* Windows: start Direct3D 12 first, and Vulkan only without a D3D12 GPU
wgpu started Vulkan, D3D12 and OpenGL together and preferred Vulkan, so
every Windows start loaded all installed Vulkan drivers. An old Intel
Vulkan driver (igvk64.dll) crashed 0.10.1 at startup inside the driver.
The app now probes D3D12, then Vulkan, then OpenGL, one backend at a
time, and starts only the first with a hardware adapter (WARP when none
has one). WGPU_BACKEND still wins; graphics.json's "backend" sets the
order by hand; the software renderer (safe mode, the retry) uses D3D12
alone. A new Windows CI job starts the real app under several settings.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* CI: a test filter YAML parses
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* CI: the startup smoke test writes graphics.json where the app reads it
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* CI: run the WARP shader tests one at a time
WARP's shader JIT (d3d10warp.dll) crashed twice in 124 runs of the gpu
tests with four test threads, with a null read inside its own optimizer
while several devices compiled compute pipelines at once. Serial runs
did not crash.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
(cherry picked from commit 95ea05e7c8)
* GPU tessellation: a pipeline the device cannot build falls back to CPU meshing instead of crashing (#1098)
Build the tessellation compute pipeline inside error scopes; TessGpu::new returns an error,
and the viewport logs it and meshes bodies on the CPU. A Windows CI job builds the pipeline
with D3D12's FXC on the WARP adapter.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* Tessellation shader: basis functions in one-dimensional local arrays, so D3D12's FXC compiles it (#1098)
FXC has to unroll loops that write an array inside a struct (Basis.n0/n1) or a 2-D array
(ndu) by a counter, and cannot with run-time trip counts (X3511). A2.2 basis functions in a
local 1-D array returned whole; derivatives from the degree p-1 functions; the pole step
offsets computed rather than indexed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Since the model's mesh went back to the reference mesher (#940), the GPU viewport drew
opaque bodies from their display plans while hover and selection lit faces with the
reference mesh's triangles. Two triangulations of the same face at equal depth fought, so
a hovered face shimmered in wedges instead of lighting whole.
Faces lit over a body drawn from its plan now use the plan's own triangles
(ModelBody::plan_mesh, evaluated once on the CPU), and the highlight pass gets a small
depth bias toward the eye so f32 rounding between the GPU and CPU evaluations of the same
triangles cannot speckle it either.
Test: every face of a box with a pocket and a round boss, hovered and rendered offscreen
through the viewport's real pipelines on the GPU path, lights the same pixels as on the
CPU path (skipped without a GPU adapter).
Co-authored-by: Claude Opus 5.5 <[email protected]>
bherbruck