PR #3351 introduced a performance regression

kirbypuff · 01-10-2016, 11:49 AM **#11**

(12-22-2015, 08:57 AM)kirbypuff Wrote: Fixed (!) in 4.0-8396...

... but the performance regression is back (with a vengeance) starting with 4.0-8634 [= the ffmpeg merge].

4.0-8631 is the last good version without the slowdown issue.

kirbypuff · (This post was last modified: 01-12-2016, 04:26 AM by kirbypuff.)

And it got fixed again in 4.0-8672? Interesting...
Definitely a weird issue with the automated build system.

It would be even more interesting to see if a "plain vanilla" VS2015u1 Win-x64 test build by someone else (not by the Dolphin auto-builder) fixes the huge performance regression mentioned in the other thread.
Ishiiruka doesn't have that issue, so it must be a certain compiler flag / switch / script / optimization that was changed/modified or even a stupid typo error.

**mbc07** · 01-12-2016, 04:09 AM **#13**

If it's really a performance regression why nobody else can reproduce it?

Looks like you're either paranoid or the one needing a bug fix, not Dolphin Dodgy

kirbypuff · (This post was last modified: 01-12-2016, 10:22 PM by kirbypuff.)

Update: Tested five different PR Test builds, all based on 4.0-8668 with trivial / mobile-only PRs that shouldn't affect the performance of the desktop version.

3 of them had the slowdown issue, 2 of them didn't.

**leolam** · 01-15-2016, 04:15 AM **#15**

But there wasn't any compiler change or flag change recently, as far as I know from looking in the commits… How about trying it with a real benchmark like the povray? Running the builds several times, perhaps on several computers, building yourself to confirm that there's a regression? As it stands, it's hard to believe that completely unrelated changes affect performance.