Closed Chatplosion closed 9 months ago
Let's be honest here: This is such an unusual hardware configuration that it is very unlikely to get help with it.
In 4.9 releases everything was working completely correct and right. This is just a Amlogic S922x SoC. Raspberry is using same type of processors and graphic processor. If you show me how I can check out on specific commits, I can show when the break began. And others can help to fix it. For me it is an honour to help.
I was able to identify the commit 7517b64aeece426cfea50db81749f680ca28b9d3 as the last working one. With trying f41f393e91579b5fe5c312926fcc0a611c8c5a5a it worked no more due to incorrect OpenGL versions. So I tracked down in which commit it happend to go bad. Something got very wrong with the "GLES: Added IQM bone support" improvement.
And one additional error was introduced with commit 37da5268e10804fc766b75c4d09a013f501faae4.
On my linux system it caused errors.
CMake Error at libraries/discordrpc/src/CMakeLists.txt:122 (add_dependencies): The dependency target "clangformat" of target "discord-rpc" does not exist.
And this is continuing since that commit with no fix.
And one additional error was introduced with commit 37da526. On my linux system it caused errors.
CMake Error at libraries/discordrpc/src/CMakeLists.txt:122 (add_dependencies): The dependency target "clangformat" of target "discord-rpc" does not exist.
And this is continuing since that commit with no fix.
Fixed in 561bce40fcc8a196d10b26c860152051fa73eeee.
- Change vid_preferbackend to vid_preferbackend=3
I believe this should be 2
now, since Softpoly has been removed. As 3 is no longer valid, it's likely defaulting to plain OpenGL.
I am using the master branch and did a complete download on commit 561bce4. The CMake Error doesn't occur anymore. Thank you alexey-lysiuk! I tried also the change in gzdoom.ini with no sucess. gzdoom_make.log
But the false api calls go on.Please look up my logs. With commit f41f393 everythin failed. I think that in this commit a flase api call is introduced that does stay until now. gzdoom-crash.log gzdoom_make.log glxinfo.txt
I did look something important up. It is about version strings. OpenGL Shading Language has special version strings. In the file /src/common/rendering/gles/gles_system.cpp is in line 184 a GLES check. This check forces version 100 (GLES 2.0) or version 330 (OpenGL 3.3). In my opinion this is a total error, because the version string for GLES 3.0 should be 300! 330 is OpenGL 3.3 that is completly incompatible to GLES 3! This renders OpenGL ES useless.
This way it actually uses GLES only features. It's a bit of a confusing setup, with USE_GLES2 set to 0 it really for when running on PC, but uses GLES renderer. Set to 1 when actually on GLES hardware.
I tried your suggestion. For me it fails. I can't even compile it. Look into the compiler log. 20221226_compiler_log.txt
Without your suggested change, it compiles perfectly. But it can't run anything.
All variants of ARM64 systems should be tested.
There is something terrible broken. But what exactly is it?
Oh yeah I see that need to be fixed, I'll find some hardware to test this. For now revert that change and try change the version string:
To something supported like "310 es".
I'm feeling very sorry for telling you, that your work arround suggestion also doesn't work. Compiling is still possible, but no more gaming. It is much more complicated than it seemed before.
I'm feeling very sorry for telling you, that your work arround suggestion also doesn't work. Compiling is still possible, but no more gaming. It is much more complicated than it seemed before.
Ah ok, what is the new problem with changing the version string? AFAIK that was the only possible breaking change, but really need to build on Raspberry or something
I am not sure what exactly was the cause. Maybe something is connected to the mesa 3d panfrost driver? You can look up all supported functions here: mesa drivers matrix. The version string of GLES2 is "100" and everything above is "3XX es". This is a big difference in the length and style of the version string.
@Chatplosion If you get a chance please try the branch I have in this PR: https://github.com/ZDoom/gzdoom/pull/1991 Branch to try: https://github.com/emileb/gzdoom/tree/fix_gles_on_embedded_pcs
It works again on my Raspberry. You can also try with CVAR "gl_es" set to 1 and 0 if any issues. Thanks a lot
Good news, your fix is working. I finally was able to compile and play!
Good news, your fix is working. I finally was able to compile and play!
Good news thanks for the update!
Seems to have been resolved.
GZDoom version
GZDoom g4.10pre-85-g6cf3dc045-m - 2022-11-27 10:08:48 -0500 - SDL version
Which game are you running with GZDoom?
Other
What Operating System are you using?
Other
Please describe your specific OS version
Linux odroid-n2plus 5.19.0-odroid-arm64 #1 SMP PREEMPT Ubuntu 5.19.17-202211021456~jammy (2022-11-02) aarch64 aarch64 aarch64 GNU/Linux
Relevant hardware info
Hardkernel N2+
Have you checked that no other similar issue already exists?
A clear and concise description of what the bug is.
The error occurs after the WAD selection. A full screen of an empty window and a changed mouse pointer are shown. It just stops here frozen. It seems to be a GL/GLES related error.
Steps to reproduce the behaviour.
Your configuration
Provide a Log
https://filehorst.de/d/etakleIuDatei gzdoom-crash.log von filehorst.de laden
https://abload.de/thumb/bildschirmfotovom2022gpik3.pngScreenshot