engine(macos-arm64): publish 73d20a40 - #250 parallel draw-list sub-recordings nest + RD texture wrappers allocate their RS texture before the render-thread bind (milky ocean under the separate render thread); GodotSharp regenerated

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018hSLRwirUk2NF4j1v2kH4Z
This commit is contained in:
ozanandClaude Opus 5.5 committed 2026-10-19 18:41:17 +01:00
1 parent 736e3e9bc3
commit c632796638
33 files changed
+50 -50

No files matched your search

+2 -2
View File
@@ -31,10 +31,10 @@ into GodotSharp, so GodotSharp had to be regenerated — and must ship here alon
| Platform | Engine | Terrain3D | GodotSharp shipped? | | Platform | Engine | Terrain3D | GodotSharp shipped? |
|---|---|---|---| |---|---|---|---|
| `engine/macos-arm64/` | **4.7.2** (`d4940046`) — `a967d12b` (parallel draw-list replay + recording) plus the parallel-replay command-pool reset (VUID-00050 fix) | built-in module | ✅ yes (`GodotSharp/`) | | `engine/macos-arm64/` | **4.7.2** (`73d20a40`) — `d4940046` plus the #250 parallel-recording nesting fix and the Texture2DRD bind-race fix | built-in module | ✅ yes (`GodotSharp/`) |
| `engine/windows-x64/` | **4.7.2** (`d49400463`) — same hash as macos-arm64: parallel draw-list replay + recording (game ships both ON) plus the VUID-00050 pool-reset fix, **verified 10/10 oracle with zero VUIDs before publish**; `template_release` here is still the `1e7be28b` tree | built-in module | ✅ yes (`GodotSharp/`) | | `engine/windows-x64/` | **4.7.2** (`d49400463`) — same hash as macos-arm64: parallel draw-list replay + recording (game ships both ON) plus the VUID-00050 pool-reset fix, **verified 10/10 oracle with zero VUIDs before publish**; `template_release` here is still the `1e7be28b` tree | built-in module | ✅ yes (`GodotSharp/`) |
Mac is engine `d4940046` as of 2026-10-19 (`--version` `4.7.2.stable.mono.custom_build.d49400463`): `a967d12b` plus a per-frame reset of the parallel-replay secondary command pools (fixes VUID-vkBeginCommandBuffer-00050). `a967d12b` is `e64890fd` plus parallel draw-list replay (Vulkan only) and parallel draw-list recording (every driver), both engine-default off and turned on by the game; GodotSharp regenerated; oracle `parallel_draw_replay` Vulkan 10/10, Metal 5/5 (`engine/.agents/wiki/parallel-draw-replay.md`). **Windows is also engine `d49400463` as of 2026-10-19** — Forge verified the fix before publishing: replay+record oracle Vulkan `--gpu-validation` **10/10 exit 0 with ZERO VUID lines** (on `a967d12bf` the same run had 10× VUID-00050 — the finding that produced this fix). The parallel features' Windows story: the fps lane's in-game A/B on `a967d12bf` recorded no fps regression (42→46-48), rGPU down 8.1→6.8 ms, 10-min soak stable, eye check PASS, and the game ships both ON (`f7ddc64a1`); timing claims carry the throttled-box caveat from the engine wiki. Earlier Windows oracle work on `a967d12bf`: `draw_indirect_count` d3d12 with debug layer 22/0 + expected SKIP line; `multimesh_cull_pool` d3d12 4524/4524 (parity with Mac, pooled-cull gate met). Vulkan SDK + Graphics Tools are installed on Forge. Still owed: a `--gpu-validation` in-game soak on `d49400463`, tests=yes unit suites, crash checklist. Underneath: `e64890fd` (Apple autorelease-drain + G4 census) on `4fde0e94` (G2 pooled cull, fixed shadow band, `STATIC_SHADOW_AT_REST`) on the `9dcc5d458` pair (SSCS contact shadows, hero shadow min-combine, upstream 4.8 shadow fixes #121775/#124348 with the game sun bias 0.014, `draw_list_draw_indirect_count`). Mac is engine `73d20a40` as of 2026-10-19 (`--version` `4.7.2.stable.mono.custom_build.73d20a40b`): `d4940046` plus `219b914c` (#250, parallel draw-list sub-recordings nest; Windows Ultra-bench FATAL) and `73d20a40` (RD texture wrappers allocate their RS texture before the render-thread bind, fixing the milky ocean under the separate render thread); GodotSharp regenerated, API unchanged. The game pin stays `d49400463` until Windows is published at the same commit. Before that, Mac was engine `d4940046` as of 2026-10-19 (`--version` `4.7.2.stable.mono.custom_build.d49400463`): `a967d12b` plus a per-frame reset of the parallel-replay secondary command pools (fixes VUID-vkBeginCommandBuffer-00050). `a967d12b` is `e64890fd` plus parallel draw-list replay (Vulkan only) and parallel draw-list recording (every driver), both engine-default off and turned on by the game; GodotSharp regenerated; oracle `parallel_draw_replay` Vulkan 10/10, Metal 5/5 (`engine/.agents/wiki/parallel-draw-replay.md`). **Windows is also engine `d49400463` as of 2026-10-19** — Forge verified the fix before publishing: replay+record oracle Vulkan `--gpu-validation` **10/10 exit 0 with ZERO VUID lines** (on `a967d12bf` the same run had 10× VUID-00050 — the finding that produced this fix). The parallel features' Windows story: the fps lane's in-game A/B on `a967d12bf` recorded no fps regression (42→46-48), rGPU down 8.1→6.8 ms, 10-min soak stable, eye check PASS, and the game ships both ON (`f7ddc64a1`); timing claims carry the throttled-box caveat from the engine wiki. Earlier Windows oracle work on `a967d12bf`: `draw_indirect_count` d3d12 with debug layer 22/0 + expected SKIP line; `multimesh_cull_pool` d3d12 4524/4524 (parity with Mac, pooled-cull gate met). Vulkan SDK + Graphics Tools are installed on Forge. Still owed: a `--gpu-validation` in-game soak on `d49400463`, tests=yes unit suites, crash checklist. Underneath: `e64890fd` (Apple autorelease-drain + G4 census) on `4fde0e94` (G2 pooled cull, fixed shadow band, `STATIC_SHADOW_AT_REST`) on the `9dcc5d458` pair (SSCS contact shadows, hero shadow min-combine, upstream 4.8 shadow fixes #121775/#124348 with the game sun bias 0.014, `draw_list_draw_indirect_count`).
Long-term: CI (Gitea Actions) builds on push and publishes here / to S3 — Long-term: CI (Gitea Actions) builds on push and publishes here / to S3 —
`docs/ai/plans/build-pipeline-architecture-2026-06-07.md`. `docs/ai/plans/build-pipeline-architecture-2026-06-07.md`.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
+6 -6
View File
@@ -4,16 +4,16 @@
"macos-arm64": { "macos-arm64": {
"binary": "engine/macos-arm64/tinqs.macos.editor.arm64.mono", "binary": "engine/macos-arm64/tinqs.macos.editor.arm64.mono",
"godotsharp": "engine/macos-arm64/GodotSharp", "godotsharp": "engine/macos-arm64/GodotSharp",
"version": "4.7.2.stable.mono.custom_build.d49400463", "version": "4.7.2.stable.mono.custom_build.73d20a40b",
"engine_repo": "tinqs/engine", "engine_repo": "tinqs/engine",
"engine_commit": "d4940046", "engine_commit": "73d20a40",
"sha256": "4fafb8bd62e1448fadd0a41ac4a938ba7b9050c1d7b6f1219fd7c13b48bdee50", "sha256": "e3f5256c7855e4faf03e1c8e53e3c02e78e1d4e146b19b7bda3105445340f6ba",
"target": "editor", "target": "editor",
"built_on": "2026-10-19", "built_on": "2026-10-19",
"built_by": "Ozan/Kraken via Claude (parallel replay command-pool reset)", "built_by": "Ozan/Kraken via Claude (parallel recording nesting + Texture2DRD bind race)",
"toolchain": "Xcode (Apple Silicon), SCons, VulkanSDK 1.4.350.0", "toolchain": "Xcode (Apple Silicon), SCons, VulkanSDK 1.4.350.0",
"scons_flags": "platform=macos target=editor arch=arm64 module_mono_enabled=yes vulkan_sdk_path=~/VulkanSDK/1.4.350.0/macOS", "scons_flags": "platform=macos target=editor arch=arm64 module_mono_enabled=yes vulkan_sdk_path=~/VulkanSDK/1.4.350.0/macOS",
"note": "Mac. Engine d4940046 = a967d12b + each parallel-replay secondary command pool is reset at the frame's graph begin (fixes VUID-vkBeginCommandBuffer-commandBuffer-00050 seen on NVIDIA with validation; Vulkan only, no effect on Metal). GodotSharp regenerated (with Tools/), API unchanged. Oracle tests/gpu_oracle/parallel_draw_replay Vulkan/MoltenVK 10/10; game hosts a world on Metal, 0 wrapper errors. Windows at a967d12bf; the game pin moves to d4940046 once Forge publishes windows-x64 at it with the oracle clean under --gpu-validation." "note": "Mac. Engine 73d20a40 = d4940046 + 219b914c (#250: parallel draw-list sub-recordings nest - a worker waiting inside its slice ran another slice's task and the inner end cleared the outer target, Windows FATAL p_idx >= _len on the Ultra bench) + 73d20a40 (Texture2DRD/TextureLayeredRD/Texture3DRD allocate the RS texture before the render-thread bind is queued - under the separate render thread a material could keep the magenta placeholder; Ariki's ocean drew milky). GodotSharp regenerated (with Tools/), API unchanged. Oracle parallel_draw_replay Vulkan/MoltenVK 10/10, Metal 5/5; game hosts the dev world on Metal with thread_model=2, ocean clear, 0 wrapper errors. Windows still at d49400463; the game pin stays at d49400463 until Forge publishes windows-x64 at 73d20a40."
}, },
"windows-x64": { "windows-x64": {
"binary": "engine/windows-x64/tinqs.windows.editor.x86_64.mono.exe", "binary": "engine/windows-x64/tinqs.windows.editor.x86_64.mono.exe",
@@ -29,7 +29,7 @@
"built_by": "Ozan/Forge via ZCode (VUID-00050 fix verification + publish)", "built_by": "Ozan/Forge via ZCode (VUID-00050 fix verification + publish)",
"toolchain": "MSVC 14.3 (VS2022 Build Tools), Windows SDK 10.0.26100, SCons 4.10.1, .NET SDK 10.0.103", "toolchain": "MSVC 14.3 (VS2022 Build Tools), Windows SDK 10.0.26100, SCons 4.10.1, .NET SDK 10.0.103",
"scons_flags": "platform=windows target=editor module_mono_enabled=yes -j32", "scons_flags": "platform=windows target=editor module_mono_enabled=yes -j32",
"note": "Windows carries d49400463 — SAME hash as macos-arm64 (built at d49400463b): the a967d12bf pair (parallel draw-list replay + recording, both shipping ON in the game since f7ddc64a1; see the a967d12bf note in git history for the full stack) PLUS d49400463b — each parallel-replay secondary command pool resets at the frame's graph begin, fixing the VUID-vkBeginCommandBuffer-00050 (10x on NVIDIA with validation) that Forge's oracle run found on a967d12bf. FIX VERIFIED BEFORE PUBLISH (RTX 5080, Khronos validation layer loaded): parallel_draw_replay oracle vulkan --gpu-validation 10/10 exit 0 with ZERO VUID lines (was 10/10 + 10x VUID-00050 on a967d12bf) — the Mac note's publish condition is met. GodotSharp regenerated from this binary — Api AND Tools, API unchanged. Headless game boot clean. In-game A/B + soak already recorded by the fps lane on a967d12bf (no fps regression 42->46-48, rGPU 8.1->6.8, soak stable, eye check PASS — ea90ea9a6); timing claims throttled-box-caveated per engine docs. Owed on Windows (carried): tests=yes unit suites, crash checklist (8 steps), template_release still 1e7be28b. D3D12 Agility DLLs not shipped (Vulkan game)." "note": "Windows carries d49400463 \u2014 SAME hash as macos-arm64 (built at d49400463b): the a967d12bf pair (parallel draw-list replay + recording, both shipping ON in the game since f7ddc64a1; see the a967d12bf note in git history for the full stack) PLUS d49400463b \u2014 each parallel-replay secondary command pool resets at the frame's graph begin, fixing the VUID-vkBeginCommandBuffer-00050 (10x on NVIDIA with validation) that Forge's oracle run found on a967d12bf. FIX VERIFIED BEFORE PUBLISH (RTX 5080, Khronos validation layer loaded): parallel_draw_replay oracle vulkan --gpu-validation 10/10 exit 0 with ZERO VUID lines (was 10/10 + 10x VUID-00050 on a967d12bf) \u2014 the Mac note's publish condition is met. GodotSharp regenerated from this binary \u2014 Api AND Tools, API unchanged. Headless game boot clean. In-game A/B + soak already recorded by the fps lane on a967d12bf (no fps regression 42->46-48, rGPU 8.1->6.8, soak stable, eye check PASS \u2014 ea90ea9a6); timing claims throttled-box-caveated per engine docs. Owed on Windows (carried): tests=yes unit suites, crash checklist (8 steps), template_release still 1e7be28b. D3D12 Agility DLLs not shipped (Vulkan game)."
} }
} }
} }