JMonkeyEngine crash on Intel video adapter

Question

I am using JME in my application and sometimes it crashes with the following message:

#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x3d601ad7, pid=168, tid=4012
#
# JRE version: 6.0_29-b11
# Java VM: Java HotSpot(TM) Client VM (20.4-b02 mixed mode, sharing windows-x86)
# Problematic frame:
# C  [ig4dev32.dll+0x21ad7]
#
# An error report file with more information is saved as:
# C:\...\hs_err_pid168.log
#
# If you would like to submit a bug report, please visit:
#   http://java.sun.com/webapps/bugreport/crash.jsp
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#

The log file can be found by this link: http://sergpank.heliohost.org/log.html

The strangest thing is that in my case I get crashes only ob builded code, but when I am launcing it from the Eclipse, everything works fine on my machine. On machines with AMD video adapters nothing crashes. On other machines with Intel videocard sometimes crashes appear and on the debug stage.

I am starting to suppose that this happens because of incorrect ant setup (in startup.ini the following path is set up: -Djava.library.path=lib/dlls so dlls is seen for the project). But still can't get why it works almost perfectly on AMD and crashes on Intel.

Maybe it is something related to the ant, and I have to add dlls to the manfest... looking through the documentation and can't find the way how it can be done.

Solution:

On 64 bit system is necessary to use the corresponding JVM(64-bit) and then nothing crashes =))

Answer 1

Crash happens because 32-bit JVM was used on 64-bit OS. It seems that in this case 32-bit dlls was loaded and that's why crash happened.

Issue is reproducible only on Intel video cards, I think it can be considered as a serious bug. If Intel would like to fix it or propose a working solution/workaround this could be great! =)

Answer 2

Avoid doing the heavy work of the OpenGL stuff in the Swing Event Dispatch Thread (note the thread which crashes the JVM: =>0x3a88e000 JavaThread "AWT-EventQueue-0" [_thread_in_native, id=5228, stack(0x3b170000,0x3b1c0000)). I believe OpenGL stuff should be done in the thread offered by JMonkeyEngine using the event dispatch mechanism it has. If you're using somebody else's API for Swing rendering, you might need to change it or do it a different way.

Edit: it looks like AWTGLCanvas does this, changing the context to the current thread. It looks like the Intel drivers may have trouble with context switches if normal fullscreen 3D stuff works. Strictly speaking, the GL thread context stuff shouldn't be necessary as you can always dispatch work to be executed to a singular OpenGL thread which should be fine as long as you only have a single OpenGL rendering viewport. The LWJGL Canvas implementation assumes you will want multiple viewports when this isn't necessarily the case. You can recode this to only support one viewport if that is fine, and you shouldn't get crashes as the code is simpler.