Meta-programming Applied to Automatic SMP Parallelization of Linear Algebra Code

We describe a software solution to the problem of automatic parallelization of linear algebra code on multi-processor and multi-core architectures. This solution relies on the definition of a domain specific language for matrix computations, a performance model for multiprocessor architectures and its implementation using C++ template metaprogramming. Experimental results asses this model and its implementation on sample computation kernels.

• Type: Conference
• Year: 2008
• Where: EUROPAR
• Authors: Joel Falcou, Jocelyn Sérot, Lucien Pech, Jean-Thierry Lapresté
• Keywords: Distributed And Parallel Computing, EUROPAR 2008, Linear Algebra Code, Multi-core Architectures, Sample Computation Kernels