Jump to content
  • Advertisement
Sign in to follow this  
staticVoid2

OpenGL OpenGL, RAII and Multi-threading

This topic is 1424 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

I'm currently porting some OpenGL code from an SFML framework to Qt. I've noticed that with Qt you have to initialize OpenGL in a callback method (extending a QGLWidget class) whereas previously I would have initialized OpenGL at the very beginning of the application. This has caused a few problems for me:

The main problem is that the code uses a lot of RAII and OpenGL resources are allocated in object constructors throughout a large hierarchy of classes. The root of this hierarchy is of-course the Application class (which contains the QGLWidget).

I now find myself having to go through all classes which use OpenGL resources and adding an 'InitializeGL' method. This ensures that the OpenGL context has been initialized and that the resources are being allocated on the correct thread but it makes the code look and feel very wrong, especially since there are classes which have nothing but OpenGL resources and could even be considered as just extending the existing API.

One solution I considered would be to Move the application initialization stage into the initializeGL method but this seems like a terrible solution.

Another would be to abstract the rendering interface to allow for OpenGL resources to delay allocation until the OpenGL context has been
established and/or the resources are actually required. For example: load a texture image onto the CPU and in the Application's render/draw method check if the texture has been uploaded to the GPU yet, if not then upload it. This ensures the resources are allocated in the correct thread and allows the resources to be aquired in the object constructors without consequence.

I would usually just go with the first option here (and have done in the past) but because this code will most likely be the foundation of a large code base I was wondering what the best solution would be.

Thanks.

Share this post


Link to post
Share on other sites
Advertisement


The root of this hierarchy is of-course the Application class (which contains the QGLWidget).

 

I don't see why you would have to use Application as the owner of the GL-dependant classes.

Wouldn't everything be solved if you didn't use the Application as root for all gl-dependant classes, but instead used your QGLWidget class as root?

 

Then you can just start creating all those classes on the callback that tells you gl has been initalized.

Share this post


Link to post
Share on other sites

I could do that but it means breaking RAII and moving the initialization of those objects into the initializeGL method, the OpenGL is also mixed in with normal code so for example:

class TerrainManager
{
	float* mHeightData[128][128];
	
	GLuint mIndexBuffer;
	
	TerrainManager()
	{
		for(int x = 0; x < 128; ++x)
		{
			for(int y = 0; y < 128; ++y)
			{
				mHeightData[x][y] = /* some value */;
			}
		}
	}
	
	
	void InitGL()
	{
		glGenBuffers(1, &mIndexBuffer);
		...
	}
	
	void DestroyGL()
	{
		glDeleteBuffers(1, &mIndexBuffer);
		...
	}
}

class Application
{
	QGLWidget* mGLWidget;
	
	TerrainManager mTerrainManager;
}

The mHeightData, even though initialized in the constructor, would not be accessible until the initializeGL callback method was called. Adding the InitGL and DestroyGL methods is a suitable solution but I feel it just adds a lot of code. I was wondering if abstracting the rendering interface to be something like:

class TerrainManager
{
	float* mHeightData[128][128];
	
	IndexBuffer* mIndexBuffer;
	
	TerrainManager()
	{
		for(int x = 0; x < 128; ++x)
		{
			for(int y = 0; y < 128; ++y)
			{
				mHeightData[x][y] = /* some value */;
			}
		}
		
		mIndexBuffer = new IndexBuffer(...);
	}
}

class IndexBuffer
{
	vector<int> mIndices;
	GLuint mIndexBuffer;

	void InitGL()
	{
		glGenBuffers(1, &mIndexBuffer);
		...
	}
	
	void DestroyGL()
	{
		glDeleteBuffers(1, &mIndexBuffer);
		...
	}
}

might be a better solution.

Share this post


Link to post
Share on other sites

Maybe I'm not understanding your problem, but can't you simply wait with creating your terrain manager or index buffer until the initialization callback? Just taking a quick glance at the Qt documentation suggests something like this (take the actual code with a grain of salt, just conveying a general idea):

class MyWidget : QGLWidget {
    ...
    void InitializeGL();
private:
    QScopedPointer<TerrainManager> terrainmanager;
    QScopedPointer<IndexBuffer> indexbuffer;
}
 
void MyWidget::InitializeGL()
{
    terrainmanager.reset(new TerrainManager);
    indexbuffer.reset(new IndexBuffer);
}

Share this post


Link to post
Share on other sites


I've noticed that with Qt you have to initialize OpenGL in a callback method (extending a QGLWidget class) ...

I'm not familiar with Qt in particular, so I may be wrong. Usually you can allocate OpenGL resources as soon as an OpenGL context exists. So, if you create a QGLContext instance on your own and later pass it into QGLWidget::QGLWidget ( QGLContext *, QWidget *, const QGLWidget *, Qt::WindowFlags ) ... would it then not be possible to allocate your OpenGL resources similarly as done with SFML?

Share this post


Link to post
Share on other sites

Another would be to abstract the rendering interface to allow for OpenGL resources to delay allocation until the OpenGL context has been
established and/or the resources are actually required. For example: load a texture image onto the CPU and in the Application's render/draw method check if the texture has been uploaded to the GPU yet, if not then upload it. This ensures the resources are allocated in the correct thread and allows the resources to be aquired in the object constructors without consequence.

This is what we do in our very huge in-house game engine.
It saves a lot of trouble, prevents retarded bugs, etc.

 

Besides, as you work with Qt you will release that each widget has its own context and you can’t share a texture (or any resource) between widgets, so it will likely be very handy to be able to create resources just-in-time across multiple widgets.

 

 

L. Spiro

Share this post


Link to post
Share on other sites

Thanks for the replies. All of the sugested methods work, I was just looking for the best way to structure this for scalability.

 

 

This is what we do in our

" title="External link">very huge in-house game engine.
It saves a lot of trouble, prevents retarded bugs, etc.

 

Can I ask, do you abstract the entire rendering interface? Or is this just for resource allocation? I have no intention of creating a layer between the application code and OpenGL but for the case I have stated it poses a lot of benefits.

 

I'll try some of the suggests methods and see which one fits best.

 

Thanks.

Share this post


Link to post
Share on other sites

do you abstract the entire rendering interface? Or is this just for resource allocation?

We support many platforms and full abstraction is necessary.
Delayed resource creation is a policy on all platforms because it avoids multi-threading issues (our renderer is multi-threaded), avoids unused resource creation (not that we have that happening a lot, but some LUT tables are generated at start-up and then only used if certain features are later enabled), and is basically the safest policy.


L. Spiro

Share this post


Link to post
Share on other sites

I could do that but it means breaking RAII and moving the initialization of those objects into the initializeGL method, the OpenGL is also mixed in with normal code so for example:


I don't see how that's breaking RAII. Where you initialize the resource is irrelevant, so long as the initialization also acquires ownership and that ownership is handled properly for cleanup.

Also keep in mind you want to be able to reload your resources on the fly either by user request or in response to a GPU reset or the like so you really want to abstract your resource handles from the GPU resources themselves. Even if you don't keep the images and whatnot in CPU memory, at least have enough information to be able to pull them back off the disk on demand.

A naive but quite usable approach would be to use a proxy resource object, something like:

class TextureResource final : public IResource {
  GpuTextureHandle m_gpuTexture;
  string m_fileName;

public:
  TextureResource(string fileName) : m_fileName(move(fileName)) {}

  bool LoadResource(GraphicsSystem& graphics, ResourceSystem& resources) override { m_gpuTexture = resources.LoadTexture(graphics, m_fileName); }
  bool UnloadResource(GraphicsSystem& graphis) override { graphics.UnloadTexture(m_gpuTexture); m_gpuTexture = {}; }
  bool IsLoaded() override const { return m_gpuTexture != {}; }

  GpuTextureHandle GetGpuTextureHandle() const { return m_gpuTexture; }
};
With that, you can do a cycle of Unload/Load for any resource you want to reload, or even for all resources as a group when the GPU resets or if you're toggling GL contexts for any reason.

More efficient resource structures are quite possible and I personally prefer the API a good data-oriented interface necessitates, but the above gets the job done.

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

We are the game development community.

Whether you are an indie, hobbyist, AAA developer, or just trying to learn, GameDev.net is the place for you to learn, share, and connect with the games industry. Learn more About Us or sign up!

Sign me up!