diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -20524,7 +20524,7 @@
   EVT PtrVT = Ptr.getValueType();
 
   SDValue Offset =
-      DAG.getNode(ISD::MUL, DL, PtrVT, Idx,
+      DAG.getNode(ISD::MUL, DL, PtrVT, DAG.getZExtOrTrunc(Idx, DL, PtrVT),
                   DAG.getConstant(EltVT.getSizeInBits() / 8, DL, PtrVT));
   SDValue NewPtr = DAG.getNode(ISD::ADD, DL, PtrVT, Ptr, Offset);
   MachinePointerInfo PointerInfo(ST->getAddressSpace());
diff --git a/llvm/test/CodeGen/AMDGPU/replace-store-of-insert-load.ll b/llvm/test/CodeGen/AMDGPU/replace-store-of-insert-load.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/replace-store-of-insert-load.ll
@@ -0,0 +1,58 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a < %s | FileCheck %s
+
+; Regression test for a bug in `DAGCombiner::replaceStoreOfInsertLoad` where
+; Idx could be smaller than PtrVT, causing a MUL to be emitted with inconsistent
+; LHS/RHS types.
+
+define void @testcase_0(ptr addrspace(1) %in, float %arg) {
+; CHECK-LABEL: testcase_0:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_store_dword v[0:1], v2, off offset:12
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %loaded = load <4 x float>, ptr addrspace(1) %in
+  %modified = insertelement <4 x float> %loaded, float %arg, i64 3
+  store <4 x float> %modified, ptr addrspace(1) %in
+  ret void
+}
+
+define void @testcase_1(ptr addrspace(1) %in, float %arg) {
+; CHECK-LABEL: testcase_1:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_store_dword v[0:1], v2, off offset:16
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %loaded = load <6 x float>, ptr addrspace(1) %in
+  %modified = insertelement <6 x float> %loaded, float %arg, i64 4
+  store <6 x float> %modified, ptr addrspace(1) %in
+  ret void
+}
+
+define void @testcase_2(ptr addrspace(1) %in, double %arg) {
+; CHECK-LABEL: testcase_2:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off offset:8
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %loaded = load <4 x double>, ptr addrspace(1) %in
+  %modified = insertelement <4 x double> %loaded, double %arg, i64 1
+  store <4 x double> %modified, ptr addrspace(1) %in
+  ret void
+}
+
+define void @testcase_3(ptr addrspace(1) %in, double %arg) {
+; CHECK-LABEL: testcase_3:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off offset:56
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %loaded = load <8 x double>, ptr addrspace(1) %in
+  %modified = insertelement <8 x double> %loaded, double %arg, i64 7
+  store <8 x double> %modified, ptr addrspace(1) %in
+  ret void
+}