Offscreen Render for all graphics cards
Author
I've been really struggling to get a way to use hardware rendering while rendering off screen and then have it work on any machine, regardless of the graphics card manufacturer and on Vista and XP. I realize that pbuffers are generally the way to go, but this doesn't work on some ATI cards and it doesn't work on older versions of OpenGL (Older than 1.4 I believe).
My Current Method is this:
1. Create a new window class
2. Create a new window with CreateWindowEx and make window invisible
3. wglCreateContext the window
4. Render
5. glReadPixels the pixels back
This works on most implimentations that I have tried, but as always, there are exceptions that I have found.
While debugging on an NVIDIA card installed on a x64 version of XP, for some reason calling the glReadPixels just gets a screenshot of where the window is currently at. Also, if i made the window visible, but it was covered or partially covered by another window, then the glReadPixel will show the other window, so once again, its like a screenshot. Anyone have any solutions or suggestions?
Thanks!
Dave
If your requirement of "hardware rendering" only means you don't want to fall back to the years old default implementation from Microsoft, then I can strongly recommend Mesa3D's off screen renderer. It's almost the ultimate off screen rendering solution and works great.
Quote:Original post by DJHoltkamp
While debugging on an NVIDIA card installed on a x64 version of XP, for some reason calling the glReadPixels just gets a screenshot of where the window is currently at. Also, if i made the window visible, but it was covered or partially covered by another window, then the glReadPixel will show the other window, so once again, its like a screenshot. Anyone have any solutions or suggestions?
Thanks!
Dave
There are two screen buffers (front buffer and back buffer). And, if you are reading front buffer, the content can be depended on the other windows. However, if you are reading back buffer, there is no way other windows get into its way and therefore, no such problem.
For p-buffer (as a offline window), there is no possible for others windows to get into its way and so, no such problem neither.
Quote:Original post by DJHoltkamp
I realize that pbuffers are generally the way to go, but this doesn't work on some ATI cards and it doesn't work on older versions of OpenGL (Older than 1.4 I believe).
My Current Method is this:
1. Create a new window class
2. Create a new window with CreateWindowEx and make window invisible
3. wglCreateContext the window
4. Render
5. glReadPixels the pixels back
This has nothing to do with pbuffers. You just create a second rendering window, and read back its framebuffer. Doing this is undefined, as you have noticed, due to the pixel ownership test. It's pretty much guaranteed to fail if the window is hidden. This has nothing to do with back versus front buffer, BTW. Even reading back the backbuffer can fail if the window is partially invisible.
So you can in fact use real pbuffers, and you won't get this problem. However, pbuffers are really, really obsolete. Just use FBOs, like everybody else. They work on pretty much every card out there, except for maybe some Intel stuff with crappy drivers. But there's not much you can do in the latter case anyway.
Author
Thanks for the replies!
Brother Bob:
I may look into the Mesa3D thing but I doubt this is the way I'm going to go due to the fact that I don't want to include other libraries and need it to work on most all windows/mac machines.
ma_hty:
I may look into the front/back buffer deal and see if there is a way to swap and then read the back... I had never thought of that but it is an intersting idea. GlReadPixels does not seem to have a way to select the back buffer. How would you go about doing this?
Yann L:
Unfortunately, I do need this to work on the all crappy intel integrated cards. Surprisingly, these cards seem to be working better than the NVIDIA and ATI cards for this method. It works flawlessly with the window hidden. In fact, this method of window creation off screen is how one of the engineers at adobe said that some of their OpenGL offscreen rendering was accomplished, but perhaps they did not know what they were talking about...
Any other ideas or suggestions?
Brother Bob:
I may look into the Mesa3D thing but I doubt this is the way I'm going to go due to the fact that I don't want to include other libraries and need it to work on most all windows/mac machines.
ma_hty:
I may look into the front/back buffer deal and see if there is a way to swap and then read the back... I had never thought of that but it is an intersting idea. GlReadPixels does not seem to have a way to select the back buffer. How would you go about doing this?
Yann L:
Unfortunately, I do need this to work on the all crappy intel integrated cards. Surprisingly, these cards seem to be working better than the NVIDIA and ATI cards for this method. It works flawlessly with the window hidden. In fact, this method of window creation off screen is how one of the engineers at adobe said that some of their OpenGL offscreen rendering was accomplished, but perhaps they did not know what they were talking about...
Any other ideas or suggestions?
Quote:Original post by DJHoltkamp
Brother Bob:
I may look into the Mesa3D thing but I doubt this is the way I'm going to go due to the fact that I don't want to include other libraries and need it to work on most all windows/mac machines.
If you don't want to use Mesa, you have to use some other OpenGL interface instead. WGL on Windows for example. Mesa doesn't add anything, it replaces, so if you go for Mesa, you don't need WGL. And platform support is not an issue for Mesa. So these can't be the issues you don't want to use Mesa.
Quote:Original post by DJHoltkamp
I may look into the front/back buffer deal and see if there is a way to swap and then read the back... I had never thought of that but it is an intersting idea. GlReadPixels does not seem to have a way to select the back buffer. How would you go about doing this?
glReadBuffer. But the operation is undefined on overlapped or hidden windows.
Quote:
Unfortunately, I do need this to work on the all crappy intel integrated cards. Surprisingly, these cards seem to be working better than the NVIDIA and ATI cards for this method. It works flawlessly with the window hidden. In fact, this method of window creation off screen is how one of the engineers at adobe said that some of their OpenGL offscreen rendering was accomplished, but perhaps they did not know what they were talking about...
Either they meant something different (like a real pbuffer), or they don't know what they're talking about. The method you are using is undefined. This means that it might work - or it might not, depending on manufacturer, driver revision, or even from one PC to another. This is almost random, and completely out of your control. The fact that it works on your Intel chip doesn't mean it will work on another. Look here. Note that with 'nonvisible window' in the last paragraph, they don't mean a hidden window, but an abstract surface. This is not accelerated under Windows.
Quote:
Any other ideas or suggestions?
Have you tried FBOs before dismissing them ? And if they don't work, why don't you use actual PBuffers ? They were meant for cases such as this one...
Author
Brother Bob:
You cna see from my first post that I am using wgl because of wglCreateContext.
Yann L:
One of my contacts from adobe said the following:
"However, your approach does match some internal code we have to bind an offscreen
window to OpenGL, with the window being a member of a static class that
is initialized on PF_Cmd_GLOBAL_SETUP, and kept in memory"
I guess this offscreen window could be a FBO?
One reason I was staying away from PBuffers is because I know that some crappy OpenGL implimentations on ATI cards do not have this extention included. Do you know what version of OpenGL that FBO's came into existence? Why do you say that it won't work on intels?
You cna see from my first post that I am using wgl because of wglCreateContext.
Yann L:
One of my contacts from adobe said the following:
"However, your approach does match some internal code we have to bind an offscreen
window to OpenGL, with the window being a member of a static class that
is initialized on PF_Cmd_GLOBAL_SETUP, and kept in memory"
I guess this offscreen window could be a FBO?
One reason I was staying away from PBuffers is because I know that some crappy OpenGL implimentations on ATI cards do not have this extention included. Do you know what version of OpenGL that FBO's came into existence? Why do you say that it won't work on intels?
Quote:Original post by DJHoltkamp
...
One reason I was staying away from PBuffers is because I know that some crappy OpenGL implimentations on ATI cards do not have this extention included. Do you know what version of OpenGL that FBO's came into existence? Why do you say that it won't work on intels?
PBuffer is something from your GPU not from OpenGL. Even the entry point functions are not declared in your opengl header file, they are there if your GPU support it. And, you can use glew library to test whether your GPU support it or not. If so, you access them using the entry point functions from glew library. ( You can download glew library from http://glew.sourceforge.net )
Therefore, you should not worry about which verion of OpenGL, instead, the model of your GPU. And, a well-written GPU driver would provide a software emulation in case a hardware support is not present (hopefully...). Anyway, the thing you should worry about should be the model of GPU.
By the way, PBuffer is difficult to be used as its initialization requires platform dependent API calls. Please use FBO instead.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
