Unlike other platforms --mfpmath=sse is the default 
gcc mode on Mac OS X Intel. So it is unnecessary
to switch FPU to double precision mode (in fact,
it even breaks some math library functions).