Comprehensive optimization of parametric kernels for graphics processing units

图形处理单元参数化内核的全面优化

• 批准号: 500717-2016
• 负责人: MorenoMaza, Marc
• 金额: $ 3.5万
• 依托单位: University of Western Ontario
• 依托单位国家: 加拿大
• 项目类别: Collaborative Research and Development Grants
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=618815
• 关键词: Comprehensive; optimization; parametric; kernels; graphics

This proposal deals with the optimization of computer programstargeting Graphics Processing Units (GPUs). The goal is to lift, fromprogrammers to optimizing compilers, the heavy burden of specifyingprogram and machine-specific details. The expected benefit is toensure robustness, portability and efficiency of the generatedcomputer programs.We address these requirements by treating machine and programparameters as unknown symbols during code generation, and generatingoptimized programs in the form of a case discussion, based on thepossible values of the machine and program parameters.This is a new approach to the design of optimizing compilers. Bytaking advantage of recent advances in the area of computer algebra,preliminary experimentation has demonstrated the merit of the aboveideas. With this proposal, we want to turn them into maturetechnology in support of commercial compilers such as the IBM XLcompiler.

该建议涉及计算机程序标记图形处理单元（GPU）的优化。目的是提起从程序员到优化编译器，指定程序和特定于机器的细节的重担。预期的好处是生成计算机程序的稳健性，可移植性和效率。我们通过在代码生成期间将机器和programparameters视为未知符号，并以案例讨论的形式将机器和程序讨论的形式产生的符号来解决这些要求，并基于机器和程序参数的可行值。计算机代数方面最新进展的综合优势，初步实验证明了上述阶段的优点。通过此提案，我们希望将它们变成成熟的技术，以支持IBM XLCOMPILER等商业编译器。