Record pinned SwiftShader P3 functional check
Native and manual checks / native (ubuntu-24.04) (push) Failing after 31s
Native and manual checks / manual (push) Successful in 27s
Windows editor and software Vulkan / windows-graphics (push) Canceled after 0s
Native and manual checks / native (windows-2025) (push) Canceled after 0s

This commit is contained in:
Emil
2026-09-24 03:04:00 +03:00
parent e2e95e1305
commit 8f4650a544
2 changed files with 21 additions and 0 deletions
@@ -39,6 +39,8 @@ Each configuration's p50 is the median of three run medians. The overhead compar
All 3,240 frames submitted the requested local-light count, reported zero Vulkan errors, and retained the requested visibility mode. Validation was **off** during the performance sweep; zero reported errors here is not a validation-layer pass. Separate GPU tests exercise validation and image correctness.
As a separate functional check, the same Release source passed [all five P3 CTests on the pinned Linux SwiftShader ICD](data/p3-lighting-2026-09-24/forward/swiftshader-release-p3-ctest.txt), including sun/local shadow images and the benchmark smoke case. SwiftShader is software Vulkan; these 64 × 64 tests are not another 1920 × 1080 performance measurement. The Windows SwiftShader CI and post-Forward+ runs remain separate acceptance steps.
With shadows on, four point lights request 24 faces but only 12 fit atomically in the 16-tile atlas (two full six-face point lights). The same 12 faces remain at 16–128 lights; 128 lights request 768 faces and explicitly drop 756. The corresponding local shadow GPU p50 stays near 0.018 ms. Thus the scaling above is mainly fragment shading, not additional shadow-map rendering. The table does not imply that 128 shadowed points are supported simultaneously.
`gpu_ms` includes the renderer's synchronous framebuffer capture copy, while `cpu_ms` includes submission and wait. Neither is an interactive game frame rate. The gate deliberately uses GPU raster timestamps; the post-occlusion raster pass is included when active. The fixed geometry and light arrangement, GPU clock state, and one driver limit generalization to other scenes and devices. The retained raw data allow this conclusion to be recomputed without trusting the prose.
@@ -0,0 +1,19 @@
Test project /home/emil/Desktop/.worktrees/Faset_Engine-p3-benchmark-release/build/linux-release
Start 7: render_lighting_sun
1/5 Test #7: render_lighting_sun ................ Passed 5.08 sec
Start 8: render_lighting_local
2/5 Test #8: render_lighting_local .............. Passed 4.26 sec
Start 9: render_lighting_policy
3/5 Test #9: render_lighting_policy ............. Passed 0.00 sec
Start 17: render_lighting_benchmark_schema
4/5 Test #17: render_lighting_benchmark_schema ... Passed 5.36 sec
Start 18: render_lighting_benchmark_smoke
5/5 Test #18: render_lighting_benchmark_smoke .... Passed 1.59 sec
100% tests passed, 0 tests failed out of 5
Label Time Summary:
gpu = 10.93 sec*proc (3 tests)
p3 = 16.29 sec*proc (5 tests)
Total Test time (real) = 16.30 sec