Нижегородский государственный университет им.Н.И.Лобачевского.


Факультет вычислительной математики и кибернетики

Лаборатория ITLabСеминар ЛабораторииСеминары 2007 г. Switch to English version  
О Лаборатории
Образовательные комплексы
Семинар Лаборатории
Семинары 2003 г.
Семинары 2004 г.
Семинары 2005 г.
Семинары 2006 г.
Семинары 2007 г.
Семинары 2008 г.
Семинары 2009 г.
Семинары 2010 г.
Семинары 2011 г.
Семинары 2012 г.
Семинары 2013 г.
Вакансии Интел
Разработчики сайта
О нас пишут
Летняя школа 2011
Видео лекции
Клуб У.М.Н.И.К.
Забыли пароль? Регистрация

Семинары 2007 г.

A new approach for optimizing multiply nested loop structures

The increasing complexity of hardware features for recent processors makes high performance code generation very challenging. One of the main issues for high performance is the optimization of memory accesses. General purpose compilers, with no knowledge of the application context and approximate memory model, seem inappropriate for this task. Combining application-dependent optimizations on the source code and exploration of optimization parameters as it is achieved with ATLAS, has been shown as one way to improve performance. Yet, hand-tuned codes such as in the MKL library still outperform ATLAS with an important speed-up and some effort has to be done in order to bridge the gap between performance obtained by automatic and manual optimizations.
We propose a new iterative compilation approach for the generation of high performance codes with compilers. This approach is not application-dependant, compared to ATLAS. The idea is to separate the memory optimization phase from the computation optimization phase. The first step automatically finds all possible decompositions of the code into kernels. With datasets that fit into the cache and simplified memory accesses, these kernels are easier to optimize, either with the compiler, at source level, or with a dedicated code generator. The best decomposition is then found by a model-guided approach, performing on the source code the required memory optimizations.
Exploration of optimization sequences and their parameters is achieved with a meta-compilation language, X language. The first results on linear algebra codes show that the performance obtained reduce the gap with those of highly optimized hand-tuned codes.
William JALBY was appointed Associate Professor at University of Rennes in 1987, then promoted Full Professor of Computer Science in 1991 and moved in 1992 to University of Versailles. His areas of research are: performance evaluation, code optimization, memory hierarchies and embedded processing. From 1987 to 1992, W. Jalby has been working closely with CSRD (CEDAR project, University of Illinois). More recently, he is collaborating with CEA DAM (French equivalent of Los Alamos) on performance evaluation and with BULL on code optimization for Itanium based SMP and with INTEL on tools for automating high performance library generation.

<< вернуться  |   Документ от: 23.09.2007 13:27



© ITLab, Нижний Новгород,  2009