emt64 performance

[pavel_jelinek]

Hi all,
I have question about the performance of different version of ifort compiler, namely l_fce_pc_8.1.027 vs. l_fc_pu_8.1.021. I have installed Suse 9.2 ,uname -a gives:
Linux pc087 2.6.8-24-smp #1 SMP Wed Oct 6 09:16:23 UTC 2004 x86_64

so emt64 should be a native platform. I've compile a code with both compilers ia-32 and emt64, but the 32-bit version runs faster by ~ 20% than emt64.
Let note, I used the same optimizations flags=-O3 -r8 -tpp7 -fno-alias -unroll. could anybody make a comment on this difference? Why does emt64 run slowly than 32-bit.

thanks in advance jel

[Steven_L_Intel1]

With our EM64T compiler, you MUST use one of the -x switches in order to get the best performance. If you are using an Intel processor, use -xP. If you are using an AMD processor, use -xW. You can also use -xW -axP to get code that will run well on both Intel and AMD.