Indiana Jones and the Staff of Kings (PPSA14388)

[BunnyFl]

Existing thread

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

Title

Indiana Jones and the Staff of Kings (PPSA14388)

Content version tested

01.000.002

AnyPS5 commit

main 352dbe7, no pull requests or local changes on top. The checkpoint runs below add #7920 and a temporary print.

System tested

Windows 11 Pro 10.0.26200, AMD Radeon RX 9070 XT (AMD proprietary driver 26.10.1), Ryzen 7 5800X3D, GCC 15.2.0 (WinLibs 15.2.0posix-14.0.0-ucrt-r7)

Furthest point

Logos or loading screen

What is shown

This is a PS2 Classics package: the PS5 emulator (eboot.bin, PS2LOGO_eeaotv1.prx) runs the US disc SLUS21885#v1.00.iso (--region-dir=SIEA) with its AOT modules SLUS21885_v1_00_eeaotv1.prx and SLUS21885_v1_00_vuaotv1.prx. The emulator's start screens run at 59.94 FPS. Every run but one then ends in VK_ERROR_DEVICE_LOST about 10 to 70 s in, before the game shows anything, with or without input:

[device-fault] result 0: Device lost detected
terminate called after throwing an instance of 'std::runtime_error'
  what():  AGC graphics: vkQueueSubmit recorder: Vulkan result -4

(the failing call is also vkAcquireNextImageKHR or vkWaitForFences present in some runs). The run that lasted, 4 minutes, had APS5_SYNC_DISPATCH=1 and APS5_TRACE_DISPATCH_IO=1; it reached the game's loading screen, where yellow dots flicker and slowly increase. APS5_SYNC_DISPATCH=1 alone does not avoid the loss, so the slowdown from the trace is the likely difference.

With #7920 and APS5_GPU_CHECKPOINTS=1, three of three runs stop at the same dispatch:

[device-fault] last dispatch reached: marker 493, program 0x140c8a000
[device-fault] last dispatch finished: marker 492, program 0x140cb8200
[device-fault] unfinished dispatch: marker 493, program 0x140c8a000
[device-fault] result 0: Device lost detected

VK_LAYER_LUNARG_crash_diagnostic also puts the GPU inside a vkCmdDispatchIndirect of a compute pipeline. With APS5_NO_GPU_INDIRECT=1 the loss stays, and a temporary print of the arguments read on the CPU shows it is the first DISPATCH_INDIRECT of 0x140c8a000, issued right after the emulator logs EE AOT: Setting up interpreter pointers, with arguments at 0x3f4051b40 of 1 x 62976 x 1 groups (two of two runs). In a traced run the dispatches around it are 1 x 3000-3200 x 1 (0x140cb8200) and 80 x 64 or 160 x 128 groups. The device fault report has no fault address, so this looks like a GPU timeout rather than a page fault. I could not tell whether 62976 is the value the title expects.

With VK_LAYER_KHRONOS_validation there is no error before the loss (only VUID-vkResetFences-pFences-01123 after it). GPU-assisted validation disables itself (Failed to submit per pre submission command buffer) and that run ends in an access violation on queue worker 0x0 instead; synchronization validation reports nothing.

Current blocker

VK_ERROR_DEVICE_LOST in the first indirect dispatch of compute program 0x140c8a000 (1 x 62976 x 1 groups).

How it was run

relinker --windows --registry --sce-module-path app0 --module-dir . --module-dir "SLUS21885#v1.00/aot" --module-dir "SLES55448#v1.01/aot" app0/eboot.bin app.exe

The dump's eboot.bin, PS2LOGO_eeaotv1.prx, sce_module/libc.prx and the four AOT modules were unwrapped with tools/unwrap_self.py. tools/import_audit.py reports 0 absent imports and 7 throwing stubs, among them sceSystemServiceLoadExec; none is reached. Warm caches. No APS5_* variables were set in the plain runs apart from APS5_LOG_TIMESTAMPS, APS5_PROFILE_DRAW, APS5_TRACE_SUMMARY_EVERY and APS5_TRACE_DIR.

Conversion, log reading, the diagnostic runs and this report were prepared with AI assistance (Claude Code); the start screens, the loading screen and the crashes were observed by me.