high performance computing on graphics processing units: hgpu.org

The need for faster computing has been around ever since the birth of the first computers. Faster hardware will almost always guarantee faster computing but occasionally the rate of hardware development is not enough for some programs to deal with the vast information they need. When these programs need to be accelerated, algorithmic optimizations have to be done that typically require changes to the program structure, in order to take advantage of parallel architectures, such as Graphics Processing Units (GPUs). Several frameworks have been developed to take advantage of the GPUs available parallelism. However, this typically requires major changes to the original program as well knowledge about the target GPU architecture. OpenACC is a recent technology which targets the simplification of this process by giving compiler hints about the parallelization strategy. This thesis targets the acceleration of an important bioinformatics application using OpenACC. Thus, two important results are taken: a) a performance comparison with CUDA; and b) a performance comparison with other parallel implementations of the program. Results show: that CUDA can have up to 2 times the performance of OpenACC regarding a single kernel, an overall 4.1 speed-up is achieved over the original serial MrBayes and that this implementation introduces some overhead when compared to the state of the art but scales much better for larger datasets than the latter.