【24h】

Efficient Memory Access Sequence Generation for Coupled Subscripts in Data-Parallel Programs

机译:数据并行程序中耦合下标的高效内存访问序列生成

获取原文
获取原文并翻译 | 示例

摘要

Generating the local memory access sequence is an important issue while compiling a data-parallel language into an SPMD code. Most studies of local memory access sequence presented to date are focused on one-dimensional arrays with CYCLIC(K) distribution onto one-dimensional processors or on multidimensional array with independent subscripts. But this problem becomes highly complicated if the subscripts contain the same induction variable, namely coupled subscripts. In this paper, we present an efficient method to compute the iterations executed on each processor by exploiting the feature of repetitive pattern of the memory accesses. Our method uses smaller iteration tables comparing to that in [10] and needs no table access cost at run-time. In order to compare with [10], extensive experiments were conducted. The results reveal the efficiency of our method, especially when the block size gets larger.
机译:在将数据并行语言编译为SPMD代码时,生成本地内存访问序列是一个重要的问题。迄今为止,对本地存储器访问序列的大多数研究都集中于将CYCLIC(K)分布到一维处理器上的一维数组或具有独立下标的多维数组。但是,如果下标包含相同的归纳变量,即耦合下标,则此问题变得非常复杂。在本文中,我们提出了一种有效的方法,可以通过利用内存访问的重复模式功能来计算在每个处理器上执行的迭代。与[10]中的方法相比,我们的方法使用较小的迭代表,并且在运行时不需要表访问成本。为了与[10]进行比较,进行了广泛的实验。结果揭示了我们方法的效率,特别是当块大小变大时。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号