Fix bug in X86 codegen: use GMOV instead of MOV for F64 w/o SSE2.