Hello,
On AMD cards, there seems to be a continuous stream of OSG OpenGL warnings of the form:
Warning: detected OpenGL error 'invalid operation' after applying GLMode 0xde1 Warning: detected OpenGL error 'invalid operation' after applying GLMode 0x806f
The second warning only appears on windows, while the first one appears on both macOS and windows.
Specs:
OS: macOS Catalina 10.15.5 (next), Windows 10 (2020.1.3)
Graphics Card: AMD Radeon R9 M390 2GB
P.S. Log file run with --log-level=debug and ai/atc messages have been removed to reduce size.
Best,
Fahim
Does it happen on both pipelines (ALS and Default/Low-Spec)?
Atleast on macOS, these warnings do not appear when starting with the following option:
--compositor=Compositor/low-spec
Ofcourse, none of the shader settings make any difference (as expected for the low-spec I think).
That's interesting, although there is nothing inherently different in the ALS pipeline apart from shadows, but you already told me enabling/disabling them doesn't make any difference.
Perhaps it has to do with FBO creation. I'll try to build a custom test pipeline to see if that throws any errors on your machine, will report back.
Yes, just to confirm, this is what I do to disable shadows:
--compositor=Compositor/als --prop:bool:/sim/rendering/als/shadows/enabled=false
I see no shadows, but do see the stream of OSG warnings.
Fahim
I'm getting the same.
See log for GPU specs.
Windows 10
Everything appears to be working as it should, regardless of error. I have interior and exterior shadows.
Commenting to get updates :)
Fahim and I tried to debug this a while ago and it seems like we got some of the errors fixed. I decided not to close this though as there were some errors exclusive to Mac which should be fixed by the new iteration of clustered shading. Once these changes are upstream we should be able to investigate this further.
Closing this as clustered shading is (temporarily) gone from next.
As James reported on the list, the first one still exists unfortunately :(
Fahim, are you getting both error opcodes now, or only 0xde1 ? (Which is the one I see)
Just 0xde1 James, so we are seeing the same behavior.
P.S. You also probably know this, but you can "hide" this endless stream by setting OSG_GL_ERROR_CHECKING=OFF while working on something that requires you to see the console!
@fahimdalvi and @fgarlin, is this still an issue?
I can confirm that on macOS, both the errorcodes still appear (tested on both Intel macOS with the same AMD card as in the report, and on Apple Silicon with M1)
I suspect this might have been fixed by SG commit fd43775b.
@fahimdalvi Does this still happen on the latest next?
Hi Fernando,
My Intel macOS machine crashed and I am in the process of getting it restored, so hopefully I'll have something to share in a few days.
I did try on Apple Silicon Mac and it still exhibits these warnings, FWIW.
Best,
Yeah, if it still shows the warnings on Apple Silicon then it's definitely not fixed. Thanks for checking Fahim.
GLMode 0xde1 = GL_TEXTURE_2D.
So the warning is indicting that you are attempting an operation that is incompatible with a GL_TEXTURE_2D. Unfortunately, the warning does not specify where this is being produced.
Have you been able to at least identify the shader that is causing this?
Shaders cannot give this type of errors. The invalid operation refers to a GPU draw command sent through gl* functions. Basically the place to fix this is the C++ Compositor/FGRenderer code, not FGData.
I suspect it has to do with clustered shading, that's why I was hopeful that they were fixed by the commit I mentioned. The correct way to identify the root cause is to use a debugger like RenderDoc, but this tool is not available in compatibility profile mode. Other tools like apitrace work in the compatibility profile but are not as powerful.
Hi @fgarlin,
I finally got a chance to test on Intel MacOS as well, and as suspected, the warnings remain. One strange thing I noticed was that in my original post, I mentioned only one warning appeared on macOS while both appeared on Windows. Now both appear on macOS as well, atleast we have consistency :D :D
Let me know if you need any testing/info from my side.
Best,
Fahim
Thanks for checking Fahim. I think I can get (temporary) access to an AMD machine to properly debug this. Unfortunately this seems to be "serious" enough to stop 2024.1 from releasing, so I'll make it a priority.