Optimise modulo comparison operation

[ilinpv]

When the result of a modulo by constant is compared against another constant we can sometimes avoid the expensive modulo instructions as described in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87290

For example

int
mod16 (int a, int b)
{
  return a % 16 == 13;
}

with -O2 using this on aarch64 GCC produces better code: https://godbolt.org/z/9K8Gd1Eaf

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

[llvmbot]

@llvm/issue-subscribers-good-first-issue

[rotateright]

This seems more like a missed generic optimization in IR. Any compare of srem by power-of-2 can be turned into an 'and' mask pattern instead:
https://alive2.llvm.org/ce/z/6daHCq

[EugeneZelenko]

@ilinpv: please read comment for beginner label.

[ilinpv]

@EugeneZelenko could we keep beginner label for the ticket to have it in old filter ( compatible with old tickets using beginner label only ) ?