共形计算:使用统一的方法在代数上连接硬件/软件边界,以实现软件和硬件应用程序的高性能计算
Conformal Computing: Algebraically connecting the hardware/software boundary using a uniform approach to high-performance computation for software and hardware applications
摘要:
我们提出了一个系统的,代数为基础的,设计方法的有效实施的计算机程序优化多层次的处理器/存储器和网络层次结构。使用一个共同的形式主义来描述问题和数据在处理器和存储器级别上的分区,允许人们在数学上证明给定算法的效率和正确性,如在一组度量(如处理器/网络速度等)方面测量的。这种方法允许普通程序员实现类似于编译器编写者所使用的高级优化(例如“平铺”的概念)。在这本专著中提出的方法利用数组数学(MoA,Mullin 1988)和索引演算(即PSI演算),使程序员能够通过代数组合和减少数组操作序列的能力,使用高级编译器式优化来开发算法。广泛的讨论和基准测试结果的快速傅立叶变换和其他重要的算法。
英文摘要:
We present a systematic, algebraically based, design methodology for efficient implementation of computer programs optimized over multiple levels of the processor/memory and network hierarchy. Using a common formalism to describe the problem and the partitioning of data over processors and memory levels allows one to mathematically prove the efficiency and correctness of a given algorithm as measured in terms of a set of metrics (such as processor/network speeds, etc.). The approach allows the average programmer to achieve high-level optimizations similar to those used by compiler writers (e.g. the notion of "tiling"). The approach presented in this monograph makes use of A Mathematics of Arrays (MoA, Mullin 1988) and an indexing calculus (i.e. the psi-calculus) to enable the programmer to develop algorithms using high-level compiler-like optimizations through the ability to algebraically compose and reduce sequences of array operations. Extensive discussion and benchmark results are presented for the Fast Fourier Transform and other important algorithms.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。