[WebGPU] Implement dynamic buffer offsets for render bundles #19025
Conversation
|
EWS run on previous version of this PR (hash 8edcd4c) |
webkit-ews-buildbot
added
the
merging-blocked
mwyrzykowski
removed
the
merging-blocked
mwyrzykowski
changed the title
mwyrzykowski
force-pushed
the
eng/WebGPU-Implement-dynamic-buffer-offsets-for-render-bundles
branch
from
8edcd4c
to
a31d465
Compare
|
EWS run on previous version of this PR (hash a31d465)
|
|
|
||
| m_bindGroupDynamicOffsets.clear(); | ||
| } | ||
|
|
||
| void RenderBundleEncoder::draw(uint32_t vertexCount, uint32_t instanceCount, uint32_t firstVertex, uint32_t firstInstance) | ||
| { | ||
| if (id<MTLIndirectRenderCommand> icbCommand = currentRenderCommand()) { | ||
| executePreDrawCommands(); | ||
| executePreDrawCommands(); |
There was a problem hiding this comment.
Does this need to be computed for every draw, or can the calculations be cached?
There was a problem hiding this comment.
I think it needs to happen before every draw because the vertex buffers can change between draws.
And Metal ICBs appear to require all buffers in use to call setVertexBuffer: for each command, whereas WebGPU maintains a persistent state across commands
There was a problem hiding this comment.
which is best illustrated through https://webgpu.github.io/webgpu-samples/samples/renderBundles
passEncoder.setPipeline(pipeline);
passEncoder.setBindGroup(0, frameBindGroup);
// Loop through every renderable object and draw them individually.
// (Because many of these meshes are repeated, with only the transforms
// differing, instancing would be highly effective here. This sample
// intentionally avoids using instancing in order to emulate a more complex
// scene, which helps demonstrate the potential time savings a render bundle
// can provide.)
let count = 0;
for (const renderable of renderables) {
passEncoder.setBindGroup(1, renderable.bindGroup);
passEncoder.setVertexBuffer(0, renderable.vertices);
passEncoder.setIndexBuffer(renderable.indices, 'uint16');
passEncoder.drawIndexed(renderable.indexCount);
if (++count > settings.asteroidCount) {
break;
}
}
setBindGroup(0 is only called once and the expectation is that bind group is persistent throughout the draw calls
| auto commandCount = m_currentCommandIndex; | ||
| m_currentCommandIndex = 0; | ||
|
|
||
| if (!m_indirectCommandBuffer) { | ||
| m_icbDescriptor.maxVertexBufferBindCount = m_device->maxBuffersPlusVertexBuffersForVertexStage(); | ||
| m_icbDescriptor.maxVertexBufferBindCount = m_device->maxBuffersPlusVertexBuffersForVertexStage() + 1; |
There was a problem hiding this comment.
Could you a comment here why + 1?
There was a problem hiding this comment.
oh because the dynamic offsets get set to index = m_device->maxBuffersPlusVertexBuffersForVertexStage() and if I set them to m_device->maxBuffersPlusVertexBuffersForVertexStage() - 1 then I need to change the WGSL compiler.
But that is kind of ugly, maybe I can clean that up
mwyrzykowski
force-pushed
the
eng/WebGPU-Implement-dynamic-buffer-offsets-for-render-bundles
branch
from
a31d465
to
99a537c
Compare
|
EWS run on current version of this PR (hash 99a537c)
|
https://bugs.webkit.org/show_bug.cgi?id=262208 <radar://116137950> Reviewed by Dan Glastonbury. Support dynamic offsets in RenderBundles / ICBs like they are supported in the RenderPassEncoder and ComputePassEncoder. * Source/WebCore/Modules/WebGPU/Implementation/WebGPURenderBundleEncoderImpl.cpp: (WebCore::WebGPU::RenderBundleEncoderImpl::setBindGroup): * Source/WebGPU/WebGPU/RenderBundleEncoder.h: * Source/WebGPU/WebGPU/RenderBundleEncoder.mm: (WebGPU::addResource): (WebGPU::RenderBundleEncoder::executePreDrawCommands): (WebGPU::RenderBundleEncoder::draw): (WebGPU::RenderBundleEncoder::drawIndexed): (WebGPU::RenderBundleEncoder::drawIndexedIndirect): (WebGPU::RenderBundleEncoder::drawIndirect): (WebGPU::RenderBundleEncoder::finish): (WebGPU::RenderBundleEncoder::setBindGroup): (WebGPU::RenderBundleEncoder::setPipeline): (wgpuRenderBundleEncoderSetBindGroup): (wgpuRenderBundleEncoderSetBindGroupWithDynamicOffsets): * Source/WebGPU/WebGPU/WebGPUExt.h: Canonical link: https://commits.webkit.org/269548@main
webkit-commit-queue
force-pushed
the
eng/WebGPU-Implement-dynamic-buffer-offsets-for-render-bundles
branch
from
99a537c
to
32f0226
Compare
|
Committed 269548@main (32f0226): https://commits.webkit.org/269548@main Reviewed commits have been landed. Closing PR #19025 and removing active labels. |
webkit-commit-queue
removed
the
merge-queue
32f0226
99a537c