Architectural Optimizations in Multi-Core Processors

The quest for greater computational power is never-ending. Recently, the architectural trend has shifted from improving single-threaded application performance to improving multi-threaded application per-formance. Thus, multi-core processors have been increasingly popular. To achieve concurrent execution of threads on multi-core processors, applications must be explicitly restructured to exploit parallelism, either by programmers or compilers. However, conventional parallel pro-gramming models may introduce overhead due to synchronization and communications among threads in multi-threaded applications. This book presents three architectural optimizations to improve thread-based synchronization and communications support in multi-core processors. Register-Based Synchronization (RBS) uses hardware registers efficiently to provide synchronization support in multi-core processors. Prepushing is a software controlled data forwarding technique to provide communications support in multi-core processors. Software Controlled Eviction (SCE) improves shared cache communications by placing shared data in shared caches.