APPLE: GPU frustum culling #2053
Closed
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
1 task
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
4 times, most recently
from
fa557df
to
2bec208
Compare
|
Filed as internal issue #USD-7669 |
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
from
2bec208
to
abb7404
Compare
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
from
abb7404
to
03cb935
Compare
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
from
03cb935
to
ed20e28
Compare
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
from
ed20e28
to
04451ef
Compare
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
5 times, most recently
from
e6d8c06
to
41ac528
Compare
2 tasks
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
from
41ac528
to
e939b02
Compare
creijon
force-pushed
the
jon/dev/culling_without_icb
branch
2 times, most recently
from
304bd1f
to
cc5a187
Compare
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
This is important for Hgi garbage collection. Contribution: Jon Creighton Fixes #2053 (Internal change: 2258046)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
Moved CPU view frustum culling to be an internal aspect of command buffer instead of split between command buffer and renderPass. Also cleaned up some out of date header includes. Contribution: Jon Creighton Fixes #2053 (Internal change: 2258054) (Internal change: 2258057)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
Fixes #2053 (Internal change: 2258193)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
Added support for generating pipeline shaders for CS in preparation for supporting GPU view frustum culling using compute. Instead of receiving drawing coord values as attribute inputs, generated CS shaders are set up to fetch drawing coordinate values directly from a dispatch buffer. The client CS needs to set the current draw index and current instance index before using any primvar accessors. Contribution: Jon Creighton, David G Yu Fixes #2053 (Internal change: 2258393)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
Implementation of GPU culling using compute shaders in the PipelineDrawBatch. The vertex-shader transform feedback approach that was used previously in the IndirectDrawBatch has not been supported in HGI since it would require multiple command buffers for synchronisation of the stages. The compute shader approach does not require a separate reset pass for instance culling since it can expand out the instances in a loop within the compute shader and populate the drawCommand buffers directly. Since the ICB changes have been merged into dev, this PR has been rebased on top of this. ICBs need to run in a separate batch, so they have been moved to the new Hd_DrawBatch::EncodeDraw method. Contribution: Jon Creighton Fixes #2053 (Internal change: 2258394)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
Currently, this affects the Storm Metal and Vulkan backends only and is guarded by HDST_ENABLE_PIPELINE_DRAW_BATCH_GPU_FRUSTUM_CULLING and the default is now set to true. View frustum culling for GL remains enabled by default using the MDI implementation. Contribution: Jon Creighton Fixes #2053 (Internal change: 2258398)
pixar-oss
pushed a commit
that referenced
this pull request
Jan 4, 2023
This provides a significant performance improvement for dispatching GPU frustum culling compute commands on Metal. Contributions: Jon Creighton Fixes #2053 (Internal change: 2258585)
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.
Description of Change(s)
Implementation of GPU culling using compute shaders in the PipelineDrawBatch. The vertex-shader transform feedback approach that was used previously in the IndirectDrawBatch has not been supported in HGI since it would require multiple command buffers for synchronisation of the stages. The compute shader approach does not require a separate reset pass for instance culling since it can expand out the instances in a loop within the compute shader and populate the drawCommand buffers directly.
Indexing into the drawCommands in the shader is done through a small emulation function in the codegen,
GetDrawingCoordField(), which has to be implemented in the kernel. This uses the thread ID to offset into the drawCommand buffer and access the encoded instance count, transform etc directly.Also, the
g_instanceIDparameter needs to be set in the culling kernel so that the codegen can access it inGetInstanceIndexCoord().For performance reasons, I've added a flag to specify that the compute encoder is created with concurrent dispatch.
To use the PipelineDrawBatch on OpenGL the option
HDST_ENABLE_HGI_RESOURCE_GENERATIONhas been changed to defaulttrueThere was also an issue around how the garbage collector was releasing resources outside of the HGI BeginFrame::EndFrame pair. This was causing a crash on OpenGL when picking a prim in the viewport. I've put a potential fix for this in
HdxPickTask.Since the ICB changes have been merged into dev, this PR has been rebased on top of this. ICBs need to run in a separate batch, so they have been moved to the new
Hd_DrawBatch::BeforeDrawmethod.Fixed the Windows build by using an
#undefon theMemoryBarrierdefine. I suggest we change the name of this function to avoid future issues.Fixes Issue(s)
Culling on instanced and non-instanced primitives on the GPU.
Tested on:
MacBook Pro M1 Max, with macOS Ventura
MacBook Pro i7 Radeon 560, with macOS Ventura
HP Xeon nVidia 3080, with Ubuntu 20.02
Update
This was originally submitted as PR: #2045
but was accidentally closed with a force-push. Reopened.