DOOM + DOOM II (PPSA21444)

[JosepBove]

Existing thread

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

Title

DOOM + DOOM II (PPSA21444)

Content version tested

03.000.000 (UP1003-PPSA21444_00-OSIRIS2000000000)

AnyPS5 commit

main 84b01f9, alone and with #5916.

System tested

Windows 11 Home (26200), AMD Ryzen 9 9950X3D, NVIDIA GeForce RTX 5090 (NVIDIA 616.92), WinLibs GCC 15.2.0 r7, Release

Furthest point

In game

What is shown

On main the title ends during audio set-up, before any frame, with exit code 0xC0000409. It redirects stderr to app0/stderr.txt, so the message only shows under gdb:

sceNgs2RackGetVoiceHandle: invalid argument

It is the same stop as Heretic + Hexen (#5912) and STAR WARS: Dark Forces Remaster (#4310), all on Nightdive's KEX engine: a one-voice reverb rack (0x2001, max_voices = 1), voice 0, then voices 1 to 3.

With #5916 the title logs kexAudioNgs2::InitReverbVoices: sceNgs2RackGetVoiceHandle failed (0x804a8305) four times and continues. A 120 s run with no input: Audio system initialized, shaders loaded, Running game session, the attract demo in game at 30 s (Master Levels, "Level 1: Attack", pickups and status bar drawn), the DOOM main menu with the controller prompts at 110 s, all at 120 FPS on the window counter (12,800 frames by 110 s). Audio and controller input were not checked.

Two intro movies do not play: kexMoviePlayerAV::StartVideoStream: sceAvPlayerAddSource failed (0x806a0002) for NightDive and Kex. This dump has no movies/ directory, so the files are absent and the title goes on without them; not an AnyPS5 stop as far as this run shows. The online parts fail as expected without a network (Playfab::signin -- Login failed, bnet_error DnsError).

Current blocker

None with #5916. On main: sceNgs2RackGetVoiceHandle: invalid argument before the first frame.

How it was run

The dump's executable and seven modules (sce_module/libc.prx, sce_module/libSceNpCppWebApi.prx, and libcohtml.Prospero.prx, libPlayFabMultiplayer.prx, libRenoirCore.PS5.prx, MediaDecoders.Prospero.prx, party.prx under prx/) were already plain ELF files; both sce_module/ and prx/ stay beside the input.

relinker --windows --registry source/input.elf run/app.exe

app0/ is the dump folder, libs/ the built libraries with the three MinGW runtime DLLs, ANYPS5_GPU='RTX 5090' to pick the discrete GPU. Nothing else beyond the usage guide. tools/import_audit.py cannot read this dump's libc.prx and libcohtml.Prospero.prx (section header table past the end of the file), but the relinker converts them and the title loads with no absent import.

[JJenkx]

Current-main graphics follow-up, 2026-10-15: #6438 fixes the duplicated/outlined collection-menu text observed on my RX 6650 XT. This supersedes my earlier cache-disable investigation: the submitted change preserves caching, pipelined draw commits and generic HTILE support.

A newly constructed depth/stencil image already has its initial clear. Repeating the depth-only clear can discard its initialized stencil on Radeon at 1024x1024. The fix uses the constructor clear for newly cleared planes, preserves inheritance of uncleared planes and keeps the pending-clear path for reused attachments. This is demonstrated by a procedural depth/stencil regression, without title assets. The lower-level driver cause has not been isolated outside the emulator sequence.

Matched native menu comparison on main ff23657f3 plus common boot prerequisites #5916/#6348 and the private dependency repair below: control 53f2e521ce26 has corrupted description text at 30 seconds; adding only #6438's production/test diff, integration 4cb5da916d57, has readable text at 30 seconds. The fixed run continues to E1M1: Hangar, with the rendered level, weapon and readable status bar visible at 170 seconds. Both use the same machine, warm cache, 2972x1672 window and pipelined draws; no forced focus or cache-disabling switches. Both finish at their scheduled watchdog (124), restoring the host configuration. This does not establish movement, firing, audio quality, sustained gameplay or a measured speedup. Earlier runs also showed clear text, so this is a reproduced configuration-dependent regression, not a claim that every previous run was corrupted.

#6438 is refreshed onto main 92e50418f, public head 9870b37f3; its production/test diff is unchanged from the native comparison. Full Linux Release libraries plus five focused depth test executables build after refresh. All five tests pass on Radeon and lavapipe, zero test skips. Restoring main's production source fails the unchanged new regression on Radeon; intentionally breaking either cleared-plane priority or uncleared-plane inheritance fails it on both drivers. All faults were removed before the final rebuild/tests. #6438 lists #6423 first for merge ordering because both edit the same clear block; it will need a rebase after that diagnostic change merges.

The title still needs the separately listed NGS2 work to pass audio initialization; the graphics fix alone does not supply that. Native and standalone validation builds use private SPIRV-Tools teardown repair b326807c0c9d8682b697fdad3153f77fc2642f5f, excluded from the public PR. These runs do not independently validate the official dependency pin. After-refresh standalone validation head: 338ff19ed, differing from the public source only by that pin.

Hardware: Ryzen 9 7950X, RX 6650 XT 8 GiB, RADV Mesa 26.2.4-arch1.1, Linux 7.2.9-arch1-1. Software-driver tests: lavapipe Mesa 26.2.4 / LLVM 23.1.1. No title assets, screenshots, binaries or raw title logs are attached. AI-assisted implementation and analysis, native tests on my machine.

---

Historical initial bring-up (retained for context):

Linux / Radeon follow-up for PPSA21444 v03.000.000, tested 2026-10-15.

On main 4e9a2a871915a507ff52679d082a9abf3b80f2d2 plus the common private dependency repair described below, DOOM aborts during audio initialization with sceNgs2RackGetVoiceHandle: invalid argument (exit 134), before its first window sample. With #5916 applied, it logs the expected reverb voice-index errors, initializes audio and renders its title/menu. A 150-second no-input run completed at its planned watchdog (124).

A further controller-driven run with #5916 and the handle-error follow-up #6348 entered E1M1: Hangar. The log records level setup/activity start, and the 130-second compositor capture shows the rendered level, weapon and status bar. Eleven scheduled Cross pulses navigated the title, collection, new-game, episode and difficulty menus. The window counter stayed around 119.9 FPS after loading, including the final in-level sample (15,070 completed frames at 131.62 seconds). This is the window counter from a bring-up run, not an official traced performance comparison or a measured speedup caused by #6348. Movement, firing, listening quality and a full level playthrough were not checked.

Ryzen 9 7950X (16 cores/32 threads), 128 GiB, RX 6650 XT 8 GiB, RADV Mesa 26.2.4, Linux 7.2.9. Native window 3072x1728. Existing converted executable and bundled modules, current built system libraries; loader and mapped driver checked against the expected clean revision. Host imports stayed enabled: import budget 16384 MiB, temporary udmabuf size limit 8192 MiB, pipelined draws depth 8. No APS5_NO_HOST_IMPORT fallback, timer-slack adjustment or forced-focus helper was used. The window became active normally.

Public integration source is 14bcda2991fa7693d42b011d2b62ce6d41efc579 in #6348, containing #5916's three commits with original authorship and this follow-up's two commits. Native build 64055de47ce2e05c83ededfc821eda995746e642 differs only by private SPIRV-Tools teardown repair b326807c0c9d8682b697fdad3153f77fc2642f5f. This does not independently validate the official dependency pin. #5916 already removes the audio initialization stop; #6348 corrects separate invalid-system/rack error values and lookup precedence, without claiming another title unblock.

The final run reached its planned 140-second timeout (124; wrapper observed completion at 141.96 seconds), restored the udmabuf setting to 64 MiB and left no title process. An earlier 180-second input attempt received SIGTERM in the wrapper at about 158 seconds, cause unknown; it is recorded as interrupted and excluded as a completed run. Another completed attempt reached difficulty selection before its deadline; the final run adds enough menu confirmations to enter the level.

Both NGS2 libraries and all twelve current NGS2 tests build and pass on the final integration, zero skips. Restoring either old handle-error value or the old validation order independently fails the unchanged new regression executable; all three faults were removed before the final tests/title run. Full Linux/Windows CI remains pending maintainer action. No game assets, screenshots, audio, shader binaries or raw title logs are attached.

AI-assisted implementation, orchestration and analysis; native runs on my machine.