Resolved-Workaround: [Bug Report] ops2m demo 33 - Dakar 2 [PAL]
#1
# PCSX2 versions used: 
  • pcsx2-v1.3.1-1270-g1c8f8a0-windows-x86 (latest dev build)
# Emulation settings:
Started with "Safest" preset;
Changes to the EE/FPU "Clamping mode" alone have no effect;
Changes to the EE/FPU "Round Mode" alone alter the severity of the bug;
The EE Interpreter is worse than the Recompiler.


# Plugins used: 
GSdx AVX2 using the OpenGL, DX11 and DX9 renderers and software mode;
cvdvdGigaherz.

# Plugin settings: GSdx - CRC Hack Level "None"

# CPU: Intel® Core™ i5-4690K CPU @ 3.50GHz

# GPU: GeForce GTX 970 (Driver: NVIDIA 358.50)

# Description: 
This was found in the Dakar 2 demo of the ops2m demo 33 disc.
With the "Safest" preset, a right trapezoid-shaped section of the race track in front of the player's car goes missing and we only see the skybox through it.
Some camera angles will make this go away.
Changing the EE/FPU rounding modes to something other than the default "Chop/Zero" will increase the severity of the bug - more stuff goes missing and camera angles will no longer make it go away, which leads me to believe this is a bug in pcsx2's EE implementation.



The following is a video of the bug, I also cycle through the different EE/FPU rounding modes:
https://vid.me/01Wc


Because I think this is EE rather than GS related, I didn't make a GS dump. What can I do to help the devs?

Sponsored links

#2
could you also check the emulog for any errors/warnings ?
We're supposed to be working as a team, if we aren't helping and suggesting things to each other, we aren't working as a team.
- Refraction
#3
Try supervu instead of microvu

You selected interpreter for ee and the bug got worse?
#4
(10-17-2015, 11:34 PM)willkuer Wrote: Try supervu instead of microvu

You selected interpreter for ee and the bug got worse?

VU settings don't influence it. Yes, the EE interpreter produces complete garbage, the car isn't even drawn IIRC.

Are warnings/errors in the emulog prefixed as such? Can I do Ctrl + F and search for warning or error?
#5
This could be interesting.

As we have the option: can you enable the full emulation of ee cache?

Can you please append the full emulog using the settings you mentioned in your first post?
#6
yeah, providing the full emulog would be better as willkuer suggested.
We're supposed to be working as a team, if we aren't helping and suggesting things to each other, we aren't working as a team.
- Refraction
#7
(10-18-2015, 12:23 AM)willkuer Wrote: This could be interesting.

As we have the option: can you enable the full emulation of ee cache?

Can you please append the full emulog using the settings you mentioned in your first post?

To recap: emulation settings on "Safest" preset but this time the "Enable EE Cache" option was used.
Here's the emuLog, but I cut out the first 86 lines, just before the first occurrence of the "Hard reset boot"

EDIT: Wait a second, I used the EE Recompiler, the EE Cache is only for the Interpreter. I'll update the post ASAP.

EDIT2: Christ that was painful. Did a full boot with the EE Interpreter and cache enabled, twiddled my fingers while the emulator ran at 5 fps, played the stage for a few seconds. I was wrong, the car is drawn and the screen looked like it was similar to when the other EE/FPU rounding modes (except chop/zero) are used i.e. more of the screen is messed up.
emuLog is attached, again the first 86 lines were snipped.


Attached Files
.txt   emuLogOld2.txt (Size: 40,36 KB / Downloads: 208)
#8
Actually i meant emulog for recompiler but doesnt matter i guess.

This could be an interesting problem for ref or gregory. I dont really think that the issue is in the ee module. Probably bad rounding and imprecision are shadowing a bug in one of the other modules. But that is just guessed here. As you have such a strong correllation between rounding/interpreter settings and gs output it might be debuggable. Unfortunately ee and gs output are seperated by some modules (in different paths) which will make it complicazed to relate a change in the ee to a change in the gs. At least its a issue worth for github. Lets see if one of the devs wants to look into it.

About searching for warnings: the messages can be passed by several functions to the console which can as well mark a message as warning. Unfortunately not all warnings are consequently marked as such. Also warning is probably hard to define from a module point of view. For some games a warning condition in a module might be an expected border case for others its a fatal error.
#9
(10-18-2015, 01:33 AM)willkuer Wrote: Actually i meant emulog for recompiler but doesnt matter i guess.

This could be an interesting problem for ref or gregory. I dont really think that the issue is in the ee module. Probably bad rounding and imprecision are shadowing a bug in one of the other modules. But that is just guessed here. As you have such a strong correllation between rounding/interpreter settings and gs output it might be debuggable. Unfortunately ee and gs output are seperated by some modules (in different paths) which will make it complicazed to relate a change in the ee to a change in the gs. At least its a issue worth for github. Lets see if one of the devs wants to look into it.

About searching for warnings: the messages can be passed by several functions to the console which can as well mark a message as warning. Unfortunately not all warnings are consequently marked as such. Also warning is probably hard to define from a module point of view. For some games a warning condition in a module might be an expected border case for others its a fatal error.

Thought you wanted the log when the interpreter and EE cache were both on. Hope it's still useful to someone and if you feel an emulog with the recompiler is also needed, that can be done relatively quick.
Also, when I said "EE interpreter produces complete garbage, the car isn't even drawn IIRC", I meant the VU1 interpreter, it produces what I call "garbage". Take a look:

This is using the VU1 dynarec.
[Image: MaAsZZB.png?1]

This is using the VU1 interpreter.
[Image: CZIFSLV.png?1]
#10
Is that log from running a developer build or the release build? (git versions are also release builds), i'd be curious to if this game gets a COP2 warning
[Image: ref-sig-anim.gif]





Users browsing this thread: 1 Guest(s)