-
-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Added visibility bitmask as an alternative SSAO method #13454
Conversation
Welcome, new contributor! Please make sure you've read our contributing guide and we look forward to reviewing your pull request shortly ✨ |
References for reviewers: |
I added a few questions to answer before merging this and also added a linear sampler for the depth texture since I noticed it slightly improves some of the ghosting noticeable in small objects that happens due to the low resolution mip that gets selected for far-away samples. |
Another good reference: https://cybereality.com/screen-space-indirect-lighting-with-visibility-bitmask-improvement-to-gtao-ssao-real-time-ambient-occlusion-algorithm-glsl-shader-implementation/ @dragostis to give you an update, I'd still like to get this merged for Bevy 0.15 if you're interested. It's just going to be a very slow process, as reviewer bandwidth is very limited, and we have a large backlog of rendering PRs. I'll try to test this out in a few weeks when I'll have more free time, and help push it forward. |
@JMS55, I'm happy to get this merged for 0.15! In it's current form, it only handles AO. Adding GI as well is quite straightforward algorithm-wise, but would require extra textures and a separate denoise pass to look good. |
Yeah lets not do GI in this PR. Feel free to open another after if you're interested though. To get this PR merged, I want to test this on some more realistic scenes and check how the perf/quality compares. If it's always better than the existing method, then imo we should remove the existing method and maintain just this one. That's basically the only blocker. Unfortunately I don't have a desktop for the next week or two to be able to run Sponza and other scenes. There's also the opportunity here to change the denoiser/downscale depth/noise patterns that I copied from XeGTAO, and try what the newer vismask impls are using. Up to you on that, we can also leave the existing ones as-is. |
The code behind the vismask paper was open sourced (NOTE: Parts, including the denoiser, are from Unity and are not open-source licensed, be careful which files you reference): https://github.com/cdrinmatane/SSRT3 |
I'm still quite busy with meshlets, but I need to find some time sometime soon to try this PR out and see how it compares to GTAO. |
@JMS55, sorry the late reply. This month has been a bit busy, but I expect o have a bit more time to push this through in September. A good scene to test this IMO is Blender's Classroom since it has a lot of thin geometry that showcases VBAO's strengths quite well. |
No worries, like I said I haven't had much time lately either. Thanks, will try. |
crates/bevy_pbr/src/ssao/ssao.wgsl
Outdated
} | ||
|
||
#if METHOD == 0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd rather not use integers for this. Better to do #ifdef VBAO #else
imo.
crates/bevy_pbr/src/ssao/mod.rs
Outdated
@@ -189,6 +190,49 @@ impl ScreenSpaceAmbientOcclusionQualityLevel { | |||
} | |||
} | |||
|
|||
#[derive(Reflect, Clone, Copy, Debug)] | |||
pub enum ScreenSpaceAmbientOcclusionMethod { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This type and GTAO vs VBAO need docs.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry for the late review, but I finally had some time to look into this,
I tried it on Sponza and it looked good! Didn't have time to compare quality or performance to GTAO, or check the code thoroughly against the paper. I'll leave that up to other reviewers. VBAO seemed better at lower-sample counts especially though based on my very quick glances.
There's also probably a conversation to be had about if we want to support both methods, or use VBAO only. @dragostis it might help to post some example screenshots of the blender classroom comparing the two methods; I didn't have time to try and get the scene setup in bevy.
Side note to other reviewers - the SSAO docs say that it doesn't work on DirectX12, but I believe the naga issues should be fixed by now? We should test and remove the note if so.
Very nice! I'd be happy to remove GTAO and merge VBAO personally. I don't see a need for both. |
@dragostis is this ready to be reviewed now? If so, please take it out of draft :) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think I agree with @JMS55 on that one. We can probably just get rid of GTAO and just keep VBAO, it looks better in pretty much every case
@dragostis the release candidate for bevy 0.15 is in about a week. I'd like to get this in before then. Code is mostly done, I think we just need to remove GTAO, and some other smaller stuff. If you're busy, I'd be happy to finish this PR out for you. Let me know if you have time, otherwise I'll probably do this myself to ensure it makes the release in time. |
@JMS55, I'll try to address the feedback and un-draft the PR tomorrow morning. Thanks for the heads up! |
Awesome, thanks! Excited to finally get this in! Summary of needed changes (iirc, might be missing some):
|
f9f464d
to
0b25789
Compare
I've crossed everything off the list apart from the changelog. Since Bevy doesn't keep a changelog file, does this mean doing a writeup in the release blog post or is it something completely different? |
Nvm looks like we removed the changelog from our PR template in favor of just the migration guide. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the work! A few small suggestions, almost ready to go!
I think the conflicts are just caused by the required components PR. Should be trivial to resolve. |
Co-authored-by: JMS55 <[email protected]>
6ba8c3e
to
5f74252
Compare
This removes the need to use the entire SSAO object as a pipeline key, instead using only the quality level. This removes the need for hand-written Eq/Hash implementations.
Yep, this will get a writeup in the blog post. I'll be pinging you over in the bevy-website repo to get your advice on how to write it :) |
Early implementation. I still have to fix the documentation and consider writing a small migration guide.
Questions left to answer:
Eq
/Hash
forSSAOMethod
?Migration guide
SSAO algorithm was changed from GTAO to VBAO (visibility bitmasks). A new field,
constant_object_thickness
, was added toScreenSpaceAmbientOcclusion
.ScreenSpaceAmbientOcclusion
also lost itsEq
andHash
implementations.