On 30 May 13:45, Jakub Jelinek wrote: > On Fri, May 30, 2014 at 03:41:22PM +0400, Kirill Yukhin wrote: > > That is definetely a bug and I see no compatibility issues in the fix. > > > > The only nit I see: maybe it'd be better to put this cpuid-less intrinsic > > into immintin.h? xmmintrin.h serves for SSE. > > Wouldn't that be an API compatibility problem? > I mean, xmmintrin.h header isn't a new header that can be only included > as part of immintrin.h or x86intrin.h, and people who include xmmintrin.h > and expect to find _mm_pause would suddenly get compiler errors. Makes sense. Moreover we already have cpuid-less _mm_prefetch () in that header. So, I take it back.
-- Thanks, K