PRAGMATA (PPSA02530)

[hugodearaujo]

Existing thread

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

Title

PRAGMATA (PPSA02530)

Content version tested

01.200.000

AnyPS5 commit

main bb04a4f with the open pull requests listed below (and, for the furthest run, an earlier integration build on main e61a7c7 with the same fixes); main 8f55414 with #10935, #10304, #10497, #12898, #4449, #9999 and #11580 (update of 2026-10-31).

System tested

Arch Linux (kernel 7.2.11), GCC 16.2.1, Release; AMD Radeon RX 9070 XT, RADV (Mesa 26.2.6)

Furthest point

In game

What is shown

The title screen, New Game and the difficulty menu, then the opening film (an AvPlayer video; the title plays its audio itself), then the real-time scene after it (astronauts walking in, with the opening credits), a black loading screen of about 2.5 minutes, and the next real-time scene in the hangar, with its dialogue ("All right, let's do it. Just get in and get out."), about 1000 s after start. With #4449 the title then reaches its first gameplay, the "Basic Movement" tutorial, and the character can be walked through the hangar area. On main 8f55414 with the pull requests below, from that save, the character went on through the round orange door, the "Delphi Corporation" lobby and the turnstiles, opened door A2 and entered the room with forklift "07", about 1200 s after start. The real-time scenes render at about 1-3 FPS. Audio was not listened to (SDL dummy driver, or a PipeWire null sink for the audio runs).

Current blocker

On main bb04a4f with the pull requests below, the loading after the credits used to end at about 1314-1359 s with a GPU hang (ring gfx_0.0.0 timeout in the kernel log, then vkQueueSubmit recorder: Vulkan result -4), with memory still free. The cause is #4449 missing on current main: the title maps its direct memory in pieces at scattered memfd offsets, and without that PR 7,000 to 12,000 host imports fail (host import of ... failed at vkAllocateMemory (-1000072003); falling back to copies) against 1 with it. With #4449 (merged by hand, it conflicts with main) and #12081 (now merged), the title reaches its first gameplay, the "Basic Movement" tutorial, about 770 s after New Game, and walks through the hangar area for about 1,000 s more without stopping.

On main 8f55414 the title also needs #11580: without it, it aborts at 276 s, before the title screen, with compute shader 0x2111c18500: reads or writes a converted float image with 16-bit data, which is not implemented. With the whole stack, a 2000 s run from New Game had no abort, hang or kill, and reached the "Basic Movement" tutorial at about 1760 s. Walking on from there, our desktop memory guard stopped the run at 1280 s on entering the forklift room (vram_free=1249 MiB gtt_free=3020 MiB; the process held 15.5 GiB of VRAM and 11.0 GiB of GTT, of which 9.9 GiB were shared imports, and the device-local heap held 2 GiB of sampled textures and 3.5 GiB of storage images). Presents fell from 8 to 4 per 10 s as memory filled. The process also peaks at 23 to 25 GiB of host memory (RSS) about 80 to 120 s after start, during start-up, then drops to about 7 GiB.

Next limits: memory at 3840x2160 (with #4449 the title's imports hold about 10 GiB of GTT, and VRAM is full during the loading after the credits), speed (about 1-3 FPS), and translucent rectangles and green and red quads over the gameplay scene, which open PR #9999 removes (same save and route: 4 of 9 screenshots show them without it, none with it; a cleared surface was read as the raw bytes of the image sharing its memory).

Stops on the way, each with the build that got past the one before it:

1. scePadOpen at 5 s (PPSA02530 stops at scePadOpen with port type 3 or 4 since #10361 #10904): fix(libScePad): return an error for unsupported port probes #10935.
2. AGC graphics: guest buffer has no GPU owner on current main: fix(agc): keep buffer owners when a staged address space dissolves #10304.
3. AGC graphics: sample-preserving guest publication of native multisample depth is not implemented on current main: fix(agc): refuse writable buffer feedback with native depth #10497.
4. The opening film plays in slow motion and the title ends it by its own clock after showing 33 to 73 s of its 245 s: the title takes about one frame per rendered frame at 1-3 FPS, and the decoder only advanced when a frame was taken. fix(libSceAvPlayer): keep synced video at the presentation clock when the title takes frames slowly #12898 keeps the video at the presentation clock; the film then ends on time (end of file).
5. The real-time scene after the film stops on one frame, on one of the credits, while rendering goes on and the title still takes input: its sound engine pushes one grain per call from a feeder thread and wakes its renderer after each push, and a blocking sceAudioOut2ContextPush returned while the queue was still full, so the mix advanced four grains per rendered frame and the scene, which follows the audio, stalled. Fixed by fix(libSceAudioOut): return from a blocking AudioOut2 push once the queue has room #12954 (merged); with it the sound engine's context pushes 46.9 grains per second and the scene runs past the credits (also checked with both AudioOut2 contexts on a PipeWire device).
6. Intermittent: some runs stop at the switch from the film to the real-time scene; cause not found yet.
7. Intermittent, on the title screen with a save: MSAA_ENABLE on a single-sample target for primitives other than triangles and rectangles is unsupported (register 0x242 = 0x00000002). Fixed by fix(agc): allow MSAA_ENABLE single-sample lines #12081 (merged).
8. The GPU hang in the loading after the credits on current main (see Current blocker): feat(libSceAgcDriver): import scattered direct memory as a dma-buf run list and probe its coherence #4449.

The builds also carry #11580, #7844, #6565, #7446 and #6782 from the integration build these runs started from; their individual effect on this title was not isolated, except that #6565 removed the still pinned notices seen around the film.

How it was run

1. relinker with the title's modules, through a local script; a save data directory copied from an earlier run.
2. Pad input through the emulator's input file (APS5_PAD_INPUT_FILE): Cross on the title screen, New Game, Standard difficulty, then Left and Cross to confirm; nothing after that.
3. APS5_TRACE_SUMMARY_EVERY=10 and, for the audio runs, APS5_TRACE_AUDIOOUT2=1.

AI-assisted: yes