I’m using matrices to calculate the rotation of points in my collision engine. I already have a user-defined matrix data type, but I wanted to know if it would be faster or even advisable to use OGL for matrix calculations. I could then copy the OGL matrix over to my own format for use.

To save some time, I know how to do it (juggling matrices and all), but is it advisable to do so, any adverse effects to rendering?

There really is no fast way of reading back transformd vertices from OpenGL. Stick to you own routines for general use, and use OpenGL when you don’t need the result.

The only “trick” I could think of that is hardware accelerated (feedback mode and gluUnProject generally are not) is filling the modelview matrix with vertices, calling glMultMatrix and reading back the result with glGet. However, that means reading from videomem, which is uncached, which is horribly slow.