Implementation of Multiple-Precision Floating-Point Arithmetic Library for GPU Computing

Takatoshi Nakayama and Daisuke Takahashi


multiple-precision arithmetic, GMP, GPU, CUDA


We have designed and implemented a fast multiple-precision floating-point arithmetic library designed to run on GPUs. The library is compatible with the GMP library and works with CUDA. We will show that multiple-precision arithmetic can be performed efficiently on GPUs by comparing the GPU performance with some of the currently available multiple-precision arithmetic libraries. Moreover, we will show that our library performs up to approximately 2.6 times better than GARPREC, which is a GPU-based multiple-precision arithmetic library based on earlier research on the NVIDIA Tesla C2050.

Important Links:

Go Back