Excessive memory usage for CORE profile on integrated GPU (Windows OS)

[DrunkyBard]

Hi,

I tried to run Core32Test on Windows OS on integrated GPU and noticed excessive memory usage on window resizing. Eventually application stops rendering process and crashes with MEMORY_ACCESS_VIOLATION exception when user continues resizing target window.
Everything is fine on Compatibility profile for integrated GPU. Also Core profile test app works correctly on dedicated GPU.

I managed to reproduce such behaviour on 2 machines with integrated GPU, they has Intel UHD graphics on board.
Idk if it is useful info but first machine had 27.20.100.8729 9/11/2020 driver, second one — 27.20.100.9168 1/8/2021.

Core profile works correctly on the third machine with Linux OS on Intel UHD graphics 620 integrated GPU.

Is it possible that there is a bug in Windows GPU driver for CORE profile or there might be some workaround to avoid enormous memory consumption on Windows for integrated GPU?

[SWinxy]

The OpenGL tests have a lot of memory leaks, but I don't think that's the issue here, as the tests don't continually (to my knowledge) leak memory.

@DrunkyBard do all the OpenGL tests cause this issue? Have you tried the Vulkan one (SimpleDemo)? How many times do you have to resize the window does it take to crash? My CPU doesn't have integrated graphics so I can't test it.

[DrunkyBard]

Other tests are ok since they use compatibility mode. I didn't try the Vulkan example.
I resize the window constantly as soon as application started. Application crashes within a minute in that case.

[SWinxy]

Do you know what is causing the crash? The logs have a stack trace which is sometimes hard to see. What's the problematic frame, and what's the stacktrace?

[DrunkyBard]

Here is the dump file generated after the crash:
CORE_PROFILE_CRASH_DUMP.txt

[SWinxy]

Done some digging and it appears that ReleaseDC() in User32 or wglMakeCurrent() in WGL may be the cause. I can't know what the exact stack trace is, as it'd been inlined by the JIT. I'm not at a Windows computer for the next few days, so could you try to run it w/o the JIT as a VM option (-Djava.compiler=NONE)?

[DrunkyBard]

I've tested with given VM options - still the same result and same stacktrace

[SWinxy]

Hm... Odd. I can't seem to reproduce the issue. @httpdigest you know about OpenGL, any thoughts?