RT Conference Proceedings
T1 Techniques for Autotuning Algorithms on Heterogenous Platforms
A1 Diéguez, Adrián P.
A1 Amor, Margarita
A1 Doallo, Ramón
A2 Carretero Pérez, Jesús
A2 García Blas, Javier
A2 Petcu, Dana
AB Current GPUs (Graphic Processing Units) can obtain high computational performance in scientific applications.Nevertheless, programmers have to use suitable parallel algorithms for these architectures and have to consideroptimization techniques in the implementation in order to achieve that performance. This thesis is focused ondesigning and implementing parallel prefix algorithms into GPU architectures with little effort. For that, we havedeveloped a very optimized library called BPLG (Tuning Butterfly Processing Library for GPUs) and based on a setof building blocks that enable to easily design well-known algorithms such as FFT, tridiagonal systems solvers, scanoperator, sorting or signal processing. This library is designed under a tuning methodology based on two-stagesindentified as GPU resource analysis and operator string manipulation. Specifically, this strategy is focused on aset of parallel prefix algorithms that can be represented according to a set of common permutations of the digitsof each of its element indices [4], denoted as Index-Digit (ID) algorithms. So far, the proposed methodology hasobtained very good results with respect to state-of-art libraries, as CUFFT, CUSPARSE, CUDPP or ModernGPU.
SN 978-84-608-6309-0
YR 2016
FD 2016-02
LK https://hdl.handle.net/10016/22888
UL https://hdl.handle.net/10016/22888
LA eng
NO Proceedings of the First PhD Symposium on Sustainable UltrascaleComputing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016.
NO European Cooperation in Science and Technology. COST
DS e-Archivo
RD 17 jul. 2024