Unifying Primary Cache, Scratch, and Register File Memories in a Throughput Processor
Unifying Primary Cache, Scratch, and Register File Memories in a Throughput Processor
复制标题
DOI：
10.1109/micro.2012.18
复制
发表时间：
2012-12
期刊：
影响因子：
--
通讯作者：
Mark Gebhart;S. Keckler;Brucek Khailany;R. Krashinsky;W. Dally
中科院分区：
文献类型：
--
作者：
Mark Gebhart;S. Keckler;Brucek Khailany;R. Krashinsky;W. Dally
现代吞吐量处理器(如GPU)使用数千个线程来驱动高带宽、长延迟的存储系统。这些线程需要大量的片上存储空间来存储寄存器、高速缓存和便签存储器。现有设计对本地存储进行硬分区，在设计时固定这些结构的容量。我们评估了现代的GPU工作负载，发现它们在这些不同的功能上有很大不同的容量需求。因此，我们提出了一种统一的本地存储器，它可以在每个应用程序的基础上动态地改变寄存器、高速缓存和便签之间的分区。这种灵活性实现的调整提高了性能和能源消耗，并扩大了可以在GPU上高效执行的应用程序的范围。与硬分区设计相比，我们发现统一本地内存提供了高达71%的性能优势以及高达33%的节能。
Modern throughput processors such as GPUs employ thousands of threads to drive high-bandwidth, long-latency memory systems. These threads require substantial on-chip storage for registers, cache, and scratchpad memory. Existing designs hard-partition this local storage, fixing the capacities of these structures at design time. We evaluate modern GPU workloads and find that they have widely varying capacity needs across these different functions. Therefore, we propose a unified local memory which can dynamically change the partitioning among registers, cache, and scratchpad on a per-application basis. The tuning that this flexibility enables improves both performance and energy consumption, and broadens the scope of applications that can be efficiently executed on GPUs. Compared to a hard-partitioned design, we show that unified local memory provides a performance benefit as high as 71% along with an energy reduction up to 33%.