Move the stage that acquires a new texture from the swap chain as late as possible in the rendering pipeline. #16964
+61
−37
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
pcwalton
added
A-Rendering
as possible in the rendering pipeline. Currently, we retrieve a new texture from the pipeline in `prepare_windows`, which executes during the `ManageViews` rendering phase. This is inefficient if we're GPU bound, because `ManageViews` happens before `queue_material_meshes` and `batch_and_prepare_binned_render_phase` among other frequently-slow systems. Retrieving a new texture from the swap chain can block for considerable time, during which we don't want the CPU to be sitting idle. Additionally, we don't want the GPU to be sitting idle while the CPU works on `queue_material_meshes` and so forth. This commit fixes the issue by moving the acquisition of a new swap chain texture late in the pipeline, to a new phase `PrepareWindows` that occurs right before render graph evaluation. Because render graph evaluation needs to bind the swap chain textures in order to issue drawing commands, this is the last possible moment where it's safe to do so.
pcwalton
force-pushed
the
stop-gpu-idling
branch
from
608cda1 to
02602fc
Compare
|
Do we want this? Not sure how the synchronization is implemented in wgpu, but for DirectX12 I did this: https://github.com/JMS55/bevy_solari_directx/blob/main/src/swapchain.rs#L70-L76 Maybe @cwfitzgerald can weigh in on when the best time to block on the swapchain in wgpu is for good frame pacing / minimizing latency / maximizing FPS. |
|
I don't really understand why you'd want to not wait until the last possible moment. Right now the GPU is sitting idle for a long time and that makes no sense to me. |
|
To be clear, not having this definitely loses us FPS in Bistro if vsync is off. |
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.
Currently, we retrieve a new texture from the pipeline in
prepare_windows, which executes during theManageViewsrendering phase. This is inefficient if we're GPU bound, becauseManageViewshappens beforequeue_material_meshesandbatch_and_prepare_binned_render_phaseamong other frequently-slow systems. Retrieving a new texture from the swap chain can block for considerable time, during which we don't want the CPU to be sitting idle. Additionally, we don't want the GPU to be sitting idle while the CPU works onqueue_material_meshesand so forth.This commit fixes the issue by moving the acquisition of a new swap chain texture late in the pipeline, to a new phase
PrepareWindowsthat occurs right before render graph evaluation. Because render graph evaluation needs to bind the swap chain textures in order to issue drawing commands, this is the last possible moment where it's safe to do so.This shows Bistro on
mainwith shadows enabled before these changes. Note thatqueue_material_meshesandbatch_and_prepare_binned_render_phaseoccur afterprepare_windowsand therefore the GPU is idle during this time.This shows Bistro with these changes applied and shadows enabled. Observe that
queue_material_meshesandbatch_and_prepare_binned_render_phaseare moved beforeprepare_windows, allowing for simultaneous GPU/CPU execution.