rendervulkan: read capture back through a buffer on NVIDIA - #2315
Open
matte-schwartz wants to merge 1 commit into
Open
rendervulkan: read capture back through a buffer on NVIDIA#2315matte-schwartz wants to merge 1 commit into
matte-schwartz wants to merge 1 commit into
Conversation
On NVIDIA, the capture path's RGB->NV12 shader stores into the host-visible linear PipeWire texture go out at a few hundred MB/s, so one 5120x2160 frame takes ~65ms regardless of GPU clocks. paint_pipewire blocks on that work in the compositor thread, capping the whole session at ~15fps whenever recording or Remote Play is active. Keep capture images device-local and move them into a host-visible buffer with the copy engine. Both the proprietary driver and NVK map host-visible system memory GPU-uncached, so shader stores bypass the GPU L2 and cross PCIe as individual snooped transactions that nothing coalesces. The copy engine moves the same data in large bursts and does not share the problem. Key on the vendor ID since the behavior is common to both drivers, and this fix results in a >10x speedup for the capture path on my 5090.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
On NVIDIA, the capture path's RGB->NV12 shader stores into the host-visible linear PipeWire texture go out at a few hundred MB/s, so one 5120x2160 frame takes ~65ms regardless of GPU clocks. paint_pipewire blocks on that work in the compositor thread, capping the whole session at ~15fps whenever recording or Remote Play is active.
Keep capture images device-local and move them into a host-visible buffer with the copy engine. Both the proprietary driver and NVK map host-visible system memory GPU-uncached, so shader stores bypass the GPU L2 and cross PCIe as individual snooped transactions that nothing coalesces. The copy engine moves the same data in large bursts and does not share the problem. Key on the vendor ID since the behavior is common to both drivers, and this fix results in a >10x speedup for the capture path on my 5090.