Bendy and the Dark Revival (PPSA27624)

[JJenkx]

Existing thread

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

Title

Bendy and the Dark Revival (PPSA27624)

Content version tested

01.000.003

AnyPS5 commit

main 44a59edc3a.

System tested

Linux 7.2.9 (Arch), Ryzen 9 7950X, Radeon RX 6650 XT, RADV Mesa 26.2.4, GCC, Release.

Furthest point

Boots (stops in the AGC driver at 90 s)

What is shown

From a fresh save with Cross pressed every 15 s, the window title reports 52, 50 and 49 FPS at 20 to 56 s, then 14 FPS at 74 s. It stops 90 s in:

[gpu] driver failure: shader 0x211584900 stage 1: reserved prepared vertex component selector

I found no open PR or issue for this message.

Current blocker

AGC driver: reserved prepared vertex component selector for a vertex-stage shader.

How it was run

relinker --registry --rpath <build>/core/libs/libs eboot.elf app.elf

Modules in sce_module are converted with the title. This dump has no PlayGo files, so it was run with APS5_PLAYGO_UNLISTED_CHUNKS=installed.

AI-assisted testing and report: yes, Claude Code.

[JJenkx]

Follow-up: I reran this title with a private driver build whose message adds the V# words to reserved prepared vertex component selector. That run stopped at 92 s on a different check of the same kind instead:

[gpu] driver failure: AGC graphics: unsupported vertex buffer descriptor flags

So two runs, on the same main base, failed on two different validity checks of a vertex buffer descriptor: a reserved channel selector (2 or 3) in one, and flag bits the driver doesn't support in the other. That suggests the descriptor the draw uses is not a well-formed V#, or the driver is reading the wrong data as one. I could not capture the words this time. Implementing selectors 2 and 3 would need a hardware measurement, which I could not take on this machine.

AI-assisted: yes (Claude Code).

[valenciana-nc]

@JJenkx I opened #10232 to add the four V# words and attribute location to both errors from these runs; the prepared-selector error also includes the selector and component. This follows your diagnostic approach without relaxing either validity check.

Could you retest PPSA27624 01.000.003 with current main + #10232, using the same setup/input, and post the complete failure line if either check fires? That should capture the descriptor even when the flag check is reached instead of the selector check. No game files or shader payload are needed.

I do not have a local dump and am not claiming a title fix. Local Windows/RTX 4060 focused tests pass (including regressions that fail against main's old messages); the descriptor's source and the underlying cause remain unknown.

AI-assisted: yes.