Use builtins for overflow math if possible

include/mruby/numeric.h

@@ -36,6 +36,47 @@ mrb_value mrb_num_div(mrb_state *mrb, mrb_value x, mrb_value y);
 # define MRB_INT_OVERFLOW_MASK ((mrb_uint)1 << (MRB_INT_BIT - 1))
 #endif
 
+/* Idea from Potion: https://github.com/perl11/potion (MIT) */
+#if (defined(__clang__) && ((__clang_major__ > 3) || (__clang_major__ == 3 && __clang_minor__ >= 4))) \
+    || (defined(__GNUC__) && __GNUC__ >= 5)
+
+static inline mrb_bool
+mrb_int_add_overflow(mrb_int augend, mrb_int addend, mrb_int *sum)
+{
+  mrb_bool of;
+
+#ifdef MRB_INT64
+  long long val;
+  of = __builtin_saddll_overflow(augend, addend, &val) ||
+#else
+  int val;
+  of = __builtin_sadd_overflow(augend, addend, &val) ||
+#endif
+  (val > MRB_INT_MAX) || (val < MRB_INT_MIN);
+
+  *sum = (mrb_int) val;
+  return of;
+}
+
+static inline mrb_bool
+mrb_int_sub_overflow(mrb_int minuend, mrb_int subtrahend, mrb_int *difference)
+{
+  mrb_bool of;
+
+#ifdef MRB_INT64
+  long long val;
+  of = __builtin_ssubll_overflow(minuend, subtrahend, &val) ||
+#else
+  int val;
+  of = __builtin_ssub_overflow(minuend, subtrahend, &val) ||
+#endif
+  (val > MRB_INT_MAX) || (val < MRB_INT_MIN);
+
+  *difference = (mrb_int) val;
+  return of;
+}
+#else
+
 static inline mrb_bool
 mrb_int_add_overflow(mrb_int augend, mrb_int addend, mrb_int *sum)
 {
@@ -56,6 +97,8 @@ mrb_int_sub_overflow(mrb_int minuend, mrb_int subtrahend, mrb_int *difference)
   return !!(((x ^ z) & (~y ^ z)) & MRB_INT_OVERFLOW_MASK);
 }
 
+#endif
+
 #undef MRB_INT_OVERFLOW_MASK
 #undef mrb_uint
 #undef MRB_UINT_MAKE