A Data Cache with Dynamic Mapping
نویسندگان
چکیده
Dynamic Mapping is an approach to cope with a loss of performance due to cache interference and to improve performance predictability of blocked algorithms for modern architectures. An example is matrix multiply: tiling matrix multiply for a data cache of 16KB using optimal tiles size achieves an average data-cache miss rate of 3%, but with peaks of 16% due to interference. Dynamic Mapping is a software-hardware approach for which the mapping in cache is determined at compile time, by manipulating the address used by the data cache. The reduction in the cache misses translates into a 2-fold speed-up for matrix multiply and FFT by eliminating data-cache miss spikes. Dynamic mapping has the same goal as other proposed approaches, but it determines the cache mapping before issuing a load. It uses the computational power of the processor instead of the memory controller or the data cache mapping and it has no effect on the access time in memory and cache. It is an approach combining several concepts, such as nonstandard cache mapping functions and data layout reorganization and, potentially, without any overhead.
منابع مشابه
Dynamic Page Mapping Policies for Cache Conflict Resolution on Standard Hardware
In computer systems with large, physically-indexed, direct-mapped caches, a poor mapping from virtual to physical pages causes excessive cache connict misses. In a previous paper we proposed a simple hardware device , the Cache Miss Lookaside (CML) Buuer, which identiies pages that are suuering from connict misses. The operating system can use this information to implement a dynamic page mappin...
متن کاملMapping Policies for Cache Con ict
In computer systems with large, physically-indexed, direct-mapped caches, a poor mapping from virtual to physical pages causes excessive cache connict misses. In a previous paper we proposed a simple hardware device , the Cache Miss Lookaside (CML) Buuer, which identiies pages that are suuering from connict misses. The operating system can use this information to implement a dynamic page mappin...
متن کاملThe Performance And Power Impact Of Using Multiple Dram Address Mapping Schemes In Multicore Processors
Lowest-level cache misses are satisfied by the main memory by mapping the cache-line physical address to the corresponding set of memory cells that contain the data. This is done by identifying the rank, bank, row and column to start fetching the data from in bursts. The address mapping can happen through a specific devised scheme in the memory controller, that receives the request from the CPU...
متن کاملSegmented Bitline Cache: Exploiting Non-uniform Memory Access Patterns
On chip caches in modern processors account for a sizable fraction of the dynamic and leakage power. Much of this power is wasted, required only because the memory cells farthest from the sense amplifiers in the cache must discharge a large capacitance on the bitlines. We reduce this capacitance by segmenting the memory cells along the bitlines, and turning off the segmenters to reduce the over...
متن کاملDynamic thread mapping of shared memory applications by exploiting cache coherence protocols
In current computer architectures, the communication performance between threads varies depending on the memory hierarchy. This performance difference must be considered when mapping parallel applications to processor cores. In parallel applications based on the shared memory paradigm, the communication is difficult to detect because it is implicit. Furthermore, dynamic mapping introduces sever...
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 2003