[Updated on Apr. 18] Integration of CiNii Articles into CiNii Research


Bibliographic Information

Other Title
  • Hastega: Parallization of Linear Regression Using SIMD Instruction for Elixir Programming

Search this article



Python is a programming language used for software development of AI and machine learning programming as a de facto standard. However, it cannot process massive data from IoT because it is not parallelized in the language processor level. To solve this problem, we focus on a parallel programming language, Elixir. We have been researching and developing Hastega, which is a language processor to parallelize a Elixir code into SIMD instructions and/or GPGPU code, and libraries of AI and machine learning written in Elixir using it. In our previous works, we have already succeeded in getting the results that it have achieved to improve the speed of the integer operation benchmarks 4-8 times faster. This presentation reports that we optimize a linear regression program written in Elixir, convert it into native code including SIMD instructions of CPU by our hands, evaluate it in the aspect of performance, and get results that it improves the speed 2.3-2.6 times faster. One of our future works is to implement Hastega that compiles Elixir code into parallelized native code of SIMD instructions and/or GPGPU by the language processor.



Report a problem

Back to top