I would like to know if it worth it to change the shader/textures of my opengl es 3.0 Image processing project to use integer sampling operations etc. Since the image data is always uint8 I though it might be faster then going to float?
If you want to prioritise speed over accuracy, use “lowp” precision, which can be implemented as fixed-point.
If anything, using integer arithmetic may be slower. The semantics are more precisely defined, so the implementation has less scope for optimisation or approximation.