China
Africa
Explore chapters and articles related to this topic
H.264/AVC Video Codec Design A Hardwired Approach
Published in Borko Furht, Syed Ahson, Handbook of Mobile Broadcasting, 2008
Tung-Chien Chen, Chung-Jr Lian, Yu-Wen Huang, Liang-Gee Chen
Software profiling is usually the first step to get more insights to algorithm complexity and to understand algorithm characteristics of computing and memory load/store behavior. It helps to find the critical parts of the whole design, and then we can spend more effort on its optimization. To focus on the target specification, here a software C model is developed by extracting all baseline profile compression tools from the reference software, and IPROF,11 an instruction-level profiler, is used for the complexity analysis.
Performance and Footprint at the Toolchain Level
Published in Ivan Cibrario Bertolotti, Tingting Hu, Embedded Software Development, 2017
Ivan Cibrario Bertolotti, Tingting Hu
This is a rather widespread approach in software design and development, in which an initial version of the code is produced first, in order to verify algorithm correctness. Then, possibly based on code inspection, benchmarking and profiling (when supported by the underlying target architecture) critical parts of the code are identified and optimized.
Parallel computing solutions for Markov chain spatial sequential simulation of categorical fields
Published in International Journal of Digital Earth, 2019
Weixing Zhang, Weidong Li, Chuanrong Zhang, Tian Zhao
Profiling enables us to find out the speed bottlenecks of a program (Gorelick and Ozsvald 2014). A computational analysis was made to profile the sequential simulation algorithm in order to identify the most time-consuming steps in executing the coMCRF model, so that those steps can be decomposed for domain parallelism or functional parallelism. A sequential simulation based on the coMCRF model was performed and profiled on a workstation (OS: Windows; CPU: Intel Xeon E3-1225 v3; Ram: 16GB), with a case study of 1000 × 1000 pixels for land cover post-classification. Similar to the findings in other studies (Vargas, Caetano, and Mata-Lima 2008; Pesquer, Cortés, and Pons 2011), our analysis finds that the quadrant nearest neighbor searching (NNS) and the sequential simulation steps consume most of the execution time (up to 95%) for generating a realization. The quadrant nearest neighbor searching and the random simulation cost approximate 43% and 52% of the total execution time, respectively. Compared to Vargas, Caetano, and Mata-Lima (2008) with respect to the time cost of the neighborhood search step, the quadrant nearest neighbor searching consumes a higher proportion of the execution time, because one quadrant neighborhood search needs to launch four independent spiral searches sequentially to identify the nearest neighbors within all quadrants. Overall, it is those two steps that limit the computational performance of the coMCRF model.