Otter: bridging the gap between MATLAB and ScaLAPACK
Otter: bridging the gap between MATLAB and ScaLAPACK
复制标题
Otter：弥合 MATLAB 和 ScaLAPACK 之间的差距
DOI：
10.1109/hpdc.1998.709963
复制
发表时间：
1998
期刊：
影响因子：
--
通讯作者：
Nagajagadeswar Seelam
中科院分区：
文献类型：
--
作者：
M. J. Quinn;Alexey G. Malishevsky;Nagajagadeswar Seelam
我们已经开发了一个绰号为“Otter”的编译器，它将普通的MATLAB脚本翻译成面向支持ScaLAPACK和其他几个并行数值库的并行计算机的C程序。我们概述了多遍编译器的各个阶段，并讨论了运行库的设计和实现。我们给出了在四个不同的并行平台上编译的十个基准脚本所获得的性能，相对于在单个CPU上执行这些脚本的MATLAB解释器的速度。最后，我们说明了不同的数据分布和块大小如何影响编译代码的性能。
We have developed a compiler, nicknamed "Otter", which translates ordinary MATLAB scripts into C programs targeting parallel computers supporting ScaLAPACK and several other parallel numerical libraries. We outline the phases of the multi-pass compiler and we discuss the design and implementation of the runtime library. We present the performance achieved by ten compiled benchmark scripts on four distinct parallel platforms, relative to the speed of the MATLAB interpreter executing these scripts on a single CPU. Finally, we illustrate how different data distributions and block sizes can affect the performance of a compiled code.