convolution texture

Accelerated Computing CUDA CUDA Programming and Performance
#1

Good afternoon.

I am trying to understand how the kernel of convolutionTexture work.

I would like to know why 0,5f is added to the thread index:

const float x = (float)ix + 0.5f;
const float y = (float)iy + 0.5f;

thank you

#2

Hi,

I think it’s done to read from “the middle” of the pixels. You get more accurate results overall like that, I think, because it’s probably interpolating over the neighboring pixels, so you want to adjust the “empirical” thread numbers to their geometric usage.

#3

Thank you for the answer.