DAVE THE DIVER (PPSA19115)

[Gashrod1]

Existing thread

• I searched this category for the title ID and no thread exists for this title on any platform.

Title

DAVE THE DIVER (PPSA19115)

Content version tested

01.041.000

AnyPS5 commit

main 956269a plus open PRs #4820 (84f6364), #5496 (213050f), #5445 (556890f), #5892 (1e5d3ab) and #5558 (079164e), and the local changes listed under Current blocker. #5445 was merged by hand on top of #4820 and #5496 (its branch carries an older #4820): conflicts in tests and CMake only, plus one call in createFilled adapted to the 3-argument guard(context, use, created.active).

System tested

Windows 11 Pro 26200, WinLibs MinGW-w64 GCC 15.2.0 r7, AMD Ryzen 7 5800X, NVIDIA RTX 5060 (596.36). Start-up also checked on Arch Linux, GCC, Intel Alder Lake-P iGPU (Mesa ANV 26.2.4).

Furthest point

In game

What is shown

Unity 6000.0.52f1, IL2CPP. Intro cutscene at 60 FPS, in game about 20 FPS, pauses during loads (shader cache was cold). Audio is clean with #5892 and #5558; without them it cut out and crackled. Keyboard input works. 0 absent imports (tools/import_audit.py), 24 imports of stubs, none hit so far.

Profile on Windows (15 gdb stack samples while in game): GPU at 20-25%, no core saturated. The submitting thread waits in Driver::holdFlipBehindWorker in 14 of 15 samples; the queue worker is busy in every sample, about a quarter of it in memcmp under GuestMemory::EqualsCommittedUnsynced / ReadCommittedUnsynced (Windows direct host imports are compared, not watched), the rest in BindExactBuffer, PrepareDrawBindings and draw resource resolution.

Current blocker

On main 956269a the title stops at its first frames:

DescriptorBindingBuilder::Populate multisampled image resources have no descriptor image shape

That is #3036. Each open PR moves it one step:

Stack	Stop
main	multisampled image resources have no descriptor image shape
+ #4820	guest texture descriptor uses MSAA which is not implemented
+ #5496	sampling the 1920x1080 multisampled texture 0x231590000 of 2 samples, which no draw rendered to, is not implemented
+ #5445	... which no draw rendered to and whose tile mode 4 has no modeled multisampled color footprint, is not implemented

Tile mode 4 is kZ64KBX: a 2-sample Z_X depth texture sampled before any draw rendered to it. Past that, the native multisample depth path of #5496 refuses, in this order:

1. native multisample depth metadata fill transition is not implemented (an HTILE fill covering the whole target)
2. native multisample stencil or depth format import is not implemented (the depth target has stencil)
3. native multisample depth import requires watched accessible guest memory, then native multisample depth requires watched accessible guest memory at sampling
4. native multisample depth fast-clear transition is not implemented
5. native multisample depth sampled identity or format mismatch (when returning to the boat)

To get in game I bypassed these locally. These are workarounds, not proposals:

• createFilled accepts the 64 KiB Z/S/D XOR tile modes with the Standard64KB footprint.
• NoteDepthMetadataFill does not refuse sampling when the fill covers the whole HTILE (the pending clear is kept).
• The native multisample depth import accepts VK_FORMAT_D32_SFLOAT_S8_UINT and a stencil plane, and skips the initial-state proof; CheckCurrent only checks the refusal.
• Fast clears on a sampled native multisample depth target are applied (ApplyFastClear) instead of refused.
• An identity or format mismatch falls back to the color path instead of throwing. With it, returning to the boat no longer crashes (checked in game). The mismatch logged is always the same: depth target 0x241510000 1920x1080, 2 samples, VK_FORMAT_D32_SFLOAT_S8_UINT; sampled texture 0x242610000 1920x1080, 1 sample, guest format 36. So the sampled texture is single-sample color at target + 0x1100000, likely matched through the target's stencil plane address, and is not a multisample depth read. It hit about 28,000 times in one session.

Also seen, not investigated further:

• Linux iGPU run: one crash in Unity's allocator (TLSF free-list unlink on a corrupted block header) after a scene load. Not seen on Windows so far.
• Linux: at exit, ld-linux runs CommonDialog.prx's fini entry with garbage arguments and it jumps into its own dynamic table.
• Shader prebuild logs multisampled array image resources have no descriptor image shape (skipped, not fatal).

How it was run

ELFs dumped decrypted from the console. Relinked on Linux:

relinker --windows --module-dir Media/Plugins --module-dir Media/Modules eboot.bin app.exe

(--registry --to-intel for the Linux run on Intel). Media/ and sce_sys/ copied into app0/ beside the converted modules, libs/ from the Windows build with the three MinGW DLLs. Started with cmd /c "app.exe > run.log 2>&1", no environment variables.

AI-assisted: yes