The following dissertation describes the design and creation of an OpenMP C compiler. The compiler targets the parallel capability of a modern series of NVIDIA graphics cards, and the generated output source code is written in the NVIDIA CUDA C language. The project also proposes and implements an extension to the OpenMP language: a single new directive and supporting runtime routines. By this route, problems with pointer indirection in the presence of a dual address space memory architecture are relieved.
|Qualification||Master of Science|
|Publication status||Published - 23 Aug 2007|
- parallel processing
- heterogeneous hardware