GCC 4.7 Link-Time Optimization Performance
Written by Michael Larabel in Software on 21 August 2012. Page 2 of 3. 17 Comments

Starting off with the BYTE Dhyrystone 2 test, this computational test running on the Ivy Bridge system did benefit greatly from GCC's Link-Time Optimization support: the LTO-optimized binary was about 34% faster.

With the TTSIOD Renderer, there was surprisingly a small drop in performance with the LTO binary.

The Himeno binary wasn't able to benefit from GCC's current Link-Time Optimization feature.

With the build-php test profile we have a look at the compile time difference with and without GCC 4.7.1 LTO. With the additional optimizations across the entire PHP program, the compile time increased by 2.8x. While this increased compile time may be a problem if applying LTO to debug builds, the increased memory use and build times shouldn't be much of a restriction for release-ready widely-distributed software given the performance improvements yielded to some workloads by LTO.

C-Ray and Smallpt, lightweight single-file ray-tracing programs, didn't see any performance gains out of the LTO-optimized binaries to not much surprise.


Related Articles
Trending Linux News