SFE GCC should support "graphite" optimizations
GCC's "graphite" optimization options benefit code which loops through large arrays of data (e.g. scientific fortran code) and can benefit C code as well. SFE GCC should support these options. Enabling these options requires a number of exotic C++ libraries which are easy to build if you know how to find them, the order that they need to be built, and the special build options. I suggest building the libraries with the same compiler which is being installed.
I have attached notes regarding how I successfully did this under OpenIndiana for GCC 4.7.1. The same should be applicable to GCC 4.6.