Repository navigation
AEW: Fight Forever (PPSA09351) #21560
PeterGrieffindor
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
AEW: Fight Forever (PPSA09351)
Content version tested
01.010.000
AnyPS5 commit
maine4caa07950with #20827 merged on top (local merge5500170ef8), built here. Unchangedmainshows the picture upside down and stops within a minute, both since #16053 (details); #20827 puts the three starting values back.Release v0.1.5 also runs the title (training mode, about 35 presents per second there on 15 November).
System tested
Windows 11 Pro (build 26300), WinLibs GCC 15.2.0 r7 (Release), Ryzen 7 9800X3D, Radeon RX 9070 XT (AMD driver 26.11.1), 32 GB RAM.
Furthest point
In game
What is shown
An Exhibition 1-on-1 match (Kenny Omega against Jon Moxley, Dynamite arena with a crowd) starts and plays. Scripted input (
tools/pad_input.py, CROSS every 2 to 3 s) goes from the title screen through the menus and the first-time tutorial dialogs into the match; the sheet below is nine screenshots of one such run, 25 s apart. The wrestlers, the referee, the crowd, shadows and lights are drawn; no visible fault was seen in them.On
0694c4cbd6with fix(agc): start the viewport control and color masks at the AGC register defaults again #20827 the same script played a match to its end: the winner screen, then back to the wrestler selection, with no stop.Menus hold 60 presents per second, the title's cap.
Frame rate in the match at 3840x2160, read from the window title once a second, with no trace switch set:
APS5_NO_COPY_HLE=1engine.inibelowengine.inibelowThe measurements, traces and where the default loses the time are in perf: AEW: Fight Forever (PPSA09351): the copy-kernel HLE waits for the GPU on queue 0 every frame; APS5_NO_COPY_HLE=1 takes a match from 53 to 60 fps #21539.
The title reads Unreal ini files it finds in
download0/aewfightforever/saved/ps5/. Thisengine.inithere is what the second and third rows use; the screenshots of the standing scene show no difference with it:The title rewrites the file and keeps these lines.
Played by hand on 15 November (v0.1.5, and
main6453229902with the three values put back): audio plays in the menu, an Xbox Series X controller works in the menus and in training (Pad: connected game controller (slot 0): Xbox Series X Controller), and training mode (CM Punk against Moxley in the empty training arena) is playable.Save data mounts and unmounts without an error (
[SAVEDATA:native] ... -> 0x00000000).The log of the played match (linked under How it was run) has no call to an unimplemented function.
The trophy loader reports unsupported rules in
sce_sys/uds/uds00.ucp, for example:Current blocker
No stop in the match. The title needs #20827, which is open.
Known problem, not investigated: an intermittent stop while the title loads, always on the title's
FAsyncLoadingThreadwhere a thread was named. It was logged in 3 of 33 runs on 15 November: on v0.1.5, entering training,FATAL: unhandled exception 0xc0000005,rip app.exe+0x1e022ea,read of 0x00000000fffffffc, with the buffer atrbxholding only0xffbytes; on twomain-side commits (6577ac79adafter 157 s,6fd2999417after 61 s) the title's ownabort()(guest abort() (int 0x45)inlibc.prx.guest.prx+0x148e0). It still happens:FATAL: unhandled exception 0xc000000586 s into one scripted run on0694c4cbd6with #20827, and one scripted run one4caa07950with #20827 ended while loading. The scripts restarted such runs and their logs were not kept, so how often was not counted on these builds. The next start of the same build gets past the same point.Not tried: other match types, more than two wrestlers, other arenas, the career mode, online play and the PlayStation Store button.
Import audit with
tools/audit_dump.pyagainst the libraries of0694c4cbd6with #20827: 14,200 references, 3,620 unique imports; 2,003 implemented, 91 partial, 0 stub, 0 absent, 1,526 provided by the title's own modules. The partial ones are 72 inlibSceJson2, 12 inlibkernel(getsockopt,setsockopt,sceKernelAioDeleteRequest,sceKernelAioPollRequests,sceKernelGetModuleInfoForUnwind,__tls_get_addr),sceAjmBatchStart,sceAjmInstanceCreate,sceHttp2Init,sceNetGetsockoptandsceUserServiceGetEvent.How it was run
The dump holds
eboot.bin.esbakand*.prx.esbakcopies beside the modules, whichtools/audit_dump.pytakes for modules and stops on (#20280). The audit and the conversion ran on a copy holding onlyeboot.bin,sce_module/*.prxandprx/*.prx:The title ships
libc.prx,libSceFace.prx,libSceFaceTracker.prx,libSceFontGsm.prx,libSceJobManager.prx,libSceNpCppWebApi.prxandlibScePfs.prxinsce_module/, andeossdk-ps5-shipping.prxinprx/; all eight were converted.Layout:
app.exe,libs/, andapp0/holding the originaleboot.bin,ue4commandline.txt,contentids.json, the original and converted modules, and directory junctions to the dump'saewfightforever/andsce_sys/.APS5_PLAYGO_UNLISTED_CHUNKSis required for this dump, which has neithersce_sys/playgo-chunk.datnorsce_sys/playgo-scenario.json. Without it the title stops 35 s in:=absentand an explicit chunk list were not tried. The other three variables are the ones the frame rates above were measured with; the default column leavesAPS5_NO_COPY_HLEout. match-run-5500170ef8.log is the log of the played match in the third row, withAPS5_NO_COPY_HLE=1; local paths in it are replaced by<run>and<out>.AI-assisted testing and report: yes, Claude Mythos 5.2.
All reactions