[Regression] FINAL FANTASY XIII [MRTC00003] Crashing to desktop again (#14822)

[Darkhost1999]

Quick Summary

Final Fantasy XIII started crashing again to the desktop. This time changing the date of regression as there was a moment when this issue was resolved or at least working.

Details

At this time it appears the broken log changes its behavior to do something with SYS where it didn't do this before.
Again the symptoms are that the cache used from working builds will continue working on master.
So to reproduce this issue users need to clear their cache. Boot the game and soon find that RPCS3 is closed.

Build with regression

RPCS3 v0.0.29-15753-fdb2fecd

Attach two log files

Broken.log
Working-Kinda.log

System configuration

Windows 11 pro
Ryzen 5 3600
GTX 1660 Super

Other details

What do you mean by kinda works?
There are 12 instances of

·F 0:02:03.346525 {SPU Worker 2} SIG: Thread terminated due to fatal error: Verification failed
(in file D:\a\1\s\rpcs3\Emu\Cell\SPURecompiler.cpp:3538[:3], in function analyse)

in the working log. The SPU DMA option only worked, before this regression, on 2nd boots and not on the first boots.
After the regression, FFXIII only works with the PPU interpreter again.

[AniLeo]

Possibly fixed by #14959, needs retesting

[Darkhost1999]

RPCS3.log
RPCS3 v0.0.29-15897-195e7e18 Closes to desktop

[AniLeo]

Linking PR that introduced the regression: #14822

[Darkhost1999]

Test of #15212
RPCS3.log
Some progress but ultimately

F {PPUW.1.7} VM: Access violation reading location 0x70a3b65d (unmapped memory)

did not close to the desktop

But here's a test after clearing LLVM cache and retrying
RPCS3.log
Closed to desktop

[elad335]

Test #15212

[Darkhost1999]

RPCS3.log
Closed to desktop

[elad335]

I fixed the SPU analyser btw. Not sure what is why the PPU crash, perhaps it's a concurrency bug in PPU LLVM.

[Darkhost1999]

F {PPU[0x1000028] Thread (white::sound::layout::load) [0x013d6d8c]} SIG: Thread terminated due to fatal error: Verification failed
(in file D:\a\1\s\rpcs3\util\serialization.hpp:135[:5], in function raw_serialize)
F {PPU[0x1000010] Thread (QixFiberWorker0) [0x013d6d8c]} SIG: Thread terminated due to fatal error: Verification failed
(in file D:\a\1\s\rpcs3\util\serialization.hpp:135[:5], in function raw_serialize)
F {PPU[0x1000027] Thread (white::sound::AsyncRegistrant) [0x013d6d8c]} SIG: Thread terminated due to fatal error: Verification failed
(in file D:\a\1\s\rpcs3\util\serialization.hpp:135[:5], in function raw_serialize)
F {PPU[0x100000d] Thread (CDevSd StreamHandler) [0x013d6d8c]} SIG: Thread terminated due to fatal error: Verification failed
(in file D:\a\1\s\rpcs3\util\serialization.hpp:135[:5], in function raw_serialize)

RPCS3.log

AMD Ryzen 7 7700X
Not closing to the desktop and fatal errors are after closing the game during the intro video

[elad335]

so this is an liblv2.sprx hle regression

[Darkhost1999]

That liblvl2.sprx HLE testing was an accident from other tests and currently seems to be appropriately resolved but not the root issue for the close-to desktop. I don't know what got me to the main menu past compiling the LLVM cache.

RPCS3.log

Here is a current log where I cleared all cache and right-clicked -> boot with default configuration. Shortly later closed to desktop.

[Darkhost1999]

Not sure why the PPU crashed, perhaps it's a concurrency bug in PPU LLVM.

Testing #15333 and the PPU crash appears to be gone now.
RPCS3.log

After that PR is merged there will be no settings other than default needed to boot FFXIII.
If for some reason anyone experiences a crash try using a clean cache.

[bigol83]

Yesterday i tested this Pr, it fixed this issue. Now that it is merged the game crashes again. Removed all caches, reset to default configuration, still crashes.

[bigol83]

It is weird how this PR #15333 fixed this issue but when it was merged it reappeared.

[bigol83]

If you want to go ingame you can only use Interpreter PPU decoder.

[AstroDot]

Hello,

FINAL FANTASY XIII [MRTC00003]

PPU Decoder:
LLVM - crashes RPCS3 on loading PPU
Static - works

RPCS3 Version: 0.0.32-16476-9ceb22a2 Alpha | master

[Darkhost1999]

Tested RPCS3 v0.0.32-16683-065a6437
RPCS3.log
Still an issue. To reproduce use a clean LLVM cache. The close to desktop happens when building a new cache not loading an old cache.

[bigol83]

With the latest main build it crashes a little later at 84/119, it creates a bit more functions compared to last week log

·! 0:00:56.213189 {PPUW.1.11} PPU: LLVM: 4176 functions generated
·! 0:00:57.684742 {PPUW.1.10} SYS:

but in the end still crashes to desktop

[elad335]

who can get stacktrace for it?

[bigol83]

who can get stacktrace for it?

I don't know how to do that, i would do it otherwise. I expected someone else to answer this request but It seems there is not enough interest

[Darkhost1999]

I've googled and searched, I've asked and asked.
All I ever have gotten answered is that stack trace is just another name for Call Stack. So I assume this is what is being asked for.
Additionally searching RPCS3 issues history when a stack trace is asked for 9/10 times nothing is provided. 1/10 it's a screenshot of the full visual studio debugging window after the exception occurred. So this should be accurate.

[bigol83]

Nice, thanks, hopefully it helps.

[TailsGuy]

I just tried this out in RPCS3, the LLVM recompiler unfortunately doesn't work still. Static works, but performance is... strange. You can't create a cache with any version from 2024 as far as I can tell.

EDIT: I can't get into game even with builds going back a year, I don't know if this is my PC but was it always an issue trying to get in game with the LLVM recompiler?

[bigol83]

I just tried this out in RPCS3, the LLVM recompiler unfortunately doesn't work still. Static works, but performance is... strange. You can't create a cache with any version from 2024 as far as I can tell.

EDIT: I can't get into game even with builds going back a year, I don't know if this is my PC but was it always an issue trying to get in game with the LLVM recompiler?

There was a small period of time when it booted fine with ppu llvm but then went back to crashing. Yeah, i know PPU performance is weird with Interpreter, it definitely is a lot slower. When i could go ingame and play it with PPU LLVM it was basically 60 fps locked while with Interpreter it struggles a lot.

[TailsGuy]

OK well, I figured out what build works with the game.
You need rpcs3-v0.0.29-15463, and you also need to enable "Accurate SPU DMA" and "Accurate XFloat".

However, this is a pretty old version of RPCS3. I'll load it up in the latest version and supply a log file.

[TailsGuy]

Well shit I was gonna put a log but I found somewhat of a security concern in logs instead

[Darkhost1999]

It's okay, the logs are in the original post

[TailsGuy]

No what I mean is I realized that in logs, you can see the user's names, and if they used their real name as their user name that's kind of an issue isn't it? I don't think I've seen that in PCSX2 or Dolphin or anything like that. That might need to get looked into.

[TailsGuy]

Here's my log but I had to edit that, that's concerning cause so many people have contributed.
0.0.32-16807 RPCS3.log
0.0.29-15463 RPCS3.log

[ReLyf47]

No what I mean is I realized that in logs, you can see the user's names, and if they used their real name as their user name that's kind of an issue isn't it? I don't think I've seen that in PCSX2 or Dolphin or anything like that. That might need to get looked into.

This is a user skill issue

If you put your RPCS3 folder inside the download folder or desktop ofc it will contain the windows install user name

Simply put it on C:/ or D:/ instead of under C:/Users/[Your Name]/[Any of the folder here]

[ReLyf47]

Or in your case somehow you put your emulators under document for some reason

[TailsGuy]

No what I mean is I realized that in logs, you can see the user's names, and if they used their real name as their user name that's kind of an issue isn't it? I don't think I've seen that in PCSX2 or Dolphin or anything like that. That might need to get looked into.

This is a user skill issue

If you put your RPCS3 folder inside the download folder or desktop ofc it will contain the windows install user name

Simply put it on C:/ or D:/ instead of under C:/Users/[Your Name]/[Any of the folder here]

Well yeah it's a user skill issue, nevertheless being clipped in log isn't something everyone has in mind when they literally make a Windows user. I sure didn't five years ago, when I was a child. But whatever I guess it's not really a problem worth the scale of rewriting how logs work. Nor is it necessarily the program's fault where the user places the emulator.

But still it wouldn't really hurt to substitute the username for something else, it's worth thinking about.

[TailsGuy]

Or in your case somehow you put your emulators under document for some reason

It's my business where I put my stuff, it's easily accessible, it's grouped with my other stuff, and that's how it's been for years.

[bigol83]

With rpcs3 0.0.32 1680 (actual master) this issue has been fixed. I managed to compile all PPU modules with LLVM.

[ItsLaguna]

Can confirm that it's working again with LLVM IN 0.0.32-16830.

[Behodar]

While this did fix the original issue, it's now freezing at "Loading PPU Modules, Progress: module 119 of 120 (1s remaining)" with 0.0.32-16830 on MacOS 14.6.1. I'm not sure whether this is a Mac issue or a me issue (deleting all the caches didn't help).

Are any other Mac users able to try it?

[bigol83]

While this did fix the original issue, it's now freezing at "Loading PPU Modules, Progress: module 119 of 120 (1s remaining)" with 0.0.32-16830 on MacOS 14.6.1. I'm not sure whether this is a Mac issue or a me issue (deleting all the caches didn't help).

Are any other Mac users able to try it?

Try deleting PPU cache. It's also weird to me it counts 120, the game PPU modules are 119

[TailsGuy]

Just went through some of the prologue scenes and gameplay, it's all in working order. Even locked 60FPS on a Ryzen 7 5700X.
Though I can't say it's fully playable until Chapter Nine's crash is fixed (or has already been fixed and no one knew since you couldn't boot into the game for a year)

[Darkhost1999]

For Future reference, if this issue gets revived make a new regression report.
The confirmed working build is 0.0.32-16830 give or take a small handful of builds if you'd like to find the exact build that fixed this issue.
A new regression report will not ask what build solved this issue, it'll ask for the last known working build and the build that caused the new regression.

I'm testing on Windows

[Behodar]

Try deleting PPU cache. It's also weird to me it counts 120, the game PPU modules are 119

I moved the rpcs3 folders out of Application Support and Caches, essentially starting from fresh, installed the firmware, booted up... and same issue.

At this point I'm wondering whether it's a Mac issue, or whether it affects Linux too (are the rest of you on Windows?). I'm also not sure whether this is a different bug or still related to this particular one.

[Edit: Your comment came in while I was typing, so I hope you don't feel like I was stepping on your toes]

[Darkhost1999]

I believe Elad was looking into this if there are still residual issues on Mac the executive decision to revive this or request a new ticket is up to Elad.
I would think a new ticket referencing this one as the original post is officially resolved. But that's an unusual one.
And I'm curious too is this only solved on Windows so far? Was anybody testing Linux?

[Behodar]

For what it's worth, I tried this on another Mac - one that's never had RPCS3 on it - and it did exactly the same thing. I'll log a "proper" ticket for it when I have time to collect logs etc.

[mtmrc]

Broken again from build 0.0.32-16848. I'm lazy to open new issue.

[AniLeo]

#15961

[devanteweary]

Try deleting PPU cache. It's also weird to me it counts 120, the game PPU modules are 119

I moved the rpcs3 folders out of Application Support and Caches, essentially starting from fresh, installed the firmware, booted up... and same issue.

At this point I'm wondering whether it's a Mac issue, or whether it affects Linux too (are the rest of you on Windows?). I'm also not sure whether this is a different bug or still related to this particular one.

[Edit: Your comment came in while I was typing, so I hope you don't feel like I was stepping on your toes]

I'm on Windows 10.

Just tried a brand new, fresh portable "install" of RPCS3 using version 0.0.33-16940-02362a48 Alpha | master
Installed PS3 firmware 4.91.

It gets pretty far into the pre-compiling of shaders / PPU (sorry not too familiar with the lingo) and then completely crashes / closes RPCS3.