The goal of this project is to build and apply open source platforms for deep learning applications.
Develop algorithm library and sample library for artificial intelligence application software, integrate two kinds of applications (biological image big data analysis, genetic data analysis), and demonstrate on sugon advanced computing platform.
Complete the function development of algorithm library and sample library. The machine learning algorithm is developed to accelerate the core library, which is deployed and integrated into sugon advanced computing service platform. Based on the home-made hugon processor, the library is available for users to call. The accelerated core library will be closely coupled to the optimization of high-performance computer system at all levels.
Implement the algorithm tool set that supports the efficient parallel execution of large-scale machine learning and partially open source. The algorithm tool set should be integrated into sugon advanced computing platform to provide application services.
Responsible for the transplantation and integration of more than two kinds of applications (big data analysis of biological image and gene data analysis) into sugon advanced computing service platform, completing the support of typical application process, and reaching the leading level in terms of performance and scalability.
The original title of our paper is "Demystifying GPU Microarchitecture to Tune SGEMM". Now it is changed to "Understanding the GPU Microarchitecture to Achieve Bare-Metal Performance Tuning".
These codes are corresponding to section 3 (Instruction solver and GPU assembler KeplerAs), section 4 (SGEMM implementation and optimizations), and section 5 (SGEMM performance evaluation). Follow the instructions in each directory, and validate the functionality and performance results.