[llvm] [Analysis] Optimise mul(const, (udiv %n, const)) during SCEV expansion (PR #212769)

[David Sherwood via llvm-commits]

david-arm wrote:

It's possible to generalise this to more cases, e.g. `shl (lshr %x, c1), c2` where c2>c1, etc. This does also show improvements in places and is what instcombine would do, however without a good way to measure use counts of the SCEV nodes during expansion it's difficult to assess the profitability of the transformation.

https://github.com/llvm/llvm-project/pull/212769