Solari: Double buffer prepass textures #22102
Conversation
JMS55
added
C-Bug
JMS55
changed the title
JMS55
requested review from
IceSentry,
SparkyPotato,
alice-i-cecile,
atlv24 and
tychedelia
| camera_commands.remove::<DeferredPrepass>(); | ||
| } | ||
|
|
||
| if depth_prepass_double_buffer { |
There was a problem hiding this comment.
i hate that we still have to do this whole dance T_T
| let mut normal_textures = <HashMap<_, _>>::default(); | ||
| let mut deferred_textures = <HashMap<_, _>>::default(); | ||
| let mut deferred_textures1: HashMap<_, _> = <HashMap<_, _>>::default(); | ||
| let mut deferred_textures2: HashMap<_, _> = <HashMap<_, _>>::default(); |
There was a problem hiding this comment.
not a problem for this particular pr but kinda a bummer in render graph we can't easily reuse allocations via locals
| fn package_double_buffered_texture( | ||
| texture1: Option<CachedTexture>, | ||
| texture2: Option<CachedTexture>, | ||
| frame_count: u32, |
There was a problem hiding this comment.
kinda minor, but i would prefer if this were just an atomic on the node rather than requiring a dependency on diagnostics
There was a problem hiding this comment.
Yeah I suppose I can use a Local counter here... Technically it could get out of sync with FrameCount which other systems rely on for temporal stuff, but, should be fine?
There was a problem hiding this comment.
What do other people think?
There was a problem hiding this comment.
If a user enables double-buffering on an odd frame, would there be a desync? Maybe worth keeping a separate counter in that case but might not be a problem at all.
There was a problem hiding this comment.
Shouldn't be an issue idt. The order doesn't really matter.
I was more worried about what if we start running schedules at different rates, and different systems start thinking there's different frame counts.
pablo-lua
added
the
S-Needs-Review
| fn package_double_buffered_texture( | ||
| texture1: Option<CachedTexture>, | ||
| texture2: Option<CachedTexture>, | ||
| frame_count: u32, |
There was a problem hiding this comment.
If a user enables double-buffering on an odd frame, would there be a desync? Maybe worth keeping a separate counter in that case but might not be a problem at all.
ecoskey
added
S-Ready-For-Final-Review
4 participants
Fixes Solari crashing with
Partial copy of 0..1600 on X dimension with size 3200 is not supported for the Source texture format Depth32Float with 1 samplesafter #21746.Also slightly better performance by avoiding the copy.
Code is extremely cludgy though, we're just piling hacks upon hacks for the prepass texture stuff.