Repository navigation
Fix WSL GPU stalls and show the active rendering device - #4
Merged
Merged
Conversation
Emit equivalent triangles for cached fans and convex field spots. Mesa D3D12 primitive conversion otherwise maps cached GPU index buffers for CPU conversion and waits on a fence. Add a real rendering-path regression and the verified WSL adapter selection command.
fred-huang122
changed the base branch from
perf/ode-0.16.6-integration
to
master
November 2, 2026 17:20
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fix WSL Mesa D3D12 rendering stalls and show the active graphics device in grSim.
This PR remains stacked on #3. Ordinary merges include its corrected robot contacts, faster CPU triangular solves and automatic island-worker policy; those changes remain in the #3 base.
The isolated topology fix previously increased RTX 4060 Laptop rendering from roughly 7 FPS to roughly 50 FPS on stock Ubuntu Mesa 25.2.8. Controlled competing GPU work reduced the fixed build from about 54 to 42 FPS and recovered after it stopped.
The latest paired default-mode comparison keeps the graphics fix and accurate collisions identical, comparing the previous serial default with #3's new CPU math and automatic workers. Three shuffled repetitions per case, 6 v 6 on Division B, normal detail, 956×545 viewport, 3-second warmup and 6-second measurement. A fresh world and atomic placement give both builds matching opposing formations and ball positions; GRSIM_ODE_THREADS is unset for both.
All nine new-default app trials pass their live gate and report three actual helpers. Pushing specifically meets ±2% real-time speed, at least 30 FPS and less than 50 ms final backlog in all three trials: 47.11–47.68 FPS, 0.9975–1.0024×, 6.54–8.71 ms final backlog. The previous default accumulates 1.86–5.52 seconds and suppresses drawing under the existing backlog rule. Forcing serial mode can still lag during cold pushing; automatic workers avoid that default-path failure.
All trials use the real D3D12 NVIDIA GeForce RTX 4060 Laptop GPU context on stock Mesa 25.2.8-0ubuntu0.24.04.2. Background GPU utilization with grSim closed ranges 14–66% (median 34%). Existing workloads remain running; small FPS differences are not universal speedup claims. Crowd ratios above 1 indicate recovery from initial backlog. Renderer and physics probes have different OS thread IDs.
A separate live default-mode collision audit records 47,841 robot contacts at every physics substep, including startup. Peak shell/planar penetration is 0.711 mm, below the existing 2 mm limit, with finite contacts and zero solver warnings/errors; its 15-second measurement maintains 1.0001× and 54.2 FPS. The fixed-step math/trajectory comparisons are documented in #3.
Validation: all 21 local regressions pass. Current CI passes Ubuntu 22.04/24.04, native arm64 macOS 15/26, and downloading/running the exact macOS 15 package on macOS 26. The shipped app reports Apple Software Renderer, status=1 and rendering_on_gui_thread=1 on the hosted Mac, validating packaging/runtime rather than physical Mac GPU throughput.
Profiling datasets, reports, contact probes and prototypes remain outside Git. System graphics drivers and existing user workloads were left unchanged.
Verified Apple Silicon app with graphics fix and updated ODE, macOS 15 or later.