Compilers, Tools and Languages for Scientists

点击翻译按钮获取中文摘要

Computation and programming are becoming increasingly important aspects of the work done by scientists. However, scientists face many challenges and do not necessarily have the proper languages, compilers, and software tools to efficiently and safely accomplish their goals. A 2010 article in Nature, entitled "Why scientific programming does not compute", describes the increasing importance of programming, the challenges faced by scientists, and the consequences when the software developed is not up to a proper standard.****Other important current trends in programming languages and computer architectures also impact on this problem. Dynamic programming languages such as MATLAB, R and Python are becoming increasingly popular as application programming languages for scientists, and JavaScript is becoming increasingly popular as a way of programming on the web. On the architecture side, scientists have at their disposal a wide range of devices including workstations, laptops, and mobile devices, as well as access to many remote devices via the web. Furthermore, almost all devices now support some combination of multicore CPU and GPUs. ******The overall objective the proposed research is to develop new compiler techniques, software development tools, and domain-specific languages which will help scientists program effectively and which will make the best use of computer resources.  Our proposed research will build upon our recent work on the McLab project (www.sable.mcgill.ca/mclab), which has concentrated on compiler/VM tools and techniques, language extensions, and tools for MATLAB. As has been the case historically in our research, we will develop new ideas, implement these as open source projects, and test them on a variety of benchmarks from a wide variety of scientific disciplines.***We have identified five related research directions:  (1) we will support automatic translation of MATLAB programs to JavaScript, so as to allow Scientists to easily run their code on the web; (2) we will continue our work on effective optimization and code generation for multicore CPUs and GPUs; (3) we will leverage static and dynamic program analysis to develop programming tools for scientists, (4) we will continue our work on benchmarks for scientific computations, expanding to a variety of dynamic languages; and (5)  we will work on new domain-specific programming languages in order to ease the programming burden on scientists.******The proposed research is intended to be used by both compiler researchers (new techniques and open source tools), and scientist end-users. The research is novel in that it is targeting an increasingly important area of enabling programming for scientists, and in generating efficient code for both modern CPU/GPUs and web-based systems.**