i'm working on part of an app the creates non-power-of-2 images for me - i want to display them on textures so i need to scale them up to power-of-2.

i'm not able to do it the sensible way and use opengl to scale UVs.

i've looked at some bilinear scaling algorithms but haven't found anything quite right - too slow or not good enough quality.

any suggestions on how i might do this with hardware - i.e. write a method that is pass in an arbitrary sized buffer and it gets scaled to a new, high quality power-of-2 one quickly..

thanks in advance.

What API are you using? If you can use GDI you can use the StretchBlt function although that doesn't do any filtering. GDI+ has some fancy high quality filtering options but I don't think current GDI+ drivers use hardware. Otherwise, what about Direct3D or DirectDraw? I guess if you can't use OpenGL then those two are out of the question..

cheers
sam.

thanks sam but there is no common API except opengl - it's a cross platform (win32, mac osx & linux) application.

How about generating the images onto a power-of-two texture that is larger than the actual image (so that if you have a 100x147 image you generate it onto a 128x256 texture, leaving parts of it blank).

Then you can either use that texture directly, ignoring the wasted space(and memory) or use hardware to draw a quad using the used part of that texture onto another texture that is power-of-two, thus using the built in filtering of the 3D hardware.

You don't need to stretch the the images!
just define a power of 2 texture width and height big enought to fit the image. Then copy the image to the texture, and place it at UV origin (0,0), then just use the currect texture coords, to map only the image, without using the extra texture area that is left out.

In a more specialized case, if your images are small and many, just create a big texture, and try to fill it (as optimally as you can with images, then use this texture, but adjust the texture coords, according to where the image you want is located in the texture.

Not fair 8) same time, nearly same content, and your post gets here first !
Touché.

I wrote a bicubic image resize routine some time ago, here's a few screenshots how it works out in practise:
http://www.liimatta.org/fusion/filter/filter.html

API example (C++):

surface* x = surface::create("test.png");
x->resize(666,42);
// ...

p.s. I didn't run sharpening filter on the bicubic (scaled up) image, if wondering why it is slightly blurry. It really depends when you want to so the code doesn't do that automatically..

that looks very interesting - thank you - i'll take a look at the library.