Multiple projection matrices

Graphics and GPU Programming Programming

Started by gchris6810 June 22, 2014 05:49 PM

3 comments, last by gchris6810 9 years, 10 months ago

gchris6810

207

Author

June 22, 2014 05:49 PM

Hi,

I am creating an application with multiple viewports. I was researching methods as to how I could store multiple projection matrices and all I found was the glPushMatrix and glPopMatrix functions which don't seem to be adequate for what I need (although I may be wrong). Would computing the individual projection matrices each frame give a large performance loss? If so then what method can I use to store and recall them? If I have to use the aforementioned functions then could I please have an example of the implementation. I understand that the projection matrix stack can store a maximum of two matrices on some hardware. How would I circumvent this? Thanks.

Samith

2,465

June 22, 2014 08:37 PM

What version of OpenGL are you using? glPushMatrix and glPopMatrix functions have long since been deprecated and full removed in OpenGL 4.0 and higher.

And no, computing a projection matrix for N viewports is not an expensive operation. Computing a projection matrix is essentially one cotangent call and a few multiplies and divides. And you should only need to do this once per viewport per frame. And if you truly suspect you're wasting a lot of time computing projection matrices, then you can go ahead and profile your application and see if that's indeed what is happening.

swiftcoder

18,997

June 22, 2014 10:17 PM

and full removed in OpenGL 4.0 and higher.

Only in Core contexts.

Regardless, pretty much all OpenGL programs these days manage their own matrices. You generally provide them as uniforms to a shader, which gives you the freedom (and responsibility) to define them any way you like.

Tristam MacDonald. Ex-BigTech Software Engineer. Future farmer. [https://trist.am]

21st Century Moose

13,459

June 23, 2014 08:38 AM

Even in old-school GL you could store multiple matrices as arrays of 16 floats (each), then glLoadMatrixf them. Example:

float pmatrix0[] = {...};

float pmatrix1[] = {...};

float pmatrix2[] = {...};

// some time later

glMatrixMode (GL_PROJECTION);

glLoadMatrixf (pmatrix0);

glMatrixMode (GL_MODELVIEW);

// draw some stuff and then:

glMatrixMode (GL_PROJECTION);

glLoadMatrixf (pmatrix1);

glMatrixMode (GL_MODELVIEW);

// draw some more stuff, then:

glMatrixMode (GL_PROJECTION);

glLoadMatrixf (pmatrix2);

glMatrixMode (GL_MODELVIEW);

Regarding performance, yes, there is going to be some (tiny) overhead from switching any kind of matrix in the scene, but also applies to modelview, texture matrices, etc. The overhead is honestly going to be so small that you won't even measure it on any performance graph, unless you're doing something flat-out crazy (like switching the matrix per-particle in a particle system).

For a practical real-world example, GLQuake switched it's projection matrix twice per-frame and didn't suffer from it.

Direct3D has need of instancing, but we do not. We have plenty of glVertexAttrib calls.

gchris6810

207

Author

June 23, 2014 09:28 AM

Thanks for the replies. Seeing as the performance loss is negligible I think I will stick with what I have.

Multiple projection matrices

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Multiple projection matrices

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Reticulating splines