![]() ![]() Which adopt a node to represent the execution of an instruction and adopt edges The first way is Data Dependence Graphs (DDG), The difference of DOALL, DOACROSS and DSWP techniques from the Fig.ĭependence graphs: There have been several ways to perform parallelism Pipelined parallelism from loops with arbitrary control flow. ( Bridges, 2008) is a more general PMT technique to extract PMT technique is DOPIPE ( Davies, 1981) which is restricted PMT also allows cross-threadĭependences but it only allows dependence of one direction. Technique is the major technique in this category. Which is suitable for general purpose applications allows cross-thread dependencesĪnd maintains correct execution dependences among threads by synchrony. Used to parallelize loops in array-based scientific programs. The typical IMT transformation is DOALL technique which usually Transformation prohibits cross-thread dependences and has restricted iteration Three categories techniques for extracting parallelism from loops of TLP. (IMT), Cyclic Multi-Threading (CMT) and Pipelined Multi-Threading (PMT) are Parallel types of thread-level parallelism: Independent Multi-Threading Some extended TM systems which support TLS techniques. This study presentsĪnd analyzes several thread partitioning techniques based on multi-core and Of sequential applications are the largest source of parallelism considerableĪttention on multi-core processors ( Hurson et al.,ġ997 Zhang et al., 2012a). Multi-core processors to speed single applications is a challenge. Therefore, how to using the hardware parallelism of Receive no benefits running on multi-core processors ( VachharajaniĮt al., 2007). That many important applications are single threaded and single-threaded applications Multi-core processors can improve system throughput and speed up The number of cores per chip is expected to rise inįuture, such as Intels 4-core Core i7, IBM 8-core Power 7, AMDsġ6-core Interlagos, Tileras 64-core Tile 64 and Intels 80-core Teraflops Multi-core processors are becoming ubiquitous with the continuous development Information Technology Journal, 12: 1673-1684. Parallel Techniques of the Sequential Codes Based on Multi-core. At last, some extended TM systems which support TLS techniques are analyzed from thread spawning mechanism, context passing mechanism and sequential ordering. SpecDSWP, SpecPS-DSWP and iSpecPS-DSWP are compared from supporting speculation types, memory version and implement steps. ![]() We introduced these speculative parallel techniques and described their execution. So many speculative techniques, such as thread level speculation (TLS), Speculation DSWP (SpecDSWP), Speculative PS-DSWP (SpecPS-DSWP) and Interprocedural SpecPS-DSWP (iSpecPS-DSWP), are proposed to break problematic dependences to enhance parallelism. These techniques can explore the parallel from sequential application, but much dependence is not easily predictable or manifests them infrequently by the non-speculative transformation. Typical DOALL, DOACROSS, DSWP and PS-DSWP techniques are described. In this paper, we introduced three parallel types of thread level parallelism. Addressing this problem, using parallel techniques improve the sequential program running on the multi-core processors. But many applications are sequential applications and they receive no benefits running on multi-core processors. Multi-core processors are becoming ubiquitous with the continuous development of hardware technology. Review Article Parallel Techniques of the Sequential Codes Based on Multi-core Information Technology Journal: Volume 12 (9): 1673-1684, 2013 ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |