[PR #396] Add party mode split rendering #441

New issue

Open

opened 2026-10-15 13:00:35 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-15 13:00:35 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/games-on-whales/wolf/pull/396
Author: @JBailes
Created: 4/20/2026
Status: 🔄 Open

Base: fix/amd-dma-buf-negotiation ← Head: codex/party-mode-split-render

📝 Commits (9)

65e0d7f fix: intersect waylanddisplaysrc and vapostproc DMA-BUF caps
dff853e fix: query waylanddisplaysrc DMA formats at runtime to intersect with vapostproc
401e840 style: wrap over-length log line to satisfy clang-format
14ec21e docker: pin gst-interpipe to context-query-forwarding fix
347e584 feat: 10-bit SDR streaming (HEVC/AV1 Main10)
bc652a8 test: update AV1 ServerCodecModeSupport to protocol-correct value
bbfda44 feat: bump config to v8 and migrate so existing installs get 10-bit encoders
d48be72 feat: add party mode split rendering
788890b feat: GPU-aware party split compositor + v9 config

📊 Changes

41 files changed (+3205 additions, -152 deletions)

View changed files

📝 .gitignore (+1 -0)
📝 docker/gstreamer.Dockerfile (+4 -0)
📝 src/core/src/core/gstreamer.hpp (+1 -47)
📝 src/core/src/core/virtual-display.hpp (+7 -1)
📝 src/core/src/platforms/linux/virtual-display/gst-wayland-display.cpp (+59 -3)
📝 src/core/src/platforms/unknown/no-virtual-display.cpp (+2 -2)
📝 src/moonlight-protocol/moonlight.cpp (+16 -4)
📝 src/moonlight-protocol/moonlight/protocol.hpp (+3 -1)
📝 src/moonlight-server/CMakeLists.txt (+1 -1)
📝 src/moonlight-server/api/api.hpp (+15 -1)
📝 src/moonlight-server/api/endpoints.cpp (+159 -9)
📝 src/moonlight-server/api/unix_socket_server.cpp (+23 -1)
📝 src/moonlight-server/control/control.cpp (+7 -6)
📝 src/moonlight-server/control/control.hpp (+7 -2)
📝 src/moonlight-server/control/input_handler.cpp (+8 -9)
📝 src/moonlight-server/events/events.hpp (+24 -0)
📝 src/moonlight-server/platforms/hw_unknown.cpp (+12 -2)
📝 src/moonlight-server/rest/endpoints.hpp (+3 -1)
📝 src/moonlight-server/rtsp/commands.hpp (+10 -4)
📝 src/moonlight-server/rtsp/net.hpp (+3 -2)

...and 21 more files

📄 Description

What changed

Adds 2-seat party-mode split rendering to Wolf, layered on top of #419's zero-copy / gst-interpipe context-sharing work.

Rebased onto #419 (fix/amd-dma-buf-negotiation) — this PR now targets #419, not stable, because the GPU-aware split compositor depends on #419's gst-interpipe GST_QUERY_CONTEXT forwarding (so VA/CUDA elements share a context across the interpipe boundary) and its generic-DMABuf runtime negotiation.
Party-mode session management + API for join, promote, and teardown; secondary "hidden seat" sessions; local-gamepad join path.
GPU-aware split compositor. The split is composited entirely in GPU memory, selected per detected vendor so solo and party sessions stay zero-copy:
- VA/QSV: vapostproc + vacompositor (VAMemory)
- NVIDIA: cudaconvertscale + cudacompositor (CUDAMemory)
- software fallback: videoconvertscale + compositor
  The old unconditional system-memory videoconvertscale could not ingest CUDAMemory/DMABuf at all, so it would have broken zero-copy streaming for every GPU session; the per-vendor sources fix that.
Config bumped to v9 (migration version <= 8) so existing v8 (10-bit) installs are re-migrated and pick up the party pipeline. config.v9.toml carries the software party source plus default_source_va / default_source_cuda; configTOML selects by vendor.
Restart-runner-in-place so the primary app relaunches at split render size without dropping the Moonlight session; render size decoupled from display_mode via render_display_mode.
Hardened session-id parsing across all session endpoints (parse_session_id), preserve producer caps features on resize, dedup the two-seat geometry.

Verification

cmake build of wolf (full Docker image) green against #419's pinned gst-interpipe.
VA (vacompositor) and CUDA (cudacompositor) split-compositor chains validated standalone with gst-launch on real hardware — both negotiate the split entirely in GPU memory (no host round-trip), reach PLAYING, no caps errors.
Real wolf binary boots, migrates config to v9, and selects the correct GPU party source per vendor: "zero copy pipeline on Nvidia" (RTX 5080, renderD128 -> CUDA source) and "zero copy pipeline on Intel" (UHD 770, renderD129 -> VA source).
Pending: end-to-end Moonlight client stream with a live party join (decode/visual check across vendors).

Depends on

#419 (fix/amd-dma-buf-negotiation) — base branch.
games-on-whales/wolf-ui#12, games-on-whales/gst-wayland-display#27 (companion changes).

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/games-on-whales/wolf/pull/396 **Author:** [@JBailes](https://github.com/JBailes) **Created:** 4/20/2026 **Status:** 🔄 Open **Base:** `fix/amd-dma-buf-negotiation` ← **Head:** `codex/party-mode-split-render` --- ### 📝 Commits (9) - [`65e0d7f`](https://github.com/games-on-whales/wolf/commit/65e0d7f36488d974e1dc30b6445a7f67debdf627) fix: intersect waylanddisplaysrc and vapostproc DMA-BUF caps - [`dff853e`](https://github.com/games-on-whales/wolf/commit/dff853e85577b5fee8f1846e26bfd88388ab2d70) fix: query waylanddisplaysrc DMA formats at runtime to intersect with vapostproc - [`401e840`](https://github.com/games-on-whales/wolf/commit/401e840241891c2b063b0f79b3940e97ab5e90d9) style: wrap over-length log line to satisfy clang-format - [`14ec21e`](https://github.com/games-on-whales/wolf/commit/14ec21e3f000fa9f68d80e58363c522ee5b6090d) docker: pin gst-interpipe to context-query-forwarding fix - [`347e584`](https://github.com/games-on-whales/wolf/commit/347e584559827b6b2a6f03ae3143fd6208bedf00) feat: 10-bit SDR streaming (HEVC/AV1 Main10) - [`bc652a8`](https://github.com/games-on-whales/wolf/commit/bc652a87b9e2be719a8ee57ea3ae5da9a01adf96) test: update AV1 ServerCodecModeSupport to protocol-correct value - [`bbfda44`](https://github.com/games-on-whales/wolf/commit/bbfda4415883b84f419dfb979101944ae69dd48b) feat: bump config to v8 and migrate so existing installs get 10-bit encoders - [`d48be72`](https://github.com/games-on-whales/wolf/commit/d48be720853b612fd758ac0bc89b731769dbd074) feat: add party mode split rendering - [`788890b`](https://github.com/games-on-whales/wolf/commit/788890bfa444260d07c74c89d6fc6cb2c0a194ad) feat: GPU-aware party split compositor + v9 config ### 📊 Changes **41 files changed** (+3205 additions, -152 deletions) <details> <summary>View changed files</summary> 📝 `.gitignore` (+1 -0) 📝 `docker/gstreamer.Dockerfile` (+4 -0) 📝 `src/core/src/core/gstreamer.hpp` (+1 -47) 📝 `src/core/src/core/virtual-display.hpp` (+7 -1) 📝 `src/core/src/platforms/linux/virtual-display/gst-wayland-display.cpp` (+59 -3) 📝 `src/core/src/platforms/unknown/no-virtual-display.cpp` (+2 -2) 📝 `src/moonlight-protocol/moonlight.cpp` (+16 -4) 📝 `src/moonlight-protocol/moonlight/protocol.hpp` (+3 -1) 📝 `src/moonlight-server/CMakeLists.txt` (+1 -1) 📝 `src/moonlight-server/api/api.hpp` (+15 -1) 📝 `src/moonlight-server/api/endpoints.cpp` (+159 -9) 📝 `src/moonlight-server/api/unix_socket_server.cpp` (+23 -1) 📝 `src/moonlight-server/control/control.cpp` (+7 -6) 📝 `src/moonlight-server/control/control.hpp` (+7 -2) 📝 `src/moonlight-server/control/input_handler.cpp` (+8 -9) 📝 `src/moonlight-server/events/events.hpp` (+24 -0) 📝 `src/moonlight-server/platforms/hw_unknown.cpp` (+12 -2) 📝 `src/moonlight-server/rest/endpoints.hpp` (+3 -1) 📝 `src/moonlight-server/rtsp/commands.hpp` (+10 -4) 📝 `src/moonlight-server/rtsp/net.hpp` (+3 -2) _...and 21 more files_ </details> ### 📄 Description ## What changed Adds 2-seat party-mode split rendering to Wolf, layered on top of #419's zero-copy / gst-interpipe context-sharing work. - **Rebased onto #419** (`fix/amd-dma-buf-negotiation`) — this PR now targets #419, not `stable`, because the GPU-aware split compositor depends on #419's gst-interpipe `GST_QUERY_CONTEXT` forwarding (so VA/CUDA elements share a context across the interpipe boundary) and its generic-DMABuf runtime negotiation. - **Party-mode session management + API** for join, promote, and teardown; secondary "hidden seat" sessions; local-gamepad join path. - **GPU-aware split compositor.** The split is composited entirely in GPU memory, selected per detected vendor so solo and party sessions stay zero-copy: - VA/QSV: `vapostproc` + `vacompositor` (VAMemory) - NVIDIA: `cudaconvertscale` + `cudacompositor` (CUDAMemory) - software fallback: `videoconvertscale` + `compositor` The old unconditional system-memory `videoconvertscale` could not ingest CUDAMemory/DMABuf at all, so it would have broken zero-copy streaming for every GPU session; the per-vendor sources fix that. - **Config bumped to v9** (migration `version <= 8`) so existing v8 (10-bit) installs are re-migrated and pick up the party pipeline. `config.v9.toml` carries the software party source plus `default_source_va` / `default_source_cuda`; `configTOML` selects by vendor. - Restart-runner-in-place so the primary app relaunches at split render size without dropping the Moonlight session; render size decoupled from `display_mode` via `render_display_mode`. - Hardened session-id parsing across all session endpoints (`parse_session_id`), preserve producer caps features on resize, dedup the two-seat geometry. ## Verification - `cmake` build of wolf (full Docker image) green against #419's pinned gst-interpipe. - VA (`vacompositor`) and CUDA (`cudacompositor`) split-compositor chains validated standalone with `gst-launch` on real hardware — both negotiate the split entirely in GPU memory (no host round-trip), reach PLAYING, no caps errors. - Real wolf binary boots, migrates config to v9, and selects the correct GPU party source per vendor: "zero copy pipeline on Nvidia" (RTX 5080, renderD128 -> CUDA source) and "zero copy pipeline on Intel" (UHD 770, renderD129 -> VA source). - Pending: end-to-end Moonlight client stream with a live party join (decode/visual check across vendors). ## Depends on - #419 (`fix/amd-dma-buf-negotiation`) — base branch. - `games-on-whales/wolf-ui#12`, `games-on-whales/gst-wayland-display#27` (companion changes). --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-15 13:00:35 +00:00

bsakaguchi referenced this issue

2026-10-15 13:00:48 +00:00

[PR #441] fix: serialize the RTSP handshake per client IP (same-NAT clients) #469