PPSA03743 Soul Hackers 2

[klosixgaming-arch]

Existing thread

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

Title

PPSA03743 Soul Hackers 2

Content version tested

01.002.000

AnyPS5 commit

main + #6441 (unwrap) + #6571

System tested

Windows 11, NVIDIA GeForce RTX 5090 (driver 617.42), AMD Ryzen 9 9950X3D, WinLibs GCC 15.2.0

Furthest point

In game

What is shown

Boots through the title screen and menus and plays the opening cutscene. The player can then walk around in game, and the pause/log menu opens.

Performance is the main problem: 9 to 11 fps in game, dropping to 7 or 8 when moving the character and the camera, and input lags by more than half a second. Shaders compile on the first run (120 in the first 10 seconds) and are read from the disk cache afterwards, so the remaining slowness is not shader compilation. Profiling is in progress and will be posted here.

Current blocker

1. On main, tools/unwrap_self.py refuses all 19 modules (segment 6 vaddr 0x1aa8260 is in no PT_LOAD: a PT_TLS with no file bytes at the end of its PT_LOAD). fix(tools): allow unwrap_self to place zero-size segments without a PT_LOAD host #6441 fixes this; details are in my comment there.

2. An intermittent abort, seen once 12 seconds into loading and once on opening the pause menu:

terminate called after throwing an instance of 'std::runtime_error'
  what():  APR: cannot open ...\app0\Media\StreamingAssets\PS5\Table\battle

The same file was read moments earlier, and the next run went past the same point. I am running a build that records the Windows error when that open fails, to find the cause.

3. Frame rate, as described above.

How it was run

1. Unwrap eboot.bin, sce_module/*.prx, Media/Modules/*.prx and Media/Plugins/*.prx with tools/unwrap_self.py from fix(tools): allow unwrap_self to place zero-size segments without a PT_LOAD host #6441 into a separate folder.
2. Relink: relinker --windows --registry --sce-module-path <unwrapped> --module-dir Media/Modules --module-dir Media/Plugins <unwrapped>/eboot.bin app.exe, with app0 holding the dump's other folders.
3. Run app.exe with ANYPS5_GPU set to the 5090.

[klosixgaming-arch]

Profile from a 15-minute session (APS5_TRACE_SUMMARY_EVERY=10, APS5_PROFILE_GPU=1). Build: 3ee50f1, which is #6571 plus a local APR diagnostic. Windows 11, RTX 5090 (617.42), Ryzen 9 9950X3D. The output is 3840x2160.

Title screen and menus (first ~800 s): ~25 presents/s, frame p50 38 ms. The only queue worker is 98% busy, and most of each frame is Driver.Packet.device_lock: 27 to 29 ms of a 38 ms frame. The GPU is busy only 1.5 ms per present.

In game (field, East Shipping District), 15 intervals:

	per frame
frame time p50 / p90	123 ms / 130 ms (~8 fps)
queue worker 0x0 busy	98.6%
Driver.Packet	122 ms
Driver.Draw.draw_enqueued	65 ms
Driver.Packet.device_lock	32 ms
Driver.Draw.shader_compile_and_link	8 ms
GPU busy	57 ms
draws	~87, totalling 55.7 ms of GPU time ([gputime] class draw)

The shader disk cache hits after the first run (0 misses in later intervals), so this is not compile stutter. The frame is limited mainly by the single queue worker on the CPU: about 65 ms per frame goes to preparing ~87 draws and about 30 ms to waiting for the device lock. The GPU time per draw (~0.64 ms at 4K) is high as well. The input lag of more than half a second matches frames queuing behind the worker. I have not narrowed it to a specific code path yet. A APS5_TRACE_SECONDS trace of the field is available if someone wants it.

The APR cannot open abort did not reproduce in this session, so I still have no Windows error code for it.

AI-assisted: yes. Written with AI assistance and reviewed by me.