Skip to content

egl: add a static trie-fetching implementation and use it for eglGetProcAddress()

Eric Engestrom requested to merge eric/mesa:trie into master

Verified with dEQP-EGL.functional.get_proc_address.* and an Intel CI run, but unfortunately I've been unable to measure any performance improvement for eglGetProcAddress().

Over 100'000 runs of eglGetProcAddress() for all the entrypoints in eglentrypoint.txt:

  • bsearch: 
    Average: 83.514 us
Mean: 73.888 us (±15.225 MAD)
Min: 72.068 us
Max: 171.644 us
  • trie: 
    Average: 82.345 us
Mean: 75.420 us (±7.586 MAD)
Min: 70.854 us
Max: 158.639 us

So basically no difference ☹️

Still, I did it and I think it's definitely nicer to use, so here goes.
Maybe other modules (gles?) could use this and would have better perfs?

Edit: I just realized that my distro is using GLVND, and I'm guessing that GLVND has its own eglGetProcAddress()? Which means I haven't been testing my own code at all?
I'll run my benchmark on another system later.

Edited by Eric Engestrom

Merge request reports