On 23/02/2014 21:22, Bill Hart wrote: > I believe it will. However, the performance issues that Core2/Penryn > have do not exist on Nehalem/Sandy Bridge. It's surprising to me now > that those platforms were so bad compared to AMD. These days it seems to > be the other way around. > > Anyway, I'm pretty sure my code gets better the lower the mul > latency/throughput and the higher the memory bandwidth. So things should > be great on later Intel machines. But I'll have to check at some point. > > I hope to have access to a nehalem in a couple of days, and I will be > buying a haswell in a few weeks time.
For interest I ran the benchmark using Core 2 code with the old and new assembler on my Ivy Bridge Extreme system and the two versions came out almost identical in performance with the old code having a small advantage. But the Windows Core 2 assembler code for addmul_1 and submul_1 is not based on Jason's code so I would not expect to see the same results. Not surprisingly, the current Sandybridge code still gives the best results on Ivy Bridge (again addmul_1 and submul_1 are not translations of their GCC equivalents) Brian -- You received this message because you are subscribed to the Google Groups "mpir-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to mpir-devel+unsubscr...@googlegroups.com. To post to this group, send email to mpir-devel@googlegroups.com. Visit this group at http://groups.google.com/group/mpir-devel. For more options, visit https://groups.google.com/groups/opt_out.