I have to use mipmapped textures in the driver I develop.
The creation of mipmapped texture takes very long time (about 300ms).
It is unacceptable.
To generate mipmaps I am setting GL_GENERATE_MIPMAP to GL_TRUE.
I am using Quadro FX4500. The textures are NPOT RGBA.
What I can do to improve the situation?
Is the mipmaps generation done in software?
Jan
Try to use power-of-two textures instead.
AFAIK, hardware NP2-mipmap creation is supported only on GeForce6 and above.
Read this:
http://download.nvidia.com/developer/Papers/2005/NP2_Mipmapping/NP2_Mipmap_Creation.pdf
Hope, this helps…
I don’t think mipmap is done in software. Maybe for better performance you can give all the detail level of the images instead of letting GL calculates it itself.
NPOT are a bit slower, but as far I know, it is on the uses not on creation: I don’t see why it could take more time to send an amount of data just because it is not power of two. Use is different because it concerns memory alignment and calculations. Here not.