The recent shift to multi-core computing has meant more programmers are required to write parallel programs. This is a daunting task even for experienced programmers due to the complexity and challenges involved in parallel programming. Key among these issues is managing concurrent accesses to shared memory. Unfortunately, most existing approaches rely on synchronization primitives that leave the complexity of protecting shared memory to the programmer and those that have attempted to automatically protect shared memory have achieved limited success. To address the issue of shared memory management while mitigating the limitations of existing approaches, we introduce a new technique called Synchronization via Scheduling (SvS). SvS provides efficient and automatic protection of shared memory by combining static and dynamic analysis to determine the set of possible memory accesses a block of code makes before it is executed and schedules these blocks such that no two blocks concurrently access the same memory.
展开▼
