Mesa OpenGL Threading Now Ready For Community Testing, Can Bring Big Wins

[shmerl]

Does this affect wine titles at all?

Yes, it does, but it can also collide with CSMT, which is a related idea. They can step on each other and make things worse. So test for yourself, each case can be different.

[smitty3268]

I am not insisting I am right about this, I just feel there's a strong possibility.

That's good, because you're wrong. That's not the way this works.

What we don't know and what I'd like to have proven is if these regressions are because of the games themselves or because of the hardware combination.

You can't prove a negative. Feel free to confirm for yourself that the hardware doesn't matter that much if you feel you need to.

I get the feeling someone could respond with a test on 20 different systems and you'd just claim that it's 1 game and doesn't prove other games aren't different.

If you can come up with a single test that proves you are correct, please let me know and I'll start taking you seriously.

[kenjitamura]

BTW, you might want to read up on Amdahl's Law - the more threads you add, the lower your potential maximum performance will be. In other words, as hardware continues to improve, GL threading has the potential to lower peak performance. In what may today run better because of it could regress on new hardware a couple years from now. This is why testing the regressions on various CPUs is important. Thats the only point I'm trying to make, and all you want to do is moan about "me me me".

I'm not sure I'm understanding your interpretation of Amdahl's law. I thought it was just saying the maximum performance gain possible by adding cores/threads is limited by the sequential operations within a computational workload. So if a game is made up of 50% sequential calls and 50% parallel calls no matter how many cores you throw at it the result will always infinitely approach halving the time of the operations without ever being faster than that.

I don't think adding threads lowers performance unless you're talking about the fact that the maximum single threaded performance is reduced by adding more cores because the overall frequency is lowered slightly to allow more cores without burning out the CPU. But because there are such massive diminished returns in the relationship between frequency and power/temperature above 4 GHz all CPU makers have universally decided it's smarter to lower the overall frequency by a few hundred Hz and add several more cores at the same frequency than it is to have fewer cores and keep pumping the frequency by tens of Hz with each hardware generation.

[dungeon]

This is just about simple Option's Law So, user enable this if it improve something or user disable it if it happen to be worse

[Zola]

@merek give a testing methodology to those wiling to help & forget the rest...

[shmerl]

@merek give a testing methodology to those wiling to help & forget the rest...

You can use GALLIUM_HUD, and for example measure the framerate.

[dungeon]

@merek give a testing methodology to those wiling to help & forget the rest...

GALLIUM_HUD=fps mesa_glthread=true appname

Watch fps before and after and then say what percentage increase or decrease you get:

https://percentagecalculator.net

Use third one there, increase/decrease bigest/lowest calculation

If you see any increase then follow what Marek said there:

I would like to ask those who use Mesa regardless of the hardware driver to
test their favourite games and nominate them for the whitelist if there is
an improvement. We would like the names of apps and their executable names.
This is what the drirc entry looks like for Bioshock.

        <application name="BioShock Infinite" executable="bioshock.i386">
            <option name="mesa_glthread" value="true"/>
        </application>

The first line of the entry is enough. Please send such lines for games
that get better performance from glthread to maraeo at gmail.com and I will
include them in Mesa with your name on the commit.

Basically he need as much as possible list of games that improve, so that he can add them to mesa as drirc profiles.

Just say how much percentage improvment you see and file that xml file with name of an GL app and its original exectutable name... if there is different executables for the same game, then send them all

Usually there is only one executable, but sometimes there is more usually 32bit and 64bit, but sometimes also something else might be, for example The Talos Principle had 'Talos' but also 'Talos_Unrestricted' executables... also game names on one service might differ on another, so same game might be named differently on Steam or GOG or ... They are mostly the same but sometimes different, for example Victor Vran game is 'VictorVran' on Steam but 'VictorVranGOG' on GOG but that is exception rather than a rule so user help is needed on all these names

Hopefully community would't be too much active, otherwise if Marek accept also exe names also and together with all duplicates he might get millions mails

[spstarr]

What this tells me is OpenGL is crap, but we knew that. The inconsistency that it has allowed developers/engines to create such hacks and a mess is why Vulkan must take over and soon, I just hope Vulkan forces developers/engine developers into conformance.

[mitch074]

Back to the question at hand: I'll test again with the games I own, nut up to now I've only noticed that while max FPS dropped with threading enabled, there was much less framerate variation with threading enabled. I did some tests both on a 4.2GHz Haswell i5 (thus still quite current), and a Athlon X4 620 (thus very old, but that goes from dual core @2GHz to quad core @3.2GHz), and the only "regression" I got was with Feral games (a 10% performance hit) with no benefit. AFAIK Feral implement a form of GL threading directly in game, but your results with Alien:Isolation seem to indicate that on shader heavy games your model is more efficient than theirs.
I'll try and see if nerfing my i5 allows me to test DeusEx:MD better on less-powerful CPUs and see if it's still true. I don't like testing this game though as I have huge slowdowns in Prague outdoors, that sometimes go away (on subsequent loads in the same gaming session). Mad Max in OpenGL mode might be worth a check too.

[dungeon]

What this tells me is OpenGL is crap, but we knew that. The inconsistency that it has allowed developers/engines to create such hacks and a mess is why Vulkan must take over and soon, I just hope Vulkan forces developers/engine developers into conformance.

Vulkan is crap also, we just need to wait to be widespreaded