[GH-ISSUE #3] Add support for zero-copy pipeline #3

New issue

Closed

opened 2026-10-15 12:47:16 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-15 12:47:16 +00:00

Owner

Originally created by @ABeltramo on GitHub (Jun 22, 2024).
Original GitHub issue: https://github.com/games-on-whales/gst-wayland-display/issues/3

In our pipeline we are currently doing one unnecessary GPU -> CPU copy in order to expose our raw frame buffer as a
gstreamer buffer. The rationale behind it is explained in
the docs.
This will cause some overhead that will eventually be a bottleneck when going with high resolutions and frame rates (on
Discord it has been reported that the current pipeline can't handle 1080p when going over 200 fps).

There are at least two ways to achieve this in Gstreamer:

OpenGL: exposing the buffer as
a GstGLBuffer, design docs
- We'll have to share the OpenGL context between our comp and the Gstreamer pipeline.
- I guess worst case this will end up doing GPU->GPU copy?
DMA-BUF: exposing the buffer as a GstBuffer backed by
a dmabuf, design docs
- Cuda doesn't seem to support this directly, (vaapi should) it should be possible to import DMABuf
  using glupload -> cudaupload.
- I guess worst case this will end up doing GPU->GPU copy?

This needs more research...

Useful links

Originally created by @ABeltramo on GitHub (Jun 22, 2024). Original GitHub issue: https://github.com/games-on-whales/gst-wayland-display/issues/3 In our pipeline we are currently doing one unnecessary GPU -> CPU copy in order to expose our raw frame buffer as a gstreamer buffer. The rationale behind it is explained in the [docs](https://games-on-whales.github.io/wolf/stable/dev/wayland.html#_gstreamer). This will cause some overhead that will eventually be a bottleneck when going with high resolutions and frame rates (on Discord it has been reported that the current pipeline can't handle 1080p when going over **200 fps**). There are at least two ways to achieve this in Gstreamer: - **OpenGL**: exposing the buffer as a `GstGLBuffer`, [design docs](https://gstreamer.freedesktop.org/documentation/additional/design/opengl.html) - We'll have to share the OpenGL context between our comp and the Gstreamer pipeline. - I guess worst case this will end up doing GPU->GPU copy? - **DMA-BUF**: exposing the buffer as a `GstBuffer` backed by a `dmabuf`, [design docs](https://gstreamer.freedesktop.org/documentation/additional/design/dmabuf.html) - Cuda doesn't seem to support this directly, (`vaapi` should) it should be possible to import `DMABuf` using `glupload -> cudaupload`. - I guess worst case this will end up doing GPU->GPU copy? This needs more research... ## Useful links - https://github.com/cmeissl/gst-wlr-screencopy-src/tree/main/src - https://github.com/cmeissl/gst_gl_test/blob/main/src/main.rs

bsakaguchi

2026-10-15 12:47:16 +00:00