Haha
Maybe he'll fix SSAA in Dolphin 4.0
Maybe he'll fix SSAA in Dolphin 4.0
Dolphin has a new website
|
11-17-2012, 08:46 PM
So dolphin 4.0 will come after 3.0-999 or will we have 3.0-1000
......?????
11-17-2012, 09:00 PM
SSAA should already be fixed in efb-scaling-fixes. NaturalViolence is just too lazy to test.
11-17-2012, 10:26 PM
11-18-2012, 12:01 AM
(11-17-2012, 08:46 PM)Zee530 Wrote: So dolphin 4.0 will come after 3.0-999 or will we have 3.0-1000With the new AX HLE business going on I really hope they postpone the 4.0 release until AX HLE is fully working and implemented. Right now the latest versions of Dolphin is working great and will be good for a couple of more months. So no rush on my end. 11-18-2012, 10:51 AM
Since I am apparently the only one capable of testing SSAA I went ahead and finally did it (even on my crappy HD 4000).
This was a tricky one to analyze. It it fixed but still not where it should be. I won't post the screenshots here, only links, since with my uber screen resolution of 1920 x 1200 they would take up way too much space . Prior behavior: Before there was very little if any difference from adding 4xSSAA. Turning on 4xSSAA may or may not make the image sharpen but appears to have little or no effect on aliasing. 2x IR + 4xSSAA produced exactly the same image as 4xIR with no SSAA. Which should not happen because the filters/shaders should be different. How it should be: Exactly like every PC game and dolphin before r6549. Turning it on should not make the image sharper or blurrier but should eliminate aliasing. How it is: Turning it on reduces aliasing without making the image blurrier/sharper but does not reduce aliasing as much as it should. It is equivalent to what 2.25xSSAA (1.5x1.5) should look like. Clearly the filter could still use some work. It does not look as good as dolphin from before r6549 and it does not look as good as forcing OGSSAA using nvidia inspector (which yields a 10% performance hit even on high end cards). Comparison shots: 3.0-804 master, fract: http://i.imgur.com/2gDAc.jpg 3.0-804 master, fract, 4xSSAA: http://i.imgur.com/N27RZ.jpg As we can see with the master build the image is extremely blurry with fractional scaling, which should not be the case. Turning on 4xSSAA fixes the blur but does not fix the aliasing, in fact the aliasing actually looks slightly worse with it on. 3.0-781 efb scaling fixes, fract: http://i.imgur.com/ZhyYg.jpg 3.0-781 efb scaling fixes, fract, 4xSSAA: http://i.imgur.com/CxRdP.jpg With fractional the image is sharp, exactly as it should be. With 4xSSAA aliasing is reduced, but the effect is pretty small considering we have 4 samples per pixel. Notice that there are no additional intermediate shades of color between the polygon edges, which is what you would expect from the proper box filter that SSAA is supposed to use. Originally is was broken at r6549, here are some before and after shots. r6546, fract, 4xSSAA: http://i112.photobucket.com/albums/n189/...4xSSAA.png r6551, fract, 4xSSAA: http://i112.photobucket.com/albums/n189/...4xSSAA.png I'm sorry that these are using photobucket and 1366 x 768, that was my screen resolution back when I took these. As you can see in 6551 the image is sharp but there is loads of aliasing. in 6546 it is perfect. The edges are as smooth as they should be at that resolution with 4xSSAA. You can see the intermediate shades of color where the polygon edges are. This is what it is supposed to look like. With 1920 x 1200 it was perfect, so perfect I literally could not see any aliasing at all, whereas with efb scaling fixes I can still see quite a bit. It seems like what happened was some kind of horrible sampling bug was fixed considering that fractional was blurry before, but the filter still isn't doing what it should be doing. This issue is fixed but still needs a lot of work. I'm going to post some integral scaling and older build shots at some point to further investigate the issue.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."
-Ron Swanson "I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. " -Mark Antony 11-18-2012, 12:27 PM
While attempting to do some additional testing I discovered a bug in the latest efb scaling fixes build. You cannot set 4x IR. Set the IR to 4x and you can clearly see that your games are running at 1x IR. Setting IR to 3x works though.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."
-Ron Swanson "I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. " -Mark Antony
Thanks a lot for testing; the fractional-scaling-no-aa screenshots are also a fairly good example of why the branch was necessary to begin with.
Fwiw, could you do another small test for me? Do you spot any difference whatsoever between using 2xIR with 4xSSAA and 4xIR with no SSAA? Preferrably, enable the "auto size render window" setting (or w/e it's called). 11-20-2012, 03:43 AM
Can I ask why SSAA is used on D3D9 whereas MSAA is used on the other backends? I'd have thought that it would be easier to start from scratch with a better system than to hunt for a tiny bug which no one can find, in much the same way as Delroth decided the best way to fix HLE audio was to start from scratch.
OS: Windows 10 64 bit Professional
CPU: AMD Ryzen 5900X RAM: 48GB GPU: Radeon 7800 XT
MSAA isn't an option in d3d9 due to API restrictions. Plus, you really can't compare the HLE situation to anything because, frankly said, an incredibly good job has been done to make the old HLE code as unreadable and nonsensical as possible.
|
« Next Oldest | Next Newest »
|