[Omp] OpenMP with different compilers

[Xiao Dong]

Good day,

I've been working on a Monte Carlo code paralleled with OpenMP. 
Compiling it with pgf90 simply gives a vergy good speed up of 2.0 on
dual processor AMD Opteron server; compiling it with xlf90_r somehow
gives a speed up of 1.3 on dual processor G5 server; while compiling
it with ifort, the worst thing happens - the speed up is 0.5 running
with 2 threads on SGI Altix (8-Itanium2).  What could be the problem? 
Supposedly, I've set the right environments on the Altix

KMP_STACKSIZE=1g
OMP_DYNAMIC=FALSE
KMP_LIBRARY=turnaround
OMP_NUM_THREADS=2


Tnanks for your comments!

Regards,