Sequential Optimizations : loop fusion, tiling, and unrolling

[said-hadjout]

OpenMP 5.1 has some loop optimization features such as loop tiling and loop unrolling, previously the user either coded the loop transformation by himself or used some compiler hints like

Unrolling Compiler Hints

Cray
#pragma [_CRI] unroll 4
icc
#pragma unroll 4
xlc
#pragma unroll(4)
HP
#pragma UNROLL_FACTOR 4
gcc
#pragma GCC unroll 4
clang
#pragma unroll 4
#pragma clang loop unroll_count(4)

We should have a unified way for the user to ask for certain optimizations and provide them either through OpenMP if available, compiler hints, or doing the loop transformation in the code generation directly.

It would be nice if we could provide an example where we use all the optimization techniques, and show the timing improvement with each technique, in the same spirit of this example in MOJO https://www.youtube.com/watch?v=6GvB5lZJqcE&t=0s