https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90600
--- Comment #4 from g.peterh...@t-online.de --- Am 23.05.19 um 20:11 schrieb glisse at gcc dot gnu.org: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90600 > > --- Comment #3 from Marc Glisse <glisse at gcc dot gnu.org> --- > Intel documents that it uses "unsigned __int64" but I don't see where they > document what __int64 is. We could take a "void *out" argument and cast it > inside the function, but that would lose useful diagnostics for people trying > to pass a 32-bit type. We could overload in C++. Not sure any of that is worth > the trouble, those interfaces are target-specific anyway. > What else should "unsigned __int64" be than a uint64_t (0..2^64-1)? Then this would look exactly like this: external __inline uint8_t __attribute __ ((__ gnu_inline__, __always_inline__, __artificial__)) _addcarry_u64 (uint8_t __CF, uint64_t __X, uint64_t __Y, uint64_t * __P) { return __builtin_ia32_addcarryx_u64 (__CF, __X, __Y, __P); } And I miss addcarry/subborrow for uint8/16/128. You could make that available as a general __builtin :-) Of course it would be better if such functions are included in the C/C++ standard ...
