4 ms·

Benchmarking the Linux Kernel with an “-O3” Optimized Build

• isaacfrond 4y ago
  tl;dr O3 doesn't make a lot of difference, and Linus doesn't like it. There was one outlier though a context switching stress test showed big improvements. The -O3 kernel being 270% better. These do not translate necessarily to real-world gains...
  • rbanffy 4y ago
    An outlier like that looks more like incorrect code being generated than performance gain from correct code. The PostgreSQL results though are interesting, if there was no data corruption ;-) It's not always you get an 11% boost from a software upgrade. Also, it'd be worth looking into other architectures other than x86 that may offer different results from the optimizations.
    • isaacfrond 4y ago
      I had the same feeling. A 200%+ speed improvement is too good to be true. There must be something wrong here.
• bick_nyers 4y ago
  I wonder if profile-guided optimization could give -O3 a more substantial edge and remove those pessimizations.