SAD Prefetching for MPEG4 Using Flux Caches
نویسندگان
چکیده
In this paper, we consider flux caches prefetching and a media application. We analyze the MPEG4 encoder workload with realistic data set in a scenario representative for the embedded systems domain. Our study shows that different well known data prefetch mechanisms can gain little reduction in the cache miss ratios when applied on the complete MPEG4 application. Furthermore, we investigate the potential improvement when dedicated prefetching strategies are applied to the sum of absolute differences (SAD) kernels in MPEG4. We propose a flux cache mechanism that dynamically invokes cache designs with dedicated prefetching engines that can fully utilize the available memory bandwidth. We show that our proposal improves the cache miss ratios by a factor close to 3x.
منابع مشابه
Stride-directed Prefetching for Secondary Caches
Thi s paper studies hardware pre fe tch ingfor second-level ( L 2 ) caches. Previous work o n prefetching has been extensive but largely directed a t p r imary caches. In some cases only L 2 prefetching i s possible or i s more appropriate. B y s tudying L2 prefetching characterist ics we show that existing stride-directed methods [l, 81 for L1 caches do no t work as well in L2 caches. W e prop...
متن کاملPattern-driven prefetching for multimedia applications on embedded processors
Multimedia applications in general and video processing, such as the MPEG4 Visual stream decoders, in particular are increasingly popular and important workloads for future embedded systems. Due to the high computational requirements, the need for low power, high performance embedded processors for multimedia applications is growing very fast. This paper proposes a new data prefetch mechanism c...
متن کاملThe Split replacement policy for caches with prefetch blocks
Prefetching is an inbuilt feature of file system and storage caches. The cache replacement policy plays a key role in the performance of prefetching techniques, since a miss occurs if a prefetch block is evicted before the arrival of the on-demand user request for the block. Prefetch blocks display spatial locality, but existing cache replacement policies are designed for blocks that display te...
متن کاملSecond-level Cache Organization for Data Prefetching
This paper studies hardware prefetching for second-level (L2) caches. Previous work on prefetching has been extensive but largely directed at primary caches. In some cases only L2 prefetching is possible or is more appropriate. We concentrate on stride-directed prefetching and study stream buuers and L2 cache prefetching. We show that proposed stride-directed organizations/prefetching algorithm...
متن کاملPerformance Issues in Integrating Temporality-Based Caching with Prefetching
This work evaluates the performance eeectiveness of combining two techniques for improving cache hit rate and reducing memory traac in small on-chip direct-mapped caches. Temporality-based caching is an eecient technique for reducing unnecessary cache block connicts in direct-mapped caches, but does not address compulsory misses. Tagged prefetching is a known technique for controlling compulsor...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2006