A Simple Guideline for Code Optimizations on Modern Architectures with OpenACC and CUDA

Ludomir Oteski (1) , Guillaume Colin-De-Verdière (2) , Sylvain Contassot-Vivier (3) , Stephane Vialle (4, 5, 6, 7) , Juliet Ryan (1)

1 ONERA - The French Aerospace Lab [Palaiseau]
2 DAM/DIF - DAM Île-de-France
3 SIMBIOT - SIMulating and Building IOT
4 Université Paris-Saclay
5 CentraleSupélec
6 Georgia Tech Lorraine [Metz]
7 ParSys - LRI - Systèmes parallèles (LRI)

Learn a simple strategy guideline to optimize applications runtime. The strategy is based on four steps and illustrated on a two-dimensional Discontinuous Galerkin solver for computational fluid dynamics on structured meshes. Starting from a CPU sequential code, we guide the audience through the different steps that allowed us to increase performances on a GPU around 149 times the original runtime of the code (performances evaluated on a K20Xm). The same optimization strategy is applied to the CPU code and increases performances around 35 times the original run time (performances evaluated on a E5-1650v3 processor). Finally, different hardware architectures (Xeon CPUs, GPUs, KNL) are benchmarked with the native CUDA implementation and one based on OpenACC.

Keywords: CUDA; guideline; OpenACC; Code optimization; Xeon-phi; GPU

Domains: Computer Science [cs]; Distributed, Parallel, and Cluster Computing [cs.DC]

Ludomir Oteski, Guillaume Colin-De-Verdière, Sylvain Contassot-Vivier, Stephane Vialle, Juliet Ryan. A Simple Guideline for Code Optimizations on Modern Architectures with OpenACC and CUDA. GPU Technology Conference 2017, NVIDIA, May 2017, San Jose, United States. ⟨hal-01527684⟩