Hello Kitty Island Adventure (PPSA18176) on Windows / NVIDIA: bring-up and rendering blockers

[RipcuBAN]

Hello Kitty Island Adventure, PPSA18176, content version 02.000.000, in-game version 2.6.100.

This is a work-in-progress bring-up report, not a verified compatibility result.

Environment and scope

Windows 11 Pro build 26300, NVIDIA RTX 3060, driver 32.0.16.1714, MinGW-w64 GCC 15.2, Release. This is a local selective-integration fork branch, not unmodified current upstream main. Its common ancestor with the upstream snapshot is 5cb747c146c67ffd498ab7ebaced9395035a50f6; local committed head is adf64086.

Current observations

• Startup reaches the animated title screen repeatedly. Logo, text and the start prompt are readable; the 3D background has extensive black speckling and excessive brightness. This was visually checked, not inferred from logs.
• Three runs before and three after the latest legacy-arithmetic integrations were logged. The inspected before captures and all three after captures show the same visible problem; screenshot coverage before the integrations is incomplete. The title-screen runs were bounded; they do not demonstrate sustained gameplay, saves or audible audio.
• The existing synchronous sampled-texture-upload diagnostic (APS5_SYNC_TEXTURE_UPLOAD=1) also leaves the visible corruption. That is one diagnostic observation, not proof that every synchronization path is correct.
• Automated Return/Space presses did not visibly advance the start prompt in the latest probe. Input and gameplay remain unverified; this is not yet an established title-side deadlock.

Relevant upstream work and verification

The branch includes the folded vertex/instance offset correction from #2285. The GPU execution test passes, but installing the patched AGC driver did not resolve the pictured corruption.

The measured legacy arithmetic corrections from #1281 and #1290 were then cherry-picked with original authorship. The manually assembled RDNA MAC regression failed before integration on lane 8: (0x00000000, 0x3f800000, 0x7fc00000) produced 0x7fffffff, expected 0x7fc00000. After integration, these pass:

agc_driver_mac_legacy
agc_driver_fma_legacy
vector_mad_legacy
agc_driver_f32_denormal_flush

The last test includes positive controls and rejection of unsupported FLOAT_MODE 0xd0/0xe0. These are host-GPU regression checks against the published AMD measurements. They do not measure PS5 gfx1013 behaviour; the source limitations remain in TechnicalDebt. The rendering defect is still present, so these commits are not being presented as its fix.

Next evidence needed

I am collecting actual draw/depth state and thread backtraces, then narrowing the corruption to a reproducible generic PM4/shader/texture case. A first gdb snapshot contains many condition waits, but animation continues and guest frames are incompletely symbolized; that snapshot alone does not establish a hang.

Has anyone reproduced this title's speckled/overbright rendering on a clean upstream tree, or isolated a relevant generic shader/texture case? Please link existing issues/PRs so fixes are not duplicated. No title-specific workaround is proposed here.

AI-assisted: yes, Codex; earlier local work also used Claude.