Neptunia ReVerse (PPSA02721)

[BunnyFl]

Existing thread

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

Title

Neptunia ReVerse (PPSA02721)

Content version tested

01.000.000

AnyPS5 commit

main 6a8f224, and main with #10556 (head a3a64a4) merged on top (local merge 345056f699, no conflicts). For audio, also main with #10556 and #8098 (head c1929ab) merged on top (c448dc17f0, no conflicts). #8098 has since been merged into main (6c48f35, 2026-10-27); the runs below predate that and were not repeated on it.

System tested

Windows 11 Pro 10.0.26200, AMD Radeon RX 9070 XT (driver 32.0.32015.7029), Ryzen 7 5800X3D, GCC 15.2.0 (WinLibs 15.2.0posix-14.0.0-ucrt-r7), Release.

Furthest point

Opening story scenes (with #10556)

What is shown

Runs from no save data, with scripted pad input (Cross every 3 s), up to 330 s each.

• main: three runs abort before the first frame (0 presents), 13-16 s after start, on sampled image descriptor uses an unsupported format (T# ffffffff ffffffff 00000019 c0017901 00000342 c8000000 c0064900 0000062f).
• With fix(shader): decode a T# with an unset format field as null #10556: four runs. All four stop 271-285 s after start on AGC graphics: guest sampler descriptor requests a LOD bias beyond the device limit, at mostly 550-600 presents per 10 s before that (350-601 overall). No log has an is not implemented line. The fourth run dumped a frame every 300 presents, and those frames show the logo animation, the CRIWARE logo, the Caution screen, the title menu (NEW GAME / CONTINUE / CONFIG), the tutorial help screens, narration over a world map, and then dialogue scenes whose name labels read Purple Heart, Black Heart, White Heart and Green Heart. Its last frames before the stop are black.
• Audio: none with main + fix(shader): decode a T# with an unset format field as null #10556; it plays with fix(audioout2): keep each port's grain when ports share a buffer #8098 merged on top (heard by me). With APS5_TRACE_AUDIOOUT2=1 the title sets one PCM buffer on all three of its AudioOut2 ports each grain. Without fix(audioout2): keep each port's grain when ports share a buffer #8098 the mixed output's peak is 0.000 for the whole run (143 one-second summaries); with it the peak is above zero in 125-129 of the 140-142 summaries of each of three runs.

Current blocker

• main: the unsupported-format T# above. fix(shader): decode a T# with an unset format field as null #10556 fixes it.
• With fix(shader): decode a T# with an unset format field as null #10556: guest sampler descriptor requests a LOD bias beyond the device limit (Sampler.cpp:12). Per @mixyt on fix(shader): decode a T# with an unset format field as null #10556, the title binds a bias of +31.67, while the S# field encodes up to +/-32 and maxSamplerLodBias is no more than 16 on any Vulkan device. mixyt has a follow-up prepared that clamps the bias to the device limit. No pull request for it is open yet (checked 2026-10-28).
• Audio: fix(audioout2): keep each port's grain when ports share a buffer #8098, which made it play in the runs above, is now merged into main (6c48f35). Without it the title gives one buffer to all its AudioOut2 ports and main played the last port's silent contents on all of them.
• Open pull requests the title needs: fix(shader): decode a T# with an unset format field as null #10556 (the runs used its head a3a64a4; its head is now 556257c).

How it was run

relinker --windows --registry src/eboot.elf app.exe

The dump's eboot.bin and its 2 modules (sce_module: libc, libSceNpCppWebApi) are SELF containers, unwrapped with tools/unwrap_self.py first; the .esbak copies and fakelib are left out. On main, tools/import_audit.py (with --modules sce_module) reports 0 absent imports and 1 throwing stub (sceAgcUpdateInterpolantMapping_0100). The runs used APS5_NO_PAD_INPUT=1 with APS5_PAD_INPUT_FILE written by tools/pad_input.py, plus APS5_PROFILE_DRAW=1 APS5_LOG_TIMESTAMPS=1 APS5_TRACE_SUMMARY_EVERY=10 and APS5_DUMP_FRAMES for the pictures below.

Dumped frames of one run with #10556 (one every 300 presents), in order:

Conversion, the runs, log reading and this report were done with AI assistance (Claude Code); the screens above come from frames AnyPS5 dumped during the runs. The audio result was heard by me.