Repository navigation
Aphelion (PPSA13446) #18128
arthur-paraibano
started this conversation in
Game bring-up
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Existing thread
Title
Aphelion (PPSA13446)
Content version tested
01.003.000
AnyPS5 commit
706573b (main) for the gameplay stop, with the executable converted by the relinker of e1f012c; title-screen measurements below on e1f012c. No pull requests on top, no local changes.
System tested
Windows 11 Pro (10.0.26200), AMD Ryzen 7 9800X3D, 32 GB RAM, AMD Radeon RX 9070 XT (AMD proprietary driver 26.8.1 LLPC, 32.0.31041.1004, Vulkan 1.4.349), 3840x2160 output, WinLibs GCC 15.2.0 (ucrt-posix-seh r7), Release
Furthest point
In game
What is shown
Photosensitivity warning, content warning, then the title screen with the animated snow background and "PRESS ANY BUTTON"; all three look correct in the dumped frames. Input works: the title screen, the menu and continuing the save respond. Gameplay starts, and the title stops 49 to 69 s after start (see Current blocker). Audio was not checked.
Title screen without input: two runs on e1f012c (120 s and 150 s) and one on 706573b (20 min), no stop. 119 to 143 presents per 10 s (about 13 FPS), frame time p50 74 to 76 ms, p99 80 to 93 ms.
anyps5-report.json:averageFps 12.376over 139.7 s (e1f012c, warm caches, first draw 8.5 s, first present 9.7 s) andaverageFps 13.473over 1183 s (706573b);missing.functionsempty,shaders.unsupportedempty, 88 shaders compiled with 0 failed, no rejected image formats.Current blocker
Update 2026-11-16,
maine4caa07: the depth refusal below (#18127) is fixed by #19846, merged. On currentmainwith the colour and depth changes listed here, gameplay now stops 61 to 123 s after start, in two runs, on a shader recompiler refusal:Rect-list SPIR-V: custom per-vertex interpolation is unsupported(exit 0xC0000409). Not investigated yet.Open pull requests and local changes the title still needs past that point:
mainstops at about 168 s ona shader buffer access ... overlaps multisampled color target ..., whose samples are only on the host.With all of these on an older
main(f59a4a6), the title ran to the end of a 600 s window; on that build an intermittent GPU page fault (device lost) was also seen in about 3 of 4 runs, cause unknown.Previous blocker (fixed by #19846): Gameplay stops on a refused native multisample depth import: a 4x
D32_SFLOATtarget with no stencil plane and no HTILE whose plane does not hold one value is sampled by a compute shader, and the AGC driver throwsnative multisample depth initial import requires pipe-aligned or depth-only HTILE without VRS(exit 0xC0000409). Tracked in #18127. Seen four times, on e1f012c and 706573b.On e1f012c with empty pipeline caches, the first gameplay run instead ended at the engine's watchdog (
GameThread timed out waiting for RenderThread after 120.00 secs, exit 0xC0000005 from the title's own abort) while the AMD driver built compute pipelines of 229k to 281k SPIR-V words in 2.5 to 3.5 s each; later runs passed that point.On the title screen the frame rate is about 13 FPS, measured with
APS5_TRACE_SUMMARY_EVERY=10andtools/trace_digest.py(e1f012c, warm caches):[gpu] storage image ... destroyed with GPU results pendingis printed 4 times at shutdown in each title-screen run. The first start with an old shader cache had its first present at 46.5 s ([pipeline-journal] loaded 292 pipelines, 305 modules (6.8 MiB, 18237 ms); replay ... in 22247 ms); the run after it loaded the same journal in 32 ms and replayed it in 166 ms.Analysis (AI-assisted, not measured): the title screen is bound by the CPU side of the AGC driver, not the GPU (queue worker 96-97 % busy, GPU about 42 ms per frame,
Driver.Packetthe largest phase); the dispatch cache evicts 68 % of dispatches on a static screen, which looks like thrashing; the 18 s journal load of run 1 may be a cold disk, antivirus scan or cold AMD driver cache, not confirmed.How it was run
Converted with
relinker --windows eboot.elf PPSA13446.exe(relinker of e1f012c) from the unwrapped dump, with its fivesce_module/modules converted alongside (libc,libSceFontGsm,libSceJobManager,libSceNpCppWebApi,libScePfs); no extra options, notools/titles.jsonentry.app0/holds the dump's files,libs/holdsbuild/core/libs/libs/*.prxpluslibgcc_s_seh-1.dll,libstdc++-6.dll,libwinpthread-1.dlland the mimalloc DLLs. Not converted with--registry, sotools/import_audit.pywas not run.Environment for every run:
plus
APS5_DUMP_FRAMES_EVERY=120or20for the title-screen runs, andAPS5_DUMP_FRAMES_EVERY=60 APS5_CRASH_INDIRECT_MEMORY=1 APS5_TRACE_EXIT=1for the gameplay runs.Default in-game settings. Title-screen runs: no input, stopped by closing the window. Gameplay runs: played by hand from the title screen into the save; five runs, the first with empty pipeline caches ending at the watchdog, the other four ending on the depth refusal (two of them with a temporary local diagnostic print, see #18127). Before the first run, AnyPS5
shader_cache/held 548 MiB from an earlier session; the driver cache state is unknown.Tested on 2026-11-10.
All reactions