[llvm] [AMDGPU] Fix SIFoldOperands miscompiling values that leave a divergent loop (PR #203256)

[Matt Arsenault via llvm-commits]

================
@@ -970,6 +982,22 @@ bool SIFoldOperandsImpl::isUseSafeToFold(const MachineInstr &MI,
   return !TII->isSDWA(MI);
 }
 
+// An SGPR->VGPR copy inside a divergent loop latches each lane value as it
----------------
arsenm wrote:

This is really a representational problem. We shouldn't have to special case a copy like this. Any other pass is free to do the same 

https://github.com/llvm/llvm-project/pull/203256