velzieandClaude Opus 5
593a5a4762
Four cases added, each in *both* senses. That is the point: a case that
only checks the skip passes on a driver that never performs the command at
all, which is a different defect wearing the same result.
clear-fails / clear-passes glClear inside a conditional
dispatch-fails / dispatch-passes glDispatchCompute inside one
any-samples-occluded / -visible GL_ANY_SAMPLES_PASSED with the depth
test as the discriminator
The first four were 7/9 before the driver change and are 11/11 after,
natively and in the browser.
The last two exist because KHR-GL45.compute_shader.conditional-dispatching
still fails with conditional render working, and GEX_WGPU_CONDRENDER_LOG
shows *both* its queries returning zero -- so the quad that should pass the
depth test recorded no samples either. These two drive exactly that shape
and pass, which says it is neither the query type nor the depth test in
isolation. The lead is narrowed rather than closed.
The probe's context moves 3.3 -> 4.3 for the dispatch case, and its
framebuffer gains a depth attachment; the earlier cases never enable the
depth test, so neither changes them.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01EmPqxxjBJ6rtNDimtJAQpW
Browser demo
The SuperTuxKart demo uses the direct compatibility graphics path:
stock application -> guest libGL/libEGL marshaller -> MobileGL -> WebGPU
It does not run virglrenderer or guest Mesa. Those remain available for the separate legacy virgl demos; the direct path intercepts the standard library ABI, so STK and other desktop applications require no source modification.
MobileGL is pinned as third_party/mobilegl-webgpu and patched at build time.
The WebGPU fork is unfinished and, candidly, much of this backend is not written
well: important GL invariants are missing from cache keys, unsupported calls
silently poison whole WebGPU command buffers, and several formats, framebuffer
operations, and texture types remain unimplemented despite advertising desktop
OpenGL 3.3. Treat it as an experimental compatibility layer, not a conformant GL
implementation.
Keeping the changes in web/mobilegl-*.patch makes those local fixes explicit
and leaves the upstream checkout reproducible.
Build the direct WebGPU renderer, then publish the assets:
git submodule update --init --recursive third_party/mobilegl-webgpu
web/build-mobilegl-wasm.sh
(cd web && npm run assets)
The first WebGPU build compiles the fork's pinned Tint and glslang trees and is
substantially slower than subsequent incremental builds. SuperTuxKart selects
gexmobilegl-webgpu.mjs; the virgl demos continue to use their own renderer.
OpenGL 4.4 validation
The complete 543-command core-4.4 loader surface and the currently transported and WebGPU-backed subsets have a structural gate:
tests/gl44-layer.sh
That is deliberately not a conformance claim. Official semantic evidence comes
from the pinned Khronos GL CTS (opengl-cts-4.6.8.1):
export GEX_CTS_TOOLCHAIN=/path/to/aarch64-linux-musl-cross
tests/gl44-cts.sh
tests/gl44-cts-browser.sh
The second command runs a real KHR-GL44 case through the persistent browser
and direct MobileGL/WebGPU path. GEX_CTS_CASE=KHR-GL44.… selects another case;
the growing pass list, rather than symbol coverage or an advertised version
string, is the compliance measure.
SuperTuxKart real-present benchmark
With the persistent browser CDP server on port 9335 and the demo server on port 8091, run:
tests/stk-webgpu-fps.sh
The gate profiles a race and separately measures browser animation ticks and
distinct WebGPU canvas frames over the final ten seconds. It requires at least
59.8 real presents per second, 99.5% fresh-frame coverage, and no more than one
missed animation tick. The game can produce frames faster than the browser can
display them, so its own Average FPS is diagnostic rather than the pass/fail
metric. A four-snapshot MobileGL presentation queue decouples guest production
from browser animation timing; this costs about 4.7 MiB at 640x480 and can add
roughly 40 ms of queueing latency when the guest runs near 95 FPS.