diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -456,8 +456,15 @@
 
       // Check to see if this stored value is of the same byte-splattable value.
       Value *StoredByte = isBytewiseValue(StoredVal, DL);
+      // If the previously stored value is undef, we can replace it with the
+      // newly stored value (which may be undef)
+      // Inversely, if the newly stored value is undef, we can replace it with
+      // the previously stored one
       if (isa<UndefValue>(ByteVal) && StoredByte)
         ByteVal = StoredByte;
+      else if (StoredByte && isa<UndefValue>(StoredByte))
+        StoredByte = ByteVal;
+
       if (ByteVal != StoredByte)
         break;
 
diff --git a/llvm/test/Transforms/MemCpyOpt/merge-undef-memset.ll b/llvm/test/Transforms/MemCpyOpt/merge-undef-memset.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/MemCpyOpt/merge-undef-memset.ll
@@ -0,0 +1,63 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -memcpyopt -opaque-pointers -S -verify-memoryssa | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+%S = type { { i8, i8 }, { i8, i8 }, { i8, i8 }, { i8, i8 }, { i8, i8 }, { i8, i8 }, { i8, i8 }, { i8, i8 } }
+
+define void @alternating(%S* %0) {
+; CHECK-LABEL: @alternating(
+; CHECK-NEXT: start
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT:  getelementptr
+; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 1 %0, i8 0, i64 16, i1 false)
+; CHECK-NEXT: ret void
+start:
+  store i8 0, ptr %0, align 1
+  %1 = getelementptr inbounds { i8, i8 }, ptr %0, i32 0, i32 1
+  store i8 undef, ptr %1, align 1
+  %2 = getelementptr inbounds %S, ptr %0, i32 0, i32 1
+  store i8 0, ptr %2, align 1
+  %3 = getelementptr inbounds { i8, i8 }, ptr %2, i32 0, i32 1
+  store i8 undef, ptr %3, align 1
+  %4 = getelementptr inbounds %S, ptr %0, i32 0, i32 2
+  store i8 0, ptr %4, align 1
+  %5 = getelementptr inbounds { i8, i8 }, ptr %4, i32 0, i32 1
+  store i8 undef, ptr %5, align 1
+  %6 = getelementptr inbounds %S, ptr %0, i32 0, i32 3
+  store i8 0, ptr %6, align 1
+  %7 = getelementptr inbounds { i8, i8 }, ptr %6, i32 0, i32 1
+  store i8 undef, ptr %7, align 1
+  %8 = getelementptr inbounds %S, ptr %0, i32 0, i32 4
+  store i8 0, ptr %8, align 1
+  %9 = getelementptr inbounds { i8, i8 }, ptr %8, i32 0, i32 1
+  store i8 undef, ptr %9, align 1
+  %10 = getelementptr inbounds %S, ptr %0, i32 0, i32 5
+  store i8 0, ptr %10, align 1
+  %11 = getelementptr inbounds { i8, i8 }, ptr %10, i32 0, i32 1
+  store i8 undef, ptr %11, align 1
+  %12 = getelementptr inbounds %S, ptr %0, i32 0, i32 6
+  store i8 0, ptr %12, align 1
+  %13 = getelementptr inbounds { i8, i8 }, ptr %12, i32 0, i32 1
+  store i8 undef, ptr %13, align 1
+  %14 = getelementptr inbounds %S, ptr %0, i32 0, i32 7
+  store i8 0, ptr %14, align 1
+  %15 = getelementptr inbounds { i8, i8 }, ptr %14, i32 0, i32 1
+  store i8 undef, ptr %15, align 1
+  ret void
+}
+
+declare void @llvm.memset.p0i8.i64(ptr, i8, i64, i1)