Investigate Methods to Decrease Compilation Time-AX-Program Code Group Computer Science R& D Project

Cottom, T

Lawrence Livermore National Laboratory

关键词: Efficiency;    Productivity;    Computers;    Processing;    Programming;   

DOI  :  10.2172/15004130
RP-ID  :  UCRL-ID-153725
RP-ID  :  W-7405-ENG-48
RP-ID  :  15004130

美国|英语

来源: UNT Digital Library

【 摘 要 】

Large simulation codes can take on the order of hours to compile from scratch. In Kull, which uses generic programming techniques, a significant portion of the time is spent generating and compiling template instantiations. I would like to investigate methods that would decrease the overall compilation time for large codes. These would be methods which could then be applied, hopefully, as standard practice to any large code. Success is measured by the overall decrease in wall clock time a developer spends waiting for an executable. Analyzing the make system of a slow to build project can benefit all developers on the project. Taking the time to analyze the number of processors used over the life of the build and restructuring the system to maximize the parallelization can significantly reduce build times. Distributing the build across multiple machines with the same configuration can increase the number of available processors for building and can help evenly balance the load. Becoming familiar with compiler options can have its benefits as well. The time improvements of the sum can be significant. Initial compilation time for Kull on OSF1 was {approx} 3 hours. Final time on OSF1 after completion is 16 minutes. Initial compilation time for Kull on AIX was {approx} 2 hours. Final time on AIX after completion is 25 minutes. Developers now spend 3 hours less waiting for a Kull executable on OSF1, and 2 hours less on AIX platforms. In the eyes of many Kull code developers, the project was a huge success.