专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

1. 发明授权

US08863116B1 Pre-storing blocks for a streamed application in a local cache on a host computer system 有权
标题翻译：将流应用程序的块预存储在主机系统上的本地缓存中
公开(公告)号：US08863116B1
公开(公告)日：2014-10-14
申请号：US12274958
申请日：2008-11-20
申请人： Ryan M. McCarten , Taiwei Yin , Yury Mazin
发明人： Ryan M. McCarten , Taiwei Yin , Yury Mazin
IPC分类号： G06F9/445 , G06F15/16
CPC分类号： G06F9/445
摘要： A streamed application includes a plurality of application blocks. A first subset of the plurality of application blocks is pre-stored in a cache on a host computer system prior to initiating execution of the streamed application. The streamed application may then be executed. One or more application blocks of the first subset of the plurality of application blocks may be retrieved from the cache during execution of the streamed application in response to one or more requests to access the one or more application blocks of the first subset. A second subset of the plurality of application blocks may be received from a server computer system during the execution of the streamed application and may be stored in the cache. The second subset may include one or more application blocks that were not stored in the cache when the streamed application was installed.
摘要翻译：流应用包括多个应用块。多个应用程序块的第一个子集在开始执行流式应用程序之前被预先存储在主计算机系统的高速缓存中。然后可以执行流应用程序。响应于访问第一子集的一个或多个应用程序块的一个或多个请求，在执行流传输的应用程序期间，可从高速缓存中检索多个应用程序块的第一子集的一个或多个应用程序块。可以在流式应用的执行期间从服务器计算机系统接收多个应用块的第二子集，并且可以将其存储在高速缓存中。当安装流传输的应用程序时，第二子集可能包括一个或多个未存储在缓存中的应用程序块。

2. 发明授权

US06295545B1 Reduction of execution times for convolution operations 失效
标题翻译：减少卷积运算的执行时间
公开(公告)号：US06295545B1
公开(公告)日：2001-09-25
申请号：US09192394
申请日：1998-11-12
申请人： William Hsu , Taiwei Yin
发明人： William Hsu , Taiwei Yin
IPC分类号： G06F1710
CPC分类号： H03H17/06
摘要： A convolution operation such as an FIR filter determines sums of products by loading first and second sequences of values into a register file of a processor, each value being in a separate register in the register file. The processor determines a sum of products, for example, by executing a series on multiply-and-accumulate instructions on the previously loaded values from the register file. For a next sum, one or more values is loaded into the register file to replace some or all of the values from the second sequence. The first sequence of values is left intact in the register file for use in calculating the next sum. For an FIR filter, only a single value from the second sequence is replaced, and the next sum uses most of the previously loaded second sequence. This reduces the number of required memory accesses and reduces processing time required for convolution operations. A sequence of sums of products for the FIR filter requires loading of only one new value per filtered value determined.
摘要翻译：诸如FIR滤波器的卷积操作通过将第一和第二序列序列加载到处理器的寄存器文件来确定乘积的总和，每个值在寄存器文件中的单独的寄存器中。处理器确定产品的总和，例如，通过对来自寄存器文件的先前加载的值的乘法和累加指令执行一系列的操作。对于下一个总和，将一个或多个值加载到寄存器文件中以替换第二个序列中的一些或全部值。值的第一个序列在寄存器文件中保持不变，用于计算下一个和。对于FIR滤波器，仅替换来自第二个序列的单个值，并且下一个和使用大部分先前加载的第二个序列。这减少了所需的存储器访问的数量，并缩短了卷积操作所需的处理时间。 FIR滤波器的乘积和序列需要每个滤波值确定一个新的值。

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式