As of now, it looks like they're ignoring it:

https://github.com/ran-j/PS2Recomp/blob/91678d19778891b4df85...

   #define FPU_ADD_S(a, b) ((float)(a) + (float)(b))

But if you wanted to handle it, you'd presumably macro expand the floating point operations to something that matches the PS2 fpu (or comes closer).