Wine 1.6 Released With 10,000+ Changes

[duby229]

Just missed Saturday by a couple hours... http://openbenchmarking.org/result/1...AH42-WINE16941
I thought it would be easier to read them, but I couldn't figure out how to convert the graphs to line-graphs. (I managed to do it locally, but when the results uploaded, they were bar graphs again).

Notes on these results:
3dmark & Shadermark have a performance drop at wine-1.5.9, due to the improved XRandR support: Prior to this version, they failed to recognise 1920x1080 as a valid resolution, and rendered at 1320x768 instead.
OpenGL performance is fairly consistent across all versions
DirectX performance increased for some applications, decreased for others.
PixelShaders: no performance change, except a small improvement at wine-1.5.28.

Oh wow!. That's awesome. Thanks for taking the time to provide these numbers. It's really good to see some valuable benchmarks of wine. Especially on this scale.

[guinea-pig]

It should be done running some time tomorrow evening, and if all is well, I'll upload the results to OpenBenchmarking on Saturday.

Just missed Saturday by a couple hours... http://openbenchmarking.org/result/1...AH42-WINE16941
I thought it would be easier to read them, but I couldn't figure out how to convert the graphs to line-graphs. (I managed to do it locally, but when the results uploaded, they were bar graphs again).

Notes on these results:
3dmark & Shadermark have a performance drop at wine-1.5.9, due to the improved XRandR support: Prior to this version, they failed to recognise 1920x1080 as a valid resolution, and rendered at 1320x768 instead.
OpenGL performance is fairly consistent across all versions
DirectX performance increased for some applications, decreased for others.
PixelShaders: no performance change, except a small improvement at wine-1.5.28.

[curaga]

Sure, but even if their binaries did run on my system (glibc versions etc), I'm not going to run a random binary off the web.

Testing in an Ubuntu VM instead would not help, as there have previously been bugs in Wine that are not visible under default Ubuntu, but only in my setup (no compositing, simple WM, etc). Whether this regression is one of them I don't know.

[GreatEmerald]

What you describe is manual bisecting, and no faster. It's in fact slower, since git can always halve the search space, while picking 1.4 would not do so.

It's faster if you don't need to compile. Not much faster, but still.

[dashcloud]

What you describe is manual bisecting, and no faster. It's in fact slower, since git can always halve the search space, while picking 1.4 would not do so.

Okay- wish there was a way to do it faster.

[curaga]

What you describe is manual bisecting, and no faster. It's in fact slower, since git can always halve the search space, while picking 1.4 would not do so.

[dashcloud]

Too bad the only thing I care about, win16 support, regressed somewhere between 1.3 and 1.6.

Also too bad that Wine is far too big to bisect at 15min a step. At 30 steps that's 7.5 hours of work.


I guess I'll stay on 1.3 and try again next year/1.8.

So, maybe it won't be as hard as you think. Wine does put out tarballs of all the releases, and since you know it worked with 1.3, I would check 1.4 to see if that release worked. If it does, you've narrowed down the search space a good deal, and could then try 1.5 to see if that works or not. If not, you can now just bisect 1.3 to 1.4.

Maybe you've already thought about that, and discarded it for good reasons, but if not, I'd consider that.

[curaga]

Too bad the only thing I care about, win16 support, regressed somewhere between 1.3 and 1.6.

Also too bad that Wine is far too big to bisect at 15min a step. At 30 steps that's 7.5 hours of work.


I guess I'll stay on 1.3 and try again next year/1.8.

[guinea-pig]

Benchmarks

I'm running a benchmark of the OpenGL/D3D stack for each release from the 1.5 development cycle. I know Stefan Doesinger has been running a bunch as well, but I wanted more. I'm using a lot of his tests, but with different settings. Specifically, I wanted to benchmark at settings (Full HD, etc) that would test a higher-end graphics card (GTX 660) than what he'd been testing against (GeForce 7600 GT), and would test the full series in one sitting.

It should be done running some time tomorrow evening, and if all is well, I'll upload the results to OpenBenchmarking on Saturday.

[synaptix]

Nice.

Also a big congrats to the Wine team for amazing work.