The bottom part of the slide highlights where the scheduling that was declared in the Halide function will occur in the implementation. The code is doing SIMD operations for the xi variable and using probably OMP for y.

Also, I have a question. What is the type of x, y, xi and yi in the Halide function? Just an uint8?

Maybe a better way of phrasing my question is what are the semantics of x, y etc.

I wonder how the Halide compiler looks -- are there certain kinds of dependencies between cells/indices it knows how to represent in some way (as "handwritten kernels") and other less common kinds of subroutines have to get compiled from scratch, inefficiently? I know that frameworks like this are common in implementations of machine learning algorithms