PowerVR 3 and 3.1 SDK under Windows crash on exit.

I have a curious unhandled exception which occurs only on exit from any application built with PowerVR SDK 3.0 and 3.1 (Demos, new applications), in debug and release builds. However, if the application doesn’t USE any OpenGLES2 functions, the crash doesn’t occur. Even if the application merely calls eglGetDisplay, and NOTHING else (and even if I add eglMakeCurrent with EGL_NO_SURFACE AND EGL_NO_CONTEXT, and an eglTerminate just prior to application exit), the crash occurs.



In Visual Studio 2010, when the exception occurs, the debugger points to __crtExitProcess, but the stack trace suggests (with some noted ambiguity) that execution is inside the gl library.



I have the older 2.10 PowerVR SDK, and when I substitute the include and library directories from which gl2.h, gl2ext.h, egl.h and associated import libraries are drawn, and when I intentionally place the associated 2.10 DLL’s so they’re loaded, the crash does not occur under the same conditions which cause it when using 3.0 and 3.1.



Graphics is ATI 4670, OS is XP x64 (I know, it’s old, but it’s rock stable here and while I won’t upgrade TO Vista, I would move to Win7 if not for all the applications I’d have to re-install).



Is this known to anyone?



Thanks

Hello JVene,



what preprocessor keyword did you use when you did this change ?



regards

david

Hi JVene



This appears to be a recurring issue for anyone running Windows XP and we’re currently looking into the cause of the crash. Thanks for your patience!

Hello JVene,

what preprocessor keyword did you use when you did this change ?



I have to admit I'm not certain what you're asking about precisely.

When I switched to the 2.10 libraries I merely adjusted the project's properties to change the include directories for C++ compilation, and the library path for the linker. Once I did, the project simply linked. To execute I placed the DLL's in the same directory as the executable (I opted not to put them in common location included in my path precisely so I could switch among the various versions).



Hi JVene

This appears to be a recurring issue for anyone running Windows XP and we're currently looking into the cause of the crash. Thanks for your patience!


Thanks for the acknowledgement.

Fortunately it's only at app exit, so most of what I need to debug and evaluate is intact. Unfortunately the only way I can see to perform leak detection is to use the 2.10 DLL's in the meantime, but since I'm focused on GLES 2.0 at the moment, it's a minor issue.