CpuWriteGpuReadBelt for fast frame by frame memory transfers
#1382
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…g of alignment. Fix issues on render context shutdown
…t end of life of resource pool
teh-cmc
approved these changes
Feb 23, 2023
| } | ||
| } | ||
|
|
||
| fn poll_device(&mut self) { |
| ); | ||
| } | ||
|
|
||
| if self.0.user_data.len() < self.0.user_data.len() { | ||
| if self.0.user_data.len() < self.0.vertices.len() { |
There was a problem hiding this comment.
that caused me some reeeaaal weird crash :D
Wumpf
force-pushed
the
andreas/re_renderer/staging-belt2
branch
from
6dee277 to
e98525c
Compare
Wumpf
force-pushed
the
andreas/re_renderer/staging-belt2
branch
from
e98525c to
f193edb
Compare
emilk
pushed a commit
that referenced
this pull request
Mar 2, 2023
* Add CpuWriteGpuReadBelt and use it for frameuniform buffer and colors as poc * Limit number of in-flight queue submissions
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Rewrite of #594
Introduces a convenient mechanism to allow writing directly to gpu readable memory which gpu then can copy into its internally gpu-only memory.
Goal is faster (on less memcpy), safer (alignment guarantee) and easier (no need to create buffers manually) transfers.
An unexpected ripple effect we got was that in order to have much higher memory use on Web, we need to limit the number of frames (technically: queue submissions) in flight. This makes it a lot less memory hungry as we accumlate a lot less per-frame data! Empirically, it looks a bit like wgpu handles resource cleanup a bit differently on native than on web which might just come down to the present-mode it chooses automatically. Given that the introduced limiter &
device.pollmakes sense regardless I haven't investigated this further.Only used as a proof of concept in a single uniform buffer as well as point colors.
Nevertheless, some early perf numbers, need to repeat this test when we have it it more places:
(Acquired by selecting a small timeline range range and letting it loop, then opening profiler, waiting a bit, going to profiler, pause it, select a range of frames where the app was still in focus)
Btw. there is a deeper rabbit hole we ignore here: One could leverage integrated gpus (or the cache of dedicated ones) where the gpu would use the memory directly, eliminating the need to schedule a gpu memcpy, however this is behind a wgpu feature flag anyways (
MAPPABLE_PRIMARY_BUFFERS).Checklist
CHANGELOG.md(if this is a big enough change to warrant it)First half of #426