On optimization techniques for the matrix multiplication on hybrid CPU+GPU platforms
Domingo Giménez Cánovas, Luis-Pedro García, Javier Cuenca · Dialnet (Universidad de la Rioja) · 2014
The use of auto-tuning techniques in a matrix multiplication routine for hybrid CPU+GPU platforms is analyzed. Basic models of the execution time of the hybrid routine and information obtained during its installation are used to optimize the execution time with a balanced assignation of the computation to the computing components in the heterogeneous system. Satisfactory results are obtained, with experimental execution times close to the lowest achievable.