[llvm] [AMDGPU] Speedup SIFormMemoryClauses live-in register set calculation. (PR #89919)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

>  This is faster because start slot indexes for clauses are ordered and LIS live segments for a register are also ordered so the getLiveRegMap finds a range of start slot indexes that fall into a given live segment of a register.

I don't understand why this is faster. There's no reordering going on here, so the segments should still be ordered. 

I still think this whole pass is using far too much machinery, trying to be precise for pressure tracking that is not actually precise 

https://github.com/llvm/llvm-project/pull/89919