In the last years, there has been much effort in commercial compilers (icc, gcc) to exploit efficiently the SIMD capabilities and the memory hierarchy that the current processors offer. However, the small numbers of compilers that can automatically exploit these characteristics achieve in most cases unsatisfactory results. Therefore, the programmers often need to apply by hand the optimizations to the source code, write manually the code in assembly or use compiler built-in functions (such intrinsics) to achieve high performance. In this work, we present source-to-source transformations that help commercial compilers exploiting the memory hierarchy and generating efficient SIMD code. Results obtained on our experiments show that our solutions achieve as excellent performance as hand-optimized vendor-supplied numerical libraries (written in assembly).
CitationBerna, A.; Jimenez, M.; Llaberia, J. "Vectorized register tiling". 2012.
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder. If you wish to make any use of the work not provided for in the law, please contact: firstname.lastname@example.org