Multi-core systems are optimized according to some high-level constraint criteria i.e., load distribution, timing constraint and data consistency. After achieving a well-balanced and fully functional multi-core system, a further performance gain is achievable by optimizing the allocation of data and code to core-local memories. This reduces the communication overhead on the shared BUS and requires lesser clock cycles to fetch the data from RAM/FLASH memories.
At Bosch we have developed a Local Core Memory Optimizer concept (LOCOMO) for optimized relocation of data and code according to core-access statistics. This talk shares our applied experience in Bosch productive multi-core systems.
Vorkenntnisse
No special know-how required
Lernziele
We want audience to understand following aspects
* Why local core memory optimizations are important
* Data and code allocation
* Problem with unoptimized allocation
* Benefits of low-level optimizations
//
Syed Aoun Raza
is working as Software Architect and multi-core system expert for Robert Bosch GmbH in Stuttgart. He has earned several years of experience in developing tools for model-driven software development. Additionally, he has research contributions in design and implementation of static analysis tools and techniques with focus on parallel/concurrent programs.
//
Christophe Kolb
is working on multi-core concepts and optimization tooling topics at Robert Bosch GmbH in Stuttgart. He has earned several years of experience in developing tools for model-driven software development.