Jump to content

  • Log In with Google      Sign In   
  • Create Account

Is it possible to step-debug D3D11 functions?


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
3 replies to this topic

#1 Husbjörn   Members   -  Reputation: 256

Like
0Likes
Like

Posted 25 April 2014 - 06:10 PM

This feels extremely off, but I just can't seem to track the issue down so I thought I'd ask in case someone else has experienced something similar.

 

So I'm writing a game engine in C++ using DX11 for rendering. Its going rather well for the moment and everything works just as intended when I build it as an executable. However, as a side project I'm also wrapping the engine functionality into a dynamic link library so that it may be used from other languages as well. Now this is also working... for the most part. For some reason that has me dumbfounded the host application using the dll ends up in a deadlock upon calling DLLMain with the DLL_PROCESS_DETACH reason. I've tracked the issue down to happen inside my final release call that results in my ID3D11Device getting a reference count of 0, so it most likely happens as the device is destroyed. Unfortunately there is no debugging information available that lets me see exactly where it all goes sour in there, except for that control goes into the final Release() call and never comes back out.

I have logged the refcounts for the device, device context, swapchain, etc. and they are released in exactly the same manner in DLL and executable build alike, yet the DLL version gets stuck while the executable happily terminates without as much as a sniffle from the DX debug layer or anything that would hint at hidden errors. I've gone through my code two times searching for lingering buffers that may be dependent on the device (however were that really the case it should maintain its own reference and thus prevent it from being destroyed before the buffer / whatever resource is).

 

Code and reference counts:

// pSwapChain has 1 reference, pDevCon 1 and pDev 3 here
if(pSwapChain)
    pSwapChain->Release(); // pSwapChain has 0 references, pDevCon 1 and pDev 2 after this line
if(pDevCon)
    pDevCon->Release();    // pSwapChain is unreferenced, pDevice context hits 0 references and pDev has 1 remaining reference
if(pDev)
    pDev->Release();       // 0 references for all three resources in executable build; control never returns from this call in dll build

So for my questions... is there any way to debug inside the ID3DDevice::Release() / ID3DDevice::dtor() bodies to see what's really happening? Is this code even open source / otherwise available for reading? I'm guessing not. Secondly, is this a known issue that can arise and, if so, what causes it? I've tried searching a good deal for it but have so far been unable to find anything relevant.

 

 

Thanks for any ideas, 

Husbjörn


Edited by Husbjörn, 25 April 2014 - 06:14 PM.


Sponsor:

#2 MJP   Moderators   -  Reputation: 11376

Like
0Likes
Like

Posted 25 April 2014 - 06:33 PM

Is your host process exiting, or is it just unloading the DLL? Because if it's exiting, then you may want to follow Raymond Chen's advice and do nothing.



#3 mhagain   Crossbones+   -  Reputation: 7979

Like
1Likes
Like

Posted 25 April 2014 - 07:00 PM

Release does 1 or both of 2 things.  First of all it decrements the reference count; if the reference count hasn't yet hit zero, it should return immediately.  If it has, the object is destroyed.

 

A possible debugging step you could take is to add an AddRef call before the Release.  See if it returns immediately with that in place (accepting that you're now going to leak the object, but that's OK for debugging/testing purposes).

 

While there is no D3D source code available (and in this case - assuming that the problem is inside the Release call (it may not be) - it may be in your GPU vendor's driver, or even the hardware, that you've got the problem) the best you can do is grab some debug symbols and test with those.  See http://msdn.microsoft.com/en-us/library/windows/desktop/ee416588%28v=vs.85%29.aspx for instructions.  Break execution when the program hangs and at the very least you'll get a nice call stack that you can work through and find out where the hang is.

 

Or you may even have a bad pointer that's throwing you to that part of the code, or it may be something else like memory allocated in the hosting exe that you're trying to free in the DLL, or vice-versa, causing your program to go nuts.

 

A useful trick is to add a "do nothng" line of code after the Release call, something like "pDev = NULL;" will do, put a breakpoint on that, and see if execution gets there.  If it does, your problem isn't in the Release, it's elsewhere.


It appears that the gentleman thought C++ was extremely difficult and he was overjoyed that the machine was absorbing it; he understood that good C++ is difficult but the best C++ is well-nigh unintelligible.


#4 Husbjörn   Members   -  Reputation: 256

Like
0Likes
Like

Posted 26 April 2014 - 02:48 PM

Is your host process exiting, or is it just unloading the DLL? Because if it's exiting, then you may want to follow Raymond Chen's advice and do nothing.

It is exiting and I did think about that, but it doesn't quite feel right. Also I want to be ready for the possibility that the dll is unloaded without the main app terminating so I'd rather keep it leak free. I suppose a separate function that has to be explicitly called before unloading the lib would work; in case it then terminates in an unexpected manner the DLLMain would just leave everything as is.

 

 

 

A possible debugging step you could take is to add an AddRef call before the Release.  See if it returns immediately with that in place (accepting that you're now going to leak the object, but that's OK for debugging/testing purposes).

Yes, I did something akin to that for finding out the reference counts I provided above.

 

 

It turned out that the problem goes away if I export my DLLMain function (I have no idea why; as far as I know you're not even supposed to export it). I suppose that will do.

 

 

Still so, thanks for the hints and especially that link mhagain :)

They will most certainly be of great use in the future.






Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS