BiCubic Intepol hardware oprerator for bicubic interpol

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I want to use the hardware operator for BiCubic interpol like the BiLinear Interpol (tex2D(tex, i, j); and tex.filterMode = cudaFilterModeLinear;).

There is the cudaFilterModeLinear and cudaFilterModePoint but not the cudaFilterModeCubic !!!

Would you have a solution ?

Thanks for your help.

PS : I know how to do it in the software way but this is not what I am looking for.

2

All you can do is hope that the shortly forthcoming CUDA 2.0 beta includes cubic interpolation. If not, your only option is to do it yourself in the kernel.

3

The texture hardware doesn’t support bicubic interpolation, but it is possible to compose a cubic filter out of 4 bilinear lookups:

[url=“http://developer.download.nvidia.com/SDK/9.5/Samples/DEMOS/OpenGL/src/fast_third_order/docs/Gems2_ch20_SDK.pdf”]http://developer.download.nvidia.com/SDK/9...s2_ch20_SDK.pdf[/url]

I guess this might be a nice sample to add to the SDK.

4

Thanks,

Looking forward to trying Cuda 2.0

i tryed the GEM2 sample but the results were false, or the approximations weren’t precise enough for my implemantation.

5

Yes, the only problem with this technique is that the texture coordinates only have 9 bits of fractional position in the hardware. If you want more accurate results you will have to do 16 texture lookups.

6

Btw, Lanczos interpolation isn’t too hard to do. It’s not especially hardware accelerated, but the results are more accurate than cubic anyway.

7

Have a look at the RenderMan interface specfication, its addition includes some useful code.