[SelectionDAG] Use unaligned store/load to move AVX registers onto stack for insertelement

Prior to this patch, SelectionDAG generated aligned move onto stacks for AVX registers
when the function was marked as a no-realign-stack function. This lead to misalignment
between the stack and the instruction generated. This patch fixes the issue. There was
a similar issue reported for `extractelement` which was fixed in #a6614ec5b7c1dbfc4b847884c5de780cf75e8e9c

Author: Nirhar

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

@@ -398,22 +398,24 @@ SDValue SelectionDAGLegalize::PerformInsertVectorEltInMemory(SDValue Vec,
   EVT EltVT = VT.getVectorElementType();
   SDValue StackPtr = DAG.CreateStackTemporary(VT);
 
-  int SPFI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
+  MachineMemOperand *AlignedMMO = getStackAlignedMMO(
+      StackPtr, DAG.getMachineFunction(), EltVT.isScalableVector());
 
   // Store the vector.
-  SDValue Ch = DAG.getStore(
-      DAG.getEntryNode(), dl, Tmp1, StackPtr,
-      MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SPFI));
+  SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, AlignedMMO);
+
 
   SDValue StackPtr2 = TLI.getVectorElementPointer(DAG, StackPtr, VT, Tmp3);
 
   // Store the scalar value.
-  Ch = DAG.getTruncStore(
-      Ch, dl, Tmp2, StackPtr2,
-      MachinePointerInfo::getUnknownStack(DAG.getMachineFunction()), EltVT);
-  // Load the updated vector.
-  return DAG.getLoad(VT, dl, Ch, StackPtr, MachinePointerInfo::getFixedStack(
-                                               DAG.getMachineFunction(), SPFI));
+  Ch = DAG.getTruncStore(Ch, dl, Tmp2, StackPtr2, EltVT, AlignedMMO);
+
+  Align ElementAlignment = std::min(cast<StoreSDNode>(Ch)->getAlign(),
+                                    DAG.getDataLayout().getPrefTypeAlign(
+                                        VT.getTypeForEVT(*DAG.getContext())));
+
+  return DAG.getLoad(VT, dl, Ch, StackPtr, MachinePointerInfo(),
+                     ElementAlignment);
 }
 
 SDValue SelectionDAGLegalize::ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val,
@@ -1489,24 +1491,27 @@ SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) {
   EVT VecVT = Vec.getValueType();
   EVT SubVecVT = Part.getValueType();
   SDValue StackPtr = DAG.CreateStackTemporary(VecVT);
-  int FI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
-  MachinePointerInfo PtrInfo =
-      MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI);
+  MachineMemOperand *AlignedMMO = getStackAlignedMMO(
+      StackPtr, DAG.getMachineFunction(), VecVT.isScalableVector());
 
   // First store the whole vector.
-  SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, PtrInfo);
+  SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, AlignedMMO);
 
   // Then store the inserted part.
   SDValue SubStackPtr =
       TLI.getVectorSubVecPointer(DAG, StackPtr, VecVT, SubVecVT, Idx);
 
   // Store the subvector.
-  Ch = DAG.getStore(
-      Ch, dl, Part, SubStackPtr,
-      MachinePointerInfo::getUnknownStack(DAG.getMachineFunction()));
+  Ch = DAG.getStore(Ch, dl, Part, SubStackPtr, AlignedMMO);
+
+  Align ElementAlignment =
+      std::min(cast<StoreSDNode>(Ch)->getAlign(),
+               DAG.getDataLayout().getPrefTypeAlign(
+                   Op.getValueType().getTypeForEVT(*DAG.getContext())));
 
   // Finally, load the updated vector.
-  return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, PtrInfo);
+  return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, MachinePointerInfo(),
+                     ElementAlignment);
 }
 
 SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) {

llvm/test/CodeGen/X86/unaligned-insert-into-vector-through-stack.ll

@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu | FileCheck %s
+
+define <8 x i32> @foo(<8 x i32> %arg1, i32 %n) #0 {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovups %ymm0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    # kill: def $edi killed $edi def $rdi
+; CHECK-NEXT:    andl $7, %edi
+; CHECK-NEXT:    movl $42, -40(%rsp,%rdi,4)
+; CHECK-NEXT:    vmovups -{{[0-9]+}}(%rsp), %ymm0
+; CHECK-NEXT:    retq
+entry:
+  %a = insertelement <8 x i32> %arg1, i32 42, i32 %n
+  ret <8 x i32> %a
+}
+
+attributes #0 = { "no-realign-stack" "target-cpu"="haswell" }