GCC 9 Compiler Tuning Benchmarks At Various Optimization Levels, Vectorize Options
Written by Michael Larabel in Software on 12 January 2019. Page 2 of 3. 78 Comments

For those experienced in compiler tuning or seeing our past compiler tuning benchmark results, the data shouldn't be much of a surprise.

Using "-ftree-vectorize -ftree-slp-vectorize" at -O2 did help boost the SciMark2 composite score by 17%. At -O3, those vectorize options are already enabled by default.

In SciMark2's Dense LU Matrix Factorization test, having "-O2 -ftree-vectorize -ftree-slp-vectorize" puts the performance up to where -O3 is performing.

But for those building your own packages, "-O3 -march=native" generally delivers the best performance unless hitting code issues or other snags.

In the case of programs like x264 that tend to already rely upon hand-tuned code, there isn't much of a difference beyond the most basic optimization levels.



Related Articles
Trending Linux News