[llvm] [DAGCombiner] Only fold shift+mask compare to rotate if the shift amount divides the bit width (PR #220743)

[Stanislav Bardyuk via llvm-commits]

================
@@ -14980,7 +14980,7 @@ SDValue DAGCombiner::visitSETCC(SDNode *N) {
   // If C0 is a mask or shifted mask and the shift amt (C1) isolates the
   // remaining bits (i.e something like `(x64 & UINT32_MAX) == (x64 >> 32)`)
   // Then:
-  // If C1 is a power of 2, then the rotate and shift+and versions are
+  // If C1 divides the bit width, then the rotate and shift+and versions are
   // equivilent, so we can interchange them depending on target preference.
----------------
kodlan wrote:

done

https://github.com/llvm/llvm-project/pull/220743