On Mon, Aug 21, 2017 at 1:48 PM, Xi Ruoyao <ryxi@xxxxxxxxxxxxxxxxx> wrote: > On 2017-08-21 19:41 +0800, Xi Ruoyao wrote: >> On 2017-08-21 07:31 -0400, Jeffrey Walton wrote: >> > On Mon, Aug 21, 2017 at 7:20 AM, Marc Glisse <marc.glisse@xxxxxxxx> wrote: >> > > On Mon, 21 Aug 2017, Jeffrey Walton wrote: >> > > >> > > > I'm trying to use _addcarryx_u64. GCC 7.1 is still generating ADD and >> > > > ADC instead of ADCX or ADOX: >> > > >> > > Does that compute wrong values? Make the program slower? >> > >> > It looks like the values are computed correctly. >> > >> > I don't know the affect on performance. The only benchmarks I have are >> > for add/adc. I can't get the adcx/adox generated to benchmark them. >> > >> > Compared to Intel ICC code generation, Intel's code is about 20% to 30% smaller. >> >> GCC once used ADCX but got silly results. So Uros disabled it. > > CC Uroš for his (or her) opinion. The compiler is not able to distingusih between OF and CF chains, since both are represented as a different mode of a single flags register. This is the limitation of the compiler. Uros.
