FPC - configure the target CPU (for general instruction extensions, like POPCNT) and FPU (for vector instruction set, like SSE). For example, on x86, you have to pass the option -Cfsse2 to the compiler to use SSE2 for fp calculations. On x64, SSE2 is default, but if you want to use something higher, like AVX, you'll have to pass the option -CfAVX. Just don't have any huge expectations about the performance boost; SSE2 on 32bit helps quite a lot, but that's about it, until FPC gets autovectorization.
In addition, there are some standard library functions that use extensions if they are detected at program startup, like Move.