Repository navigation
ARMORED CORE VI FIRES OF RUBICON (PPSA06774) #13424
Tehshortbus
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.
Status: boots, menus and intro movie at 60 fps, audio works, the first mission is playable at about 9 fps. No stop seen in 5 runs of 182 to 462 s, all exit 0: 3 on
main, 2 with one PR each (below).Build:
mainda026c7065, no PRs.Host: CachyOS Linux 7.3.0-4-cachyos, AMD Ryzen 7 5800X (16 threads), 64 GiB, NVIDIA GeForce RTX 4070 SUPER (
Selected GPU name=NVIDIA GeForce RTX 4070 SUPER), driver 615.78.08, AC power. Wayland with XWayland.Dump and conversion:
eboot.bin,sce_module/libc.prxandsce_module/libSceNpCppWebApi.prxunwrapped withtools/unwrap_self.py, converted withrelinker --registry input/eboot.elf app.elf. 2 converted modules.ampr_emu.indexat its root but noampr_assets-*.pak; every file the title opened was present.sce_sys/playgo-scenario.jsonis present; noAPS5_PLAYGO_UNLISTED_CHUNKSneeded.tools/import_audit.pyfailed on thismainwithFAIL: more than 256 preprocessor branch combinations in one function(the stub scanner, not the title), so no audit.Command:
APS5_PROFILE_DRAW=1 APS5_PROFILE_GPU=1 APS5_LOG_TIMESTAMPS=1 APS5_TRACE_SUMMARY_EVERY=10 setarch x86_64 -R ./app.elf, default in-game settings. First run with empty caches, later runs warm.Notes from the log:
sound/pck/enus/*.bnkopens fail, but the title then reads its banks fromsound/pck/bnk.pck(an AKPK package whose index lists them under the language-neutral entry), music streams fromstm_bgm.pckand voices fromstm_voice_enus.pck. The misses are the title's loose-file probe.NPC001.bnkis in neither place.[trophy]: 30 trophies read;uds00.ucpusescollectionandeachLatestaggregations, which are reported as unsupported.Frame rate in the first mission (warm caches, standing still, 4 × 10 s):
[gputime]): draws 31.7 ms, shadow publish 4.1 ms, storage write-back 1.2 ms, fills 1.0 ms, DCC clears 0.28 ms.PRs tried on the same scene: #12478 (bypass never starts here, no change), #10844 (pre-sync waits shorter, under 3 % of the worker, no change). #5609 targets this case but needs a rebase.
Open: no open PR changes the frame rate here yet.
AI-assisted: yes (Claude Code). The interpretation of the draw-cache misses is AI-generated; the numbers are measured.
All reactions