Hi all,

I'm working on a software project that uses OpenGL to create high-resolution results of an image stiching process. For that job, we engage FBOs or pBuffers, depending on the availability.

Recently, we've been told that there are compatibility problems on systems with support for switchable graphics. Apperantly, this technology gains importance, also on Windows platforms. The technology has been supported from Vista onwards. My reference system (Lenovo W500) runs Win7.

Since we are using the Qt framework for software development, I did a few tests with examples coming along with Qt (4.7.4 and 4.8.0). I tested the OpenGL examples "framebufferobject" and "pBuffers" on a system with switchable graphics.

In the Bios, I chose the type of GPU driver. On a system that supports switchable graphics, you have usually three different choices:
1) Discrete Graphics (ATI Mobility FireGL V7500)
2) Integrated Graphics (Mobile Intel 4 Express Chipset Family)
3) Switchable Graphics (switch between ATI and Intel graphics without reboot)

The Qt examples ran smoothly in modes 1) or 2). Case 3) was special: If the "first" adapter (discrete GPU) was attached, the examples ran successfully. If I switched to the "second" adapter however (integrated GPU) the applications quitted with an error message stating that the required extension (fbo, pbuffer respectively) do not exist.

I know, the first thing everyone will think of is "driver problem". Of course, this might be the case also for my system. (Be assured that I checked for the most recent drivers thoroughly enough!) But I would like to learn more about the general state of support for switchable graphics:

• Do the examples work on your machines?
  
• Do you know something one has to take into consideration in order to ensure support for switchable graphics?
  
• It seems that the OpenGL capabilities are not updated correctly for the new device after switching graphics. Some Intel tech-sheet (section 3.9) recommends to handle somehow D3DERR_DEVICELOST or WM_DISPLAYCHANGE messages - but is there some way to re-connect OpenGL to new hardware explicitly?
  
• On Windows, it's possible to query the available adapters with Win32_VideoController class. As with all WMI Classes, one gets a list of objects. For switchable graphics this list is at least of length two. But which device refers to the current adapter?
  

From bug-reports, we get the impression that there are a lot of compatibility problems related to systems with switchable graphics. However, searching the internet about this issue lead me only to support pages of HP or Lenovo offering some specific driver or BIOS updates for specific products. I wonder if there is some general "what-you-need-to-know-about" for switchable graphics and OpenGL...

THANKS to everyone contributing to this topic!

Thank you very much for your response.

I've access to a switchable graphics machine and have tested code using a more-or-less completely-featured GL3.3 program on it, but I did disable switchable graphics in the Bios, and did use an AMD-provided driver (rather than the vendor-provided one) that is not supported for use with switchable graphics but which worked despite that. It's a bit of a hairy thing to expect end-users to have to do, but I considered it fine enough (and accepted the risks involved) for my own personal use.

This is also the best practice we have found until now. But the problem is that for our users (and we have to consider our typical user to be very numb) this is not really an option...

The Intel-recommended D3DERR_DEVICELOST handling would be roughly equivalent - in this case - to destroying and recreating the GL context, and could be handled by your standard display mode changing code. You would have to get a new HDC for your window, create a new pixel format, etc too. This is also something I can test later on.

Ok, since we create an OpenGL context only for the stitching and rendering purpose and close it right away (so the OpenGL context does not have to persist over a long time anyway), recovering from a "device loss" it is not really the biggest concern we have now. For the same reason, the switching from one adapter to another is not an issue now. The problem we currently face at is that our app works only for one of the two display adapters in switchable mode. For the "second" adapter (the integrated one) it looks as if the supported OpenGL version is 1.1, but we require to have at least 1.4. Look at the attachement "SwitchableGraphicsDrivers.png": This is what our software and the Qt examples sees. However, if I check an OpenGL extension viewer such as realtech VR's one can read support for OpenGL 2.1 (see screenshot). Also, if I start the rendering tests, all tests for OpenGL 1.1 up to 2.1 run all successfully. In other words: all the OpenGL functionality is around for 2.1, but Qt cannot access it.

I checked the Qt functions that try to assess the OpenGL capabilities. But apparently it is not a problem of Qt, it simply gets the wrong information when calling to glGetString[color=#000000](GL_EXTENSIONS[color=#000000])...

How is that possible???

Hi mhagain,

I investigated a little further based on your inputs. On the problematic system, Mircrosoft's Generic GDI renderer is in use. I used following code for the context creation (so no Qt code is involved anymore):


void EnableOpenGL(HWND hWnd, HDC * hDC, HGLRC * hRC)
{
PIXELFORMATDESCRIPTOR pfd;
int format;
// get the device context (DC)
*hDC = GetDC( hWnd );
// set the pixel format for the DC
ZeroMemory( &pfd, sizeof( pfd ) );
pfd.nSize = sizeof( pfd );
pfd.nVersion = 1;
pfd.dwFlags = PFD_DRAW_TO_WINDOW | PFD_SUPPORT_OPENGL | PFD_DOUBLEBUFFER;
pfd.iPixelType = PFD_TYPE_RGBA;
pfd.cColorBits = 24;
pfd.cDepthBits = 16;
pfd.iLayerType = PFD_MAIN_PLANE;
format = ChoosePixelFormat( *hDC, &pfd );
SetPixelFormat( *hDC, format, &pfd );
// create and enable the render context (RC)
*hRC = wglCreateContext( *hDC );
wglMakeCurrent( *hDC, *hRC );
}


What might be the reason that OpenGL is connected to the GDI renderer and not to the the vendor's one???
Just to recall: Based on realtech's extension viewer the functionality of OpenGL 2.1 is available on that system..!

As a next step, I'll check the web for an open-source equivalent of this viewer to see how the OpenGL context has to be set up on such a system...

I checked out your hints, but I did not really get a step further.

Querying the available pixel formats with DescribePixelFormat just returns the pixel formats of the GDI renderer. This is how I verified this:

1. Query the pixel formats in a loop (see attached file)
   
2. Compared with pixel formats visible to OpenGL Extension Viewer 4.0
   
3. Compared with pixel formats visible to OpenGL Extension Viewer 2.11 (also developed by realtech VR)
   

On point 3) : I found out that extension viewer 2.11 has the same problem as the vast majority of the tested software so far (Qt examples, some 3D games, our software). If I want to check the extensions with this viewer, it shows only the capabilities of the GDI renderer.
However, if I check with version 4.0 I get the caps of the integrated device (see screenshot)!

So there must be a way to correctly create a render context (unless those guys cheat in some way...)!!!

Let me link my problem with this gamedev topic as it seems to be related... I've added a few comments there.

Hmm. Seems to be a really awkward situation with switchable graphics. Getting slowly frustrated with this topic. This technology is up-and-coming, but yet very immature.

I found more resources here (bug tracker of Second Life) and here (forum frictionalgames).

It turned out that the reported issue indeed WAS a driver problem. The ATI switchable graphics driver runs fine for 64-bit applications, however there seems to be a problem with 32bit OpenGL apps on 64bit systems. I found no 32bit example that would run correctly.

The described utility "OpenGL Extension Viewer" (which I claimed to run smoothly on the problematic system above) was 64bit, this is why it always worked. (Mean about this tool: it installs by default in the x86 program folder - so this is why I always assumed that it is a 32bit app...)

I verified the above statement with a test application that I've built for 32bit and 64bit architectures. For the 64bit version of that app, all hardware supported OpenGL functionality was available on the problematic system (see above). For the 32bit version however, I got the same behavior (OpenGL is connected to the Generic GDI driver, most of the OpenGL functionality is not available).

Since the most recent driver update for the switchable graphics driver (dated on 27th May 2012) did not improve the problem, we decided to render things in software in case of missing OpenGL support.