Programming graphics processors with extensible languages

This paper presents a GPU programming system, NUDA, built on top of an extensilbe language, Nemerle. NUDA provides control over porting application to GPU, while making the compiler do the mechanical work. Macros and annotations simplify the porting process, and increase efficiency without sacrificing code size and readability. On a number of problems and architectures, use of NUDA increased performance several times compared to initial implementation.