diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -5708,7 +5708,8 @@
       // Create a scalar GEP if there are more than 2 operands.
       if (Ops.size() != 2) {
         // Replace the last index with 0.
-        Ops[FinalIndex] = Constant::getNullValue(ScalarIndexTy);
+        Ops[FinalIndex] =
+            Constant::getNullValue(Ops[FinalIndex]->getType()->getScalarType());
         Base = Builder.CreateGEP(SourceTy, Base, ArrayRef(Ops).drop_front());
         SourceTy = GetElementPtrInst::getIndexedType(
             SourceTy, ArrayRef(Ops).drop_front());
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/masked-gather-struct-gep.ll b/llvm/test/Transforms/CodeGenPrepare/X86/masked-gather-struct-gep.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/masked-gather-struct-gep.ll
@@ -0,0 +1,23 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S -codegenprepare < %s | FileCheck %s
+; REQUIRES: x86-registered-target
+target triple = "x86_64-pc-linux"
+
+%s = type <{ float, i32, i8, [3 x i8] }>
+
+declare <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr>, i32 immarg, <4 x i1>, <4 x float>)
+
+define <4 x float> @foo(ptr %p) {
+; CHECK-LABEL: define <4 x float> @foo
+; CHECK-SAME: (ptr [[P:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast ptr [[P]] to ptr
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr float, ptr [[TMP1]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[GATHER:%.*]] = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> [[TMP2]], i32 0, <4 x i1> zeroinitializer, <4 x float> zeroinitializer)
+; CHECK-NEXT:    ret <4 x float> [[GATHER]]
+;
+  %base.splatinsert = insertelement <4 x ptr> poison, ptr %p, i32 0
+  %base = shufflevector <4 x ptr> %base.splatinsert, <4 x ptr> poison, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
+  %gep = getelementptr %s, <4 x ptr> %base, <4 x i64> zeroinitializer, <4 x i32> zeroinitializer
+  %gather = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> zeroinitializer, <4 x float> zeroinitializer)
+  ret <4 x float> %gather
+}