sort by pipeline then mesh for non transparent passes for massively better batching #11671
Conversation
JMS55
added
A-Rendering
JMS55
added
the
S-Ready-For-Final-Review
|
Actually we should test additionally batching/sorting by bind group, either before or after the mesh. |
|
It seems like it should be good overall unless the frag overdraw cost is very expensive for a particular shader. So expensive custom shaders might suffer perf loss but it would almost certainly be mitigated by the author implementing a prepass frag shader and adding depth prepass, giving better perf overall even then. |
| @@ -201,7 +203,8 @@ impl PhaseItem for Opaque3d { | |||
|
|
|||
| #[inline] | |||
| fn sort_key(&self) -> Self::SortKey { | |||
| Reverse(FloatOrd(self.distance)) | |||
| // Sort by pipeline, then by mesh to massively decrease drawcall counts in real scenes. | |||
There was a problem hiding this comment.
Need to update this comment to mention bind groups. Also, was it not helpful to add bind group sorting for the prepasses?
There was a problem hiding this comment.
For the prepass and deferred it kept panicking on the unwrap when getting the material bindgroup.
We also may need to remove the unwrap for the opaque3d pass when getting the material bindgroup id, but I'm not sure if it makes sense that we'd be drawing things in the forward pass with no material bind group?
I'd like to remove the unwrap on getting the bind group id, but I'm not sure how the sorting would deal with a wrapped Option<NonZeroU32>.
There was a problem hiding this comment.
If you can't figure it out, I'd just add a TODO comment and we can do a followup PR.
For sorting, you can make a wrapper type implementing Ord, and then do sort_by(pipeline, Wrapper(maybe_bind_group), mesh).
JMS55
removed
the
S-Ready-For-Final-Review
|
I've reverted sorting by bind group ID as I don't think I'll have time to fix the unwrap, and I really want to get this in for 0.13. Sorting by the bindgroupid also probably needs more varied testing than just 2 scenes on my system, and should be pretty easy to add in a followup PR late. |
|
Sounds good to me. Can you open an issue and add it to the 0.14 milestone so that we don't forget please? |
JMS55
added
the
S-Ready-For-Final-Review
Head branch was pushed to by a user without write access
Elabajaba
force-pushed
the
sort_by_mesh
branch
from
964b6ac
to
264e533
Compare
…etter batching (bevyengine#11671) # Objective Bevy does ridiculous amount of drawcalls, and our batching isn't very effective because we sort by distance and only batch if we get multiple of the same object in a row. This can give us slightly better GPU performance when not using the depth prepass (due to less overdraw), but ends up being massively CPU bottlenecked due to doing thousands of unnecessary drawcalls. ## Solution Change the sort functions to sort by pipeline key then by mesh id for large performance gains in more realistic scenes than our stress tests. Pipelines changed: - Opaque3d - Opaque3dDeferred - Opaque3dPrepass  --- ## Changelog - Opaque3d drawing order is now sorted by pipeline and mesh, rather than by distance. This trades off a bit of GPU time in exchange for massively better batching in scenes that aren't only drawing huge amounts of a single object.
|
Any reason not to sort by distance within groups of the same mesh and pipeline? Seems like it would recover whatever performance this PR lost in scenes where there really are huge amounts of the same object. |
…#12117) # Objective - followup to #11671 - I forgot to change the alpha masked phases. ## Solution - Change the sorting for alpha mask phases to sort by pipeline+mesh instead of distance, for much better batching for alpha masked materials. I also fixed some docs that I missed in the previous PR. --- ## Changelog - Alpha masked materials are now sorted by pipeline and mesh.
…bevyengine#12117) # Objective - followup to bevyengine#11671 - I forgot to change the alpha masked phases. ## Solution - Change the sorting for alpha mask phases to sort by pipeline+mesh instead of distance, for much better batching for alpha masked materials. I also fixed some docs that I missed in the previous PR. --- ## Changelog - Alpha masked materials are now sorted by pipeline and mesh.
…bevyengine#12117) # Objective - followup to bevyengine#11671 - I forgot to change the alpha masked phases. ## Solution - Change the sorting for alpha mask phases to sort by pipeline+mesh instead of distance, for much better batching for alpha masked materials. I also fixed some docs that I missed in the previous PR. --- ## Changelog - Alpha masked materials are now sorted by pipeline and mesh.
Objective
Bevy does ridiculous amount of drawcalls, and our batching isn't very effective because we sort by distance and only batch if we get multiple of the same object in a row. This can give us slightly better GPU performance when not using the depth prepass (due to less overdraw), but ends up being massively CPU bottlenecked due to doing thousands of unnecessary drawcalls.
Solution
Change the sort functions to sort by pipeline key then by mesh id for large performance gains in more realistic scenes than our stress tests.
Pipelines changed:
Changelog