On 4/12/2010 6:47 PM, Rayne wrote:
I mostly just want to know if an option is avaliable for the Nehalem processors. If it happens that the newest version of GCC has such an option, then I can always upgrade to that version. I believe that Barcelona is an AMD processor, so I'm confused as to how that would work for an Intel Nehalem processor. As for -mtune=nocona, I've tried and there was no performance gain.
Barcelona was the first of the recent CPUs which benefited consistently
from 128-bit moves, aligned or unaligned, and Nehalem continued the
trend. Hence the recommendation of the same option, if you are using
auto-vectorization. With current gcc, you could add -msse4.2 if you are
hoping occasionally to generate code specific to Nehalem.
