Programming Environments for Scientific Computation

目前，为超级计算机编程既困难又耗时。使用这些机器的科学家和工程师迫切需要一种可移植、可交互、灵活和方便的编程环境。有一个原型作为存在性证明，为构建一个有用的科学计算环境的可行性。这个软件工具被称为CLAMth（计算线性代数机），目前以Fortran 77编写的原型版本存在。虽然CLAM很成功，但我们认为它是超级计算机的“第一代”编程环境。使用一种现代编程语言(T)，我们建议建立一个“第二代”环境，它将：o接受程序员首选的语言，无论是T、CLAM还是Fortran。拥有一个解释性调试器，可以促进快速的原型设计。从工作站到大型机再到超级计算机完全可移植。利用最先进的算法进行大规模的数值分析任务。它将自动处理向量和矩阵数据结构，并包含“专家”能力，从自己的内部库中选择适当的算法，以使解决技术与问题的大小和特征相匹配。简而言之，这种环境将为面向应用的工程师或科学家提供任何特定计算的最快解决方案，而不需要用户决定要使用的适当算法和数据结构。---------------------------------------------------------------- 本提案的重点是开发的编程 科学家和工程师的工具。具体来说，该提案改进了与CLAM（先前开发的软件包）相关的编程语言，以改进编程环境。这个新版本的CLAM将有几个改进，如可移植性，自动并行化等，可以很容易地合并到其他语言生成的代码中。因此，36个月的提案将一次性资助如下：15万美元CSE (CISE) 5万美元Comp. Math (DMS) 5万美元新技术(DASC) --------- 25万美元36万美元。第一年全额付清。＊＊＊

Programming supercomputers is currently difficult and time consuming. Scientists and engineers who use these machines have an urgent need for a programming environment that is portable, interactive, flexible, and convenient. There is a prototype that serves as an existence proof for the feasibility of constructing a useful environment for scientific computing. This software tool, called CLAMth (the Computatinal Linear Algebra Machine), currently exists as a prototype version written in Fortran 77. Although CLAM is successful, we view it as a "first generation" programming environment for supercomputers. Using a modern programming language (T), we propose to build a "second generation" environment that will: oAccept the programmer's preferred language, be it T, CLAM, or Fortran. oHave an interpretive debugger that will promote fast prototyping. oBe fully portable from workstation to mainframe to supercomputer. oUtilize state-of-the-art algorithms for large scale numerical analysis tasks. It will handle vector and matrix data structures automatically and also contain an "expert" capability to select appropriate algorithms from its own internal library to match the solution techniques to the problem size and characteristics. In short, this environment will provide the applications- oriented engineer or scientists with the fastest possible solution for any particular computation, without requiring the user to make decisions about the appropriate algorithms and data structures to be employed. ---------------------------------------------------------------- This proposal focuses on the development of programming tools for scientists and engineers. Specifically the proposal refines the programming language associated with CLAM, a previously developed software package, for improved programming environments. This new version of CLAM will have several improvements such a portability, automatic parallelization, etc. and can be easily incorporated into code generated by other languages. Thus 36 month proposal will be funded all at one time as follows: $150.0k CSE (CISE) $ 50.0k Comp. Math (DMS) $ 50.0k New Tech (DASC) --------- $250.0k 36 mos. paid entirely in first year. ***