From a future perspective of heterogeneous multicore processors, we studied several optimization processes specified for floating-point numbers in this internship. To adjust this new generation processors, we believe {¥it split compilation} has an important role to receive the maximum benefits from these processors. A purpose of our internship is to know a potential of tradeoff between accuracy and speedup of floating-point computations as a prior-study to promote the notion of split compilation. In many case, we can more successfully optimize a target program if we are able to timely apply appropriate optimizations in dynamic. However, an online compiler cannot always apply aggressive optimizations because of its constraints like memory resources or compilation time, and the online compiler has to respect these constraints. To overcome these constraints, the split compilation can combine statically analyzed information and dynamic timely information. For this reason, in the bibliographic part, we mainly studied the several compilation method and floating-point numbers to smoothly move to our internship.