From 17df86f0f3681fd91175af77eaf0e77c17e13523 Mon Sep 17 00:00:00 2001 From: Valentin Popov Date: Sat, 18 Jul 2026 18:38:10 +0400 Subject: feat(game): export static Vulkan readback --- docs/tomes/05-render.md | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'docs/tomes/05-render.md') diff --git a/docs/tomes/05-render.md b/docs/tomes/05-render.md index 82ed843..a32d68c 100644 --- a/docs/tomes/05-render.md +++ b/docs/tomes/05-render.md @@ -1498,6 +1498,15 @@ errors. This validates the offline handoff for one captured instant; it does not claim live camera tracking, material parity, or an original-frame pixel comparison. +For visual regression work, `fparkan-game --backend static-vulkan` also accepts +`--readback-out `. It writes the final synchronized Vulkan readback bytes +only after the renderer has completed its normal teardown evidence; the JSON +report records the raw Vulkan format, byte count, hash, and requested path. A +three-frame diagnostic AutoDemo run wrote 7,372,800 bytes at format `50` +(`VK_FORMAT_B8G8R8A8_UNORM`) with the same frame hash reported by the renderer. +The artifact is a Vulkan-side comparison input, not an asserted original-frame +image. + The first bounded launch showed that repeated fingerprinting, rather than Vulkan initialization, was the load-path bottleneck: each new MAT0/TEXM request re-opened an already decoded archive and re-hashed its entire source file. -- cgit v1.2.3