OpenCL implementation of the 3D CRS optimization algorithm
Paolo Marchetti, Alessandro Prandi, Bruno Stefanizzi, Herve Chevanne, Ernesto Bonomi, Antonio M. Cristini · 2011
The simultaneous computation of the eight parameters in the 3D ZO CRS formula is a very intensive computational task with a natural data parallelism that can be increased up to several thousands of computing units. In this paper we discuss this improvement as a key solution for an efficient use of computing resources, either with conventional units or with hybrid hardware for highly parallel applications, such as CRS. Another goal of this work was to address the migration of the computational kernel of the existing application, written in C/Fortran. The use of a language such as OpenCL makes now the CRS application portable even on future high-end computing systems with hybrid multicore technology.