XashPS5 (PPSA19111)

[Lorkydey]

Existing thread

• Searched Game bring-up for XashPS5 and PPSA19111; no existing thread found on any platform.

Title and content version

XashPS5 (PPSA19111), public v1.3 release. sce_sys/param.json reports titleName: XashPS5 V1.3 and contentVersion: 01.030.000.

First report: Windows

• Baseline: main 1acfa7f.
• Follow-up: 6d5ee52b7, based on main 2f4cc89, with only the inactive vertex point-size fix from this branch.
• Windows 11 Pro 26100, NVIDIA GeForce GTX 1660 SUPER, Windows driver version 32.0.15.5585, WinLibs MinGW-w64 GCC 15.2.0 r7, Ninja, Release, BUILD_TESTING=ON.
• Furthest point: presentation is attempted, but fails before a usable menu or gameplay is reached. No FPS measurement or successful input/audio validation is claimed.

Current blockers

• fix(agc): accept inactive vertex point size on non-point draws #4715 fixes the inactive point-size rejection described below. Its proposed merge order is after feat(agc): vertex clip and cull distances #423.
• After fix(agc): accept inactive vertex point size on non-point draws #4715, the dynamic s_setpc_b64 shader stop below remains; no menu or gameplay is claimed.

On the baseline, all imports load and the engine initializes its GPU/textures before presentation stops at:

clip distances, layer, viewport or auxiliary vertex exports is unsupported (register 0x207 = 0x01210000)

The captured draw is a triangle list (VGT_PRIMITIVE_TYPE=4), with VGT_SHADER_STAGES_EN=0x02012010 and PA_SU_SC_MODE_CNTL=0x00083a40. The rejected bit is USE_VTX_POINT_SIZE (bit 16).

The inactive point-size fix removes that stop. With that fix alone at 6d5ee52, the next observed stop is:

[videoout] presentation failed: shader 0x200220200 stage 1: ShaderRecompiler::Recompile: unsupported dynamic s_setpc_b64 at program counter 0x0000003c

The process exits with 0xc0000409 after 7.38 s, without a timeout. The menu and its Previews action have not been reached. The point-size fix has state and pixel-readback regression tests; this remains a blocked title.

#423 addresses clip/cull distances and overlaps the same validation lines, but does not admit bit 16. A local integration of its head 66e1669 onto main 1acfa7f still produced the 0x01210000 stop. That integration required resolving the debt-file split and moving its serialized request addition to version 19 because main already used version 18; this was not a run of the unmodified PR head. The successful point-size run above does not include #423.

How it was run

1. Extracted the public release and its unencrypted ELF payloads, retaining the title's bundled sce_module/libc.prx for relinking.
2. Relinked with relinker.exe --windows --windows-diagnostics --registry input/eboot.elf run/app.exe. No unused-import filtering or lazy binding was used.
3. Copied release resources to run/app0, runtime libraries to run/libs, and the Steam Half-Life 1 valve data into run/app0/valve, preserving the release's autoexec.cfg, userconfig.cfg and extras.pk3.
4. Ran app.exe from run with no game arguments and no rendering bypass variables. The normal run reproduced the failure above. A separate diagnostic run used APS5_DUMP_QUEUE=0 and APS5_CAPTURE_TRACE=1 to inspect the rejected draw.

No game assets are attached. Earlier missing-import reports for this title are in #1578. Linux was used for the focused regression tests, not for a title run.

AI-assisted: yes, OpenAI Codex.

[SandstromPL]

@Lorkydey with #4821 and #4840 merged, XashPS5's stage-7 stop from #4840's test run (unsupported dynamic s_setpc_b64 at program counter 0x3c on the mesh path) looks like the next step: the user-data tail is only linked for vertex and fragment programs. Are you planning that follow-up? If not, I'd be glad to take it. And if you can re-run current main, the stage-7 line from it would help.