((overflow (a, b) \
|| (EXPR_SIGNED ((a) op (b)) && ((a) op (b)) < (tmin)) \
|| (tmax) < ((a) op (b))) \
- ? (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t, tmin, tmax), 1) \
- : (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t, tmin, tmax), 0))
-
-/* Return A <op> B, where the operation is given by OP. Use the
- unsigned type UT for calculation to avoid overflow problems.
- Convert the result to type T without overflow by subtracting TMIN
- from large values before converting, and adding it afterwards.
- Compilers can optimize all the operations except OP. */
-#define _GL_INT_OP_WRAPV_VIA_UNSIGNED(a, b, op, ut, t, tmin, tmax) \
- (((ut) (a) op (ut) (b)) <= (tmax) \
- ? (t) ((ut) (a) op (ut) (b)) \
- : ((t) (((ut) (a) op (ut) (b)) - (tmin)) + (tmin)))
+ ? (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t), 1) \
+ : (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t), 0))
+
+/* Return A <op> B, where the operation is given by OP. Return the
+ low-order bits of the mathematically-correct answer. Use the
+ unsigned type UT for calculation to avoid undefined behavior on
+ signed integer overflow. Assume that conversion to the result type
+ T yields the low-order bits in the usual way. UT and T are the
+ same width, T is two's complement, and there is no padding or trap
+ representations.
+
+ According to the C standard, converting UT to T yields an
+ implementation-defined result or signal for values outside T's range.
+ So, the standard way to convert UT to T is to subtract TMIN from
+ greater-than-TMAX values before converting them to T, and to add
+ TMIN afterwards, where TMIN and TMAX are T's extrema.
+ However, in practice there is no need to subtract and add TMIN.
+ E.g., GCC converts to signed integers in the usual way; see:
+ https://gcc.gnu.org/onlinedocs/gcc/Integers-implementation.html
+ All other known C compilers are similar to GCC in this respect.
+ Furthermore, Oracle Studio Studio 12.3 x86 has a bug when
+ implementing the standard way; see:
+ http://lists.gnu.org/archive/html/bug-gnulib/2017-04/msg00049.html
+
+ So, implement this operation in the usual way rather than in
+ the standard way. */
+
+#define _GL_INT_OP_WRAPV_VIA_UNSIGNED(a, b, op, ut, t) \
+ ((t) ((ut) (a) op (ut) (b)))
#endif /* _GL_INTPROPS_H */