGET_LIST(V)                                             \
   V(_StringBase, [], StringBaseCharAt, 1512210677)                             \