Multi-target c code generation from MATLAB

• Tipo: Artigo em Livro de Atas de Conferência Internacional
• Data: 2014

Autores

João Bispo (Autor) FEUP; Reis, L (Autor) Outra; João M. P. Cardoso (Autor) FEUP

Ata de Conferência Internacional

Título: Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI)

Páginas: 95-100

1st ACM SIGPLAN International Workshop on Libraries, Languages and Compilers for Array Programming, ARRAY 2014 - Part of PLDI 2014

12 June 2014 through 13 June 2014

Indexação

Scopus - 6 Citações

Outras Informações

ID Authenticus: P-009-YTZ

DOI: 10.1145/2627373.2627389

Abstract (EN): This paper describes our recent work on MATISSE, a framework for MATLAB to C compilation. We focus on the new optimizations and transformations, as well as on OpenCL generation. MATISSE is controlled with LARA, an aspect-oriented language, able to specify transformations to the input MATLAB code (e.g., insertion of code for variable initialization and for monitoring) and to express information concerning types and shapes of variables. We evaluate the compiler with a set of benchmarks when targeting both an embedded system and a desktop system. The results show that we were able to achieve a speedup up to 1.8× by employing information provided by LARA aspects. We also compare the execution time of the generated C code with the original code running on MATLAB, and we achieve a geometric mean speedup of 19×. The geometric mean speedup reduces to 12× when optimizing the MATLAB code with LARA aspects. Finally, we present a preliminary version of a fully-functioning pragma-based OpenCL generator, built over the MATISSE framework..

Idioma: Inglês

Tipo (Avaliação Docente): Científica