-mfpmath=sse : This one tells the compiler to use the SSE capabilities of your chip for mathematical stuff. This options can make a different to *any* code that uses maths.. Its this that makes the 20% difference to my 32 bit cuilds. (you have to also include any one of -msse, -msse2 or -msse3 at the same time).
For more details, try a 'man gcc' and search for the section on -mfpmath. Chris
