Repository navigation
Demon Slayer -Kimetsu no Yaiba- The Hinokami Chronicles (PPSA03490) #7257
klosixgaming-arch
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.
Existing thread
Title
Demon Slayer -Kimetsu no Yaiba- The Hinokami Chronicles (PPSA03490)
Content version tested
01.000.000
AnyPS5 commit
main at 66702f1 (merge of #7060), unmodified. For the PlayGo setting see #7255.
System tested
Windows 11, WinLibs GCC 15.2.0 (x86_64-posix-seh, ucrt, r7), RTX 5090 (NVIDIA 617.42), AMD Ryzen 9 9950X3D, 3840x2160
Furthest point
In game
What is shown
The logos play, the "autosave" notice, the main menu, a rendered intro movie and a fight in game (around frame 2800 of the fight, as counted by the player) were reached with no emulator crash in a run of about 16 minutes (957 s of log); the player closed it because it was too slow. Player-observed frame rates: about 0.8 fps on the notice, 18 fps after the menu, 16 fps in the intro movie, 0.2 fps for a long freeze at a scene switch and at a face loading in (0.19 fps), then 12 fps again. In the [present] lines the best 10 s window shows 220 presents (GPU busy 30.2 ms, idle gaps 15.2 ms per frame); slow windows show 1 to 3 presents per 10 s with idle gaps of 2,676 ms per frame and GPU busy 0.0 to 8.1 ms. A first action in game (for example the first attack) stutters, the second time is smooth. Audio and the controller were not checked separately.
Current blocker
Performance only, and the GPU is idle during the stalls rather than busy. Numbers from the log: the [draw] driver phases line shows recompile at 182.9 us per draw packet (capture 16.9 us); the [capture] line counts 1,662,668 captures with 865,780 pages fetched; the [recorder] line counts 19,303 syncs that waited 33.8 s for the GPU in total (pending write 5,094 syncs, 25.6 s); the [texture] lines report 2,200 textures in the run (994 copied from storage images), with allocate+image at 1,569 ms in total. The log also contains 6 lines "Shader 0x200cbc0000 ... has unsupported registered shader type 8; it is not prepared at registration" (0x200cbc0000, 0x200cbd0000, 0x200cbf0000, 0x200cc00000, 0x200cc10000, 0x200cc20000); I did not check whether anything draws with them. I have not found the cause of the freezes. Before the PlayGo setting was corrected the title aborted at startup (see How it was run).
How it was run
eboot.bin and the six sce_module prx files (libSceFace, libSceFaceTracker, libSceJobManager, libSceNpCppWebApi, libScePfs, libc) are SELF files; each was unpacked with tools/unwrap_self.py into an anyps5-elf folder. Relink: relinker --windows --windows-diagnostics --trace-imports --registry --sce-module-path /eboot.bin app.exe, with an app0 folder that links apk and sce_sys to the dump, holds eboot.bin, and holds ue4commandline.txt (contents: ../../../apk/apk.uproject, which the dump does not include). Environment: ANYPS5_GPU=5090, VK_LOADER_DRIVERS_DISABLE=dzn, APS5_LOG_TIMESTAMPS=1, APS5_PROFILE_DRAW=1 and APS5_PLAYGO_UNLISTED_CHUNKS=absent. The dump has no playgo-* files in sce_sys, so without the variable PlayGo stops with the missing-metadata error, and with APS5_PLAYGO_UNLISTED_CHUNKS=installed the title's chunk-ID walk never ends and startup aborts with "Ran out of memory allocating 5565054976 bytes with alignment 65536"; that is documented in #7255.
All reactions