Improve performance #631
Improve performance #631
Conversation
Unpacking the components of the pixel only once per blend and delaying the divisions until all operations are finished gives a 25% speedup.
This results in a 10% speedup
|
I reviewed it and I like what I see. The performance improvements are the real deal. There are some things that I'd like to change, but we can accept the PR as-is right now, as the changes I have are pretty minor. |
|
Is there any problem preventing this PR ans the SPU's PR from being submitted? I'd like to see how many more FPS i can get. |
Maybe (: |
In regards to the SPU PR, I tested for audio artifacts and regressions for about 3 weeks but savestates probably need more testing as I don't use them. I have new hardware now so I'll do retesting sometime soon hopefully. Maybe if anyone is willing to test that branch with savestates it might get merged faster. Probably most valuable would be testing if very old savestates still work properly. EDIT: I guess I can test this PR later too but when I experimented with xbrz (as texture filter only) and deposterizing before there wasn't any noticeable difference in FPS and I never used xbrz as a filter for everything because fps was unusable and increasing internal res + xbrz textures looked better to me. |
These changes decrease the number of skipped frames reported in Order of Ecclesia from 23 to 9, and are mathematically equivalent to the original code.