Automatic Generation of Highly Optimized Computer Code for Complex Many-Body Algorithms on High-Performance Computing Platforms (A02)

在高性能计算平台上自动生成复杂多体算法的高度优化的计算机代码（A02）

• 批准号: 537404418
• 金额: --
• 依托单位: Max-Planck-Institut für Kohlenforschung
• 依托单位国家: 德国
• 项目类别: Collaborative Research Centres
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/537404418?language=en
• 关键词: Automatic; Generation; Highly; Optimized; Computer

This project aims to achieve large-scale implementations of many-body electronic structure theories using automatic code generation. This development aims at addressing a number of pressing issues in the development of quantum chemical ab initio methods including legacy code, code of uncertain correctness, code of overwhelming complexity, and hardware adaptability. The main goal is to arrive at highly optimised codes for large scale computing platforms while reducing development times by many orders of magnitude or enabling the possibility of the implementation of theories that are too complex for humans. The proposed code generation chain consists of three steps: an equation generator, a factorizer for optimal generation of intermediates, and the actual code generator producing host- and hardware-specific high-level code.

本项目旨在利用自动代码生成实现多体电子结构理论的大规模实现。这一发展旨在解决量子化学从头算方法发展中的一些紧迫问题，包括遗留代码、不确定正确性的代码、压倒性复杂性的代码和硬件适应性。主要目标是为大规模计算平台提供高度优化的代码，同时减少许多数量级的开发时间，或者使实现对人类来说过于复杂的理论成为可能。所提出的代码生成链由三个步骤组成：一个方程生成器，一个用于最优生成中间体的因子生成器，以及生成特定于主机和硬件的高级代码的实际代码生成器。