Section 8.2 Addressing mode and filtering

I have a query about this section.
The spec mentions it usees Manhattan distance for finding nearest neighbor but the formula used is floor(), which is not correct.

Also for Linear filtering, 0.5f is being deducted from unnormalized coordinates, which also seems to be inconsistant.

amI missing something here? was these equations taken from another statndard?
I’m not familiar with any other standard.


In both OpenCL and OpenGL, pixels and texels are sampled in the center of the area they cover. explains it.

I believe this accounts for both of your observations.