基于CUPTI接口的GPU程序负载特征分析.pdfVIP

CUPTI GPU , , , ( , , 100084) Workload Analysis for Typical GPU Programs using CUPTI Interface 1 1 1 1 ZHENG Zhen , ZHAI Ji-Dong , LI Yan , CHEN Wen-Guang 1(Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China) Abstract : GPU-based high performance computers have become an important trend in the area of high performance computing. However, developing efficient parallel programs on current GPU devices is very complex. To address this problem, we summarize several key metrics that reflect programs performance. Then we design and implement a performance analysis tool based on underlying CUPTI interfaces provided by NVIDIA, which can analyze the performance behaviors of GPU programs effectively. Finally, we analyze 17 programs in Rodinia benchmark, which is a famous benchmark for GPU programs, and a real application using our tool. We summarize the common reasons for typical performance bottlenecks and give some high-level suggestions for developing efficient GPU programs. Key words: GPU; workload analysis; Rodinia : GPU GPU GPU NVIDIA CUPTI GPU GPU GPU Rodinia 17 GPU : GPU; ; Rodinia 1 2014 6 TOP500 48 GPU Graphic Processing Unit 15 5 GPU NVIDIA GPU GPU TOP500 48 GPU 46 NVIDIA GPU 2 AMD GPU NVIDIA GPU NVIDIA CUDA Compute Unified Device Architecture CUDA GPU global memory constant memory texture memory GPU shared memory CUDA grid thread block warp GPU GPU Zhang [1] GPU Yang Supported by the National Natural Science Foundation of China under Grant No. ); the National High-Tech Research and Development Plan of China under Grant No. 2012AA010901 ( (863)) : (1991 ) ,