EsoErik

Saturday, November 14, 2009

 

Profiling and Rebuilding the Python 2.6.4 Interpreter for a 31% Speed Improvement on Windows x64

Coding in Python is sufficiently faster than coding in C++ that I do so wherever possible. Graphical interfaces and most file IO performs acceptably when implemented in Python; I use C++ only to implement the portions of algorithms that would increase application run time by an order of magnitude were they implemented in Python. Any increase in Python interpreter speed reduces the amount of C++ required to achieve acceptable performance.

The Intel x86-64 C++ compiler version 11.1 is expensive, but I've been pleased with the optimizations it delivers and its robust C++ support (I have written valid templated code that it compiles and MSVC 9 does not). Suspecting that I could significantly improve the Python interpreter's run times for some of my applications, I grabbed the Python 2.6.4 source and associate dependency sources. All of these I rebuilt with ICC 11.1 x64 targeting Core2 with MMX, SSE, SSE2, SSE3, SSSE3, and SSE4.1 enabled with /03 and global optimizations.

This resulted in ~15% run time reduction for my pure Python test application. Hoping for more improvement, I again rebuilt Python and its dependencies, this time with profile guided optimization logging enabled. I then re-ran my test cases and rebuilt again using the resulting logs. The resulting interpreter binaries execute my pure Python test case 31% faster vs the official distribution!

Labels: ,


Comments:

Post a Comment

Subscribe to Post Comments [Atom]





<< Home

Archives

July 2009   August 2009   September 2009   October 2009   November 2009   December 2009   January 2010   September 2010   December 2010   January 2011   February 2011   April 2011   June 2011   August 2011   February 2012   June 2012   July 2012   August 2012   October 2012   November 2012   January 2014   April 2014   June 2014   August 2014   September 2014   October 2014   January 2015   March 2015   April 2015   June 2015   November 2015   December 2015   January 2016   June 2016   August 2016   January 2017   March 2017  

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]