GAMA crashes when displaying a chart with opengl on windows with a radeon graphic card setup

[LBIRE]

Describe the bug When running an experiment which displays a chart using opengl, GAMA crashes.

To Reproduce Run the following code, which is incremental model 2 from Library models with the display changed to opengl:

/**
* Name: Charts
* Author: GAMA team
* Description: Second part of the tutorial : Incremental Model
* Tags: chart, tutorial
*/
model SI_city

global {
    int nb_people <- 500;
    float agent_speed <- 5.0 #km/#h;        
    float step <- 1 #minutes;
    geometry shape <- envelope(square(500 #m));
    float infection_distance <- 2.0 #m;
    float proba_infection <- 0.05;
    int nb_infected_init <- 5;
    int nb_people_infected <- nb_infected_init update: people count (each.is_infected);
    int nb_people_not_infected <- nb_people - nb_infected_init update: nb_people - nb_people_infected;
    float infected_rate update: nb_people_infected / nb_people;

    init {
        create people number: nb_people {
            speed <- agent_speed;
        }

        ask nb_infected_init among people {
            is_infected <- true;
        }
    }

    reflex end_simulation when: infected_rate = 1.0 {
        do pause;
    }
}

species people skills: [moving] {
    bool is_infected <- false;

    reflex move {
        do wander;
    }

    reflex infect when: is_infected {
        ask people at_distance infection_distance {
            if (flip(proba_infection)) {
                is_infected <- true;
            }
        }
    }

    aspect default {
        draw circle(5) color: is_infected ? #red : #green;
    }
}

experiment main_experiment type: gui {
    parameter "Infection distance" var: infection_distance;
    parameter "Proba infection" var: proba_infection min: 0.0 max: 1.0;
    parameter "Nb people infected at init" var: nb_infected_init;
    output {
        monitor "Current hour" value: current_date.hour;
        monitor "Infected people rate" value: infected_rate;
        display map {
            species people;
        }

        display chart refresh: every(10 #cycles) type:opengl {
            chart "Disease spreading" type: series style: spline {
                data "susceptible" value: nb_people_not_infected color: #green marker: false;
                data "infected" value: nb_people_infected color: #red marker: false;
            }
        }
    }
}

Desktop (please complete the following information):

• OS: Windows 10
• PC Model: Lenovo YOGA
• GAMA version: 1.8.2
• Graphics cards / Display system: AMD RADEON

Additional context We tried the 4 values for the flags : use_old_animator use_native_opengl_window with no difference

[lesquoyb]

To add more information: I tried the example on Ubuntu with an AMD RADEON and everything worked well

[hqnghi88]

Hi, I explain something in https://github.com/gama-platform/gama/issues/3398#issuecomment-1120661720 regard my experience on windows, some case of using GPU can cause crash. Hope it help.

[AlexisDrogoul]

Can you try testing again with the new changes made to the display infrastructure in the 1.8.2 branch ? Thanks

[lesquoyb]

I tried on Windows+amd radeon and it's still crashing

[AlexisDrogoul]

I did some changes in #3442 to the way charts are displayed and compute their images. Can you try if it changes anything ?

[lesquoyb]

Tested again on Windows, it's still crashing. I tried with and without AMD software: adrenalin edition as it solved some problems related to my radeon before, but no difference. In eclipse I get this error message:

#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x00007ffce03a4486, pid=1636, tid=12044
#
# JRE version: OpenJDK Runtime Environment Temurin-17.0.2+8 (17.0.2+8) (build 17.0.2+8)
# Java VM: OpenJDK 64-Bit Server VM Temurin-17.0.2+8 (17.0.2+8, mixed mode, tiered, compressed oops, compressed class ptrs, g1 gc, windows-amd64)
# Problematic frame:
# C  0x00007ffce03a4486
#
# No core dump will be written. Minidumps are not enabled by default on client versions of Windows
#
# An error report file with more information is saved as:
# C:\Program Files\eclipse\hs_err_pid1636.log
#
# If you would like to submit a bug report, please visit:
#   https://github.com/adoptium/adoptium-support/issues
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#

And here is the log: hs_err_pid1636.log

[AlexisDrogoul]

Things to test :

• Does it change the behaviour to be synchronised with the simulation ?
• Does it change the behaviour to refresh the display every X cycles ?
• If the first seconds correctly appear before crashing, is there an increase of memory used by GAMA ?

[lesquoyb]

Here is a minimal (non-)working experiment:

model crash_radeon

experiment main_experiment {
    output {
        display chart type:java2D {
            chart "hop"  {}
        }
    }
}

The problem really seem to come from the chart + opengl as commenting the chart inside the display or switching to java2D solves the problem.

to sum up what was already tried:

• synchronize the output: no difference
• type of chart: no difference
• playing with the flags use_old_animator and use_native_opengl_window: no difference
• Installing AMD software: adrenalin edition that solves other radeon related crashes: no difference
• refreshing every X cycle: no difference
• forcing windows to use the radeon gpu for in the os preferences (I thought that maybe leaving the choice to windows could lead to a kind of error in certain case): no difference
• switching my laptop into "better use of resources" (low consumption/performance): the crash occurs later: the simulation has the time to initialize and stay stable instead of crashing directly, but it crashes as soon as the simulation starts

For the question about memory used by gama : no more memory usage than usual

[lesquoyb]

A few updates on that issue: there seem to have been a windows update that makes things a bit better (maybe also helped by the changes in the code), in my particular case most models work now at the exception of displaying an empty chart (adding data fixes it) and zooming. I conducted a few different test on my computer and @LBIRE's and the issue seems to be related to performances, as it's still not working on leo's computer (which is less performant than mine) though it can now work a few seconds in some cases. Another thing that makes me think it's performance related is that it crashes way less on the release than on the git version run through eclipse

[lesquoyb]

After further investigations the bug really seems to be rooted in the code of the gpu driver, so not much we can do on our side. And other softwares having the same issue all just recommend the user to play with the computer settings (energy settings, switch to an other gpu etc.) to try and make the software more stable, but none seem to have solved it. The good news is that last week AMD released a new update of AMD software: adrenaline edition and it solved the bug completely on my computer. So I'll wait a bit to try more complex models and scenarios, but it's possible that the solution to it could simply be "download the latest AMD software version".

[AlexisDrogoul]

@lesquoyb that would be a fantastic good news !

[lesquoyb]

Tried on another computer with the same issue, installing AMD Software: Adrenalin edition version 22.10.1 solved it. I thus consider the problem solved and added a few lines about it in the Troubleshooting page of the wiki.

[hqnghi88]

In https://github.com/gama-platform/gama/wiki/Troubleshooting/b4b827633ef4aa7567d94442fbd0b3e5dc24a1ca "solved the problem but at the cost of very slower rendering". If it is slow rendering, it is not using 3D card or even the installation is not right.

[lesquoyb]

Yes, it is slow precisely because it is not using a proper gpu but an integrated chipset. Basically that's a last resort solution

[hqnghi88]

Yes, that's what i want to say: installing the Adrenalin supposed to fix the GPU driver, not to prevent OS to use properly the GPU.