Hello,
Question regarding suitable optimization options in compiling a fortran code with Composer XE Update 11(344) in new machine: 64bit Intel Xeon E5-2687W 3.1GHz 8 cores. I have tried /Os, /O3, /Qxhost, /QxSSE2. For some reason, it ran slower than in 32 bit Intel Core 2 Duo T9550, 2.66 GHz machine.
Could you recommend optimization options for this Xeon configuration or possibily to exploit 8 cores other than OpenMP/MKL?
Thanks in advance for your help.
Sagar
Are you perhaps not taking into account the HyperThreading on the E5-2687? If you are going to run 8 OpenMP or Windows threads with auto-parallel or OpenMP but leave HT enabled, you should set KMP_AFFINITY so that 1 thread is pinned to each core. The alternatives of tbb and Cilk+ are available only with C and C++, so I doubt you really want to avoid OpenMP.
I think ifort /Os is equivalent to /O1, but I haven't seen it fully confirmed.
I don't think you are giving enough information to expect a better level of advice.
Thank you TimP.
These are my first iterations on new Xeon machine and I was hoping at least the same(or improved) level of speed with options like /Os, as in Intel Core 2 Duo T9550. But it was slower.. For example, the runtime was about 4 hours on Xeon as compared to 2.75 hours on Intel Core 2 Duo T9550 with /Os option.
I did try /Qparallel too but no improvement.
So, from your suggestion, OpenMP would improve the performance utilizing hyperthreading on E5-2687?
Thanks,
Sagar