[llvm] [Codegen][LegalizeIntegerTypes] Improve shift through stack (PR #96151)

[Matt Arsenault via llvm-commits]

================
@@ -4551,10 +4560,11 @@ void DAGTypeLegalizer::ExpandIntRes_ShiftThroughStack(SDNode *N, SDValue &Lo,
 
   // Get a temporary stack slot 2x the width of our VT.
   // FIXME: reuse stack slots?
-  // FIXME: should we be more picky about alignment?
-  Align StackSlotAlignment(1);
-  SDValue StackPtr = DAG.CreateStackTemporary(
-      TypeSize::getFixed(StackSlotByteWidth), StackSlotAlignment);
+  Align StackAlign = DAG.getReducedAlign(StackSlotVT, /*UseABI=*/false);
+  assert(DAG.getReducedAlign(LoadVT, /*UseABI=*/false) <= StackAlign);
----------------
arsenm wrote:

Don't see the point of this assert 

https://github.com/llvm/llvm-project/pull/96151