On Friday 27 February 2009 23:30:09 Jeff Gilchrist wrote: > On Thu, Feb 26, 2009 at 9:31 PM, Bill Hart <goodwillh...@googlemail.com> wrote: > > As far as we know MPIR 0.9.0 is highly stable and correct, so you > > shouldn't have any issues using it in production code. It is used in > > Sage for example, which has thousands of users. So far there have been > > no reported issues (except build issues on strange hardware, which you > > obviously haven't had). > > Good to hear that, I'm switching things over now. These extra > speedups are going to save me *days* of computing time in the long > run. For my real-word apps I'm seeing a 1.3x speedup with MPIR 0.9 on > 64bit AMD vs GMP 4.2.4 and between 1.2x and 1.7x speedup on 64bit > Core2. > > The squaring code with 105000 digit integers: > AMD Opteron 248 @ 2.2GHz (Linux 64bit) > MPIR 0.9 = 83m11.177s > GMP 4.2.4 = 108m41.370s > > Intel Xeon E5405 @ 2.00GHz (Linux 64bit) > MPIR 0.9 = 121m20.354s > GMP 4.2.4 = 208m40.348s > > > Using GMP-ECM to ECM a 244 digit number with B1=11e6: > AMD Opteron 248 @ 2.2GHz (Linux 64bit) > MPIR 0.9 = 2m32.406s > GMP 4.2.4 = 3m20.908s > > Intel Xeon E5405 @ 2.00GHz (Linux 64bit) > MPIR 0.9 = 3m45.447s > GMP 4.2.4 = 4m40.356s > > So thanks for all your hard work! With all the Core2 Windows stuff > happening after 0.9.0 I will probably wait until 1.0 to test out my > Windows binaries. > > Jeff. > >
This could make a nice "real world" benchmark for MPIR , where we have cache misses and branch mispredictions. The current mpirbench and speed don't have these features. For example if we pipeline the entire mul_basecase to increase the speed , it turns out to be 1.5x larger in the code cache. Now speed and mpirbench would probably not show any indication of this, whereas in the real world it may be a slowdown. I suppose we also have want two types of real world benchmark , one that stays the same and one that evolves along with MPIR. For example an ECM which uses mpz_mod full stop, and another one which uses mpz_mod and later on when MPIR gets mpz_mod_constant_divisor it uses that. Jason --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
