diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -5546,6 +5546,10 @@
 def DUPv8i8lane  : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
 
+def : Pat<(int_aarch64_neon_pmull64 (extractelt (v2i64 V128:$Rn), (i64 1)),
+                                    V64:$Rm),
+          (PMULLv2i64 V128:$Rn, (v2f64 (DUPv2i64lane (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), V64:$Rm, dsub), (i64 0))))>;
+
 // DUP from a 64-bit register to a 64-bit register is just a copy
 def : Pat<(v1i64 (AArch64dup (i64 GPR64:$Rn))),
           (COPY_TO_REGCLASS GPR64:$Rn, FPR64)>;
diff --git a/llvm/test/CodeGen/AArch64/pmull-ldr-merge.ll b/llvm/test/CodeGen/AArch64/pmull-ldr-merge.ll
--- a/llvm/test/CodeGen/AArch64/pmull-ldr-merge.ll
+++ b/llvm/test/CodeGen/AArch64/pmull-ldr-merge.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -verify-machineinstrs  -mtriple=aarch64-linux-gnu -mattr=+aes -o - %s| FileCheck %s --check-prefixes=CHECK
+; RUN: llc -O3 -verify-machineinstrs  -mtriple=aarch64-linux-gnu -mattr=+aes -o - %s| FileCheck %s --check-prefixes=CHECK
 
 ; Tests that scalar i64 arguments of llvm.aarch64.neon.pmull64 are
 ; loaded into SIMD registers, as opposed to being loaded into GPR followed by a mov.
@@ -60,4 +60,36 @@
   ret void
 }
 
+define void @test4(ptr %0, ptr %1) {
+; CHECK-LABEL: test4:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr q0, [x1]
+; CHECK-NEXT:    mov w9, #56824
+; CHECK-NEXT:    mov w10, #61186
+; CHECK-NEXT:    movk w9, #40522, lsl #16
+; CHECK-NEXT:    movk w10, #29710, lsl #16
+; CHECK-NEXT:    mov x8, v0.d[1]
+; CHECK-NEXT:    fmov d1, x9
+; CHECK-NEXT:    fmov d3, x10
+; CHECK-NEXT:    fmov d2, x8
+; CHECK-NEXT:    pmull v0.1q, v0.1d, v3.1d
+; CHECK-NEXT:    pmull v1.1q, v2.1d, v1.1d
+; CHECK-NEXT:    ldr q2, [x0]
+; CHECK-NEXT:    eor v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    eor v0.16b, v2.16b, v0.16b
+; CHECK-NEXT:    str q0, [x1]
+; CHECK-NEXT:    ret
+  %3 = load <2 x i64>, ptr %1
+  %4 = extractelement <2 x i64> %3, i64 1
+  %5 = tail call <16 x i8> @llvm.aarch64.neon.pmull64(i64 %4, i64 2655706616)
+  %6 = load <2 x i64>, ptr %0
+  %7 = extractelement <2 x i64> %3, i64 0
+  %8 = tail call <16 x i8> @llvm.aarch64.neon.pmull64(i64 %7, i64 1947135746)
+  %9 = xor <16 x i8> %8, %5
+  %10 = bitcast <16 x i8> %9 to <2 x i64>
+  %11 = xor <2 x i64> %6, %10
+  store <2 x i64> %11, ptr %1
+  ret void
+}
+
 declare <16 x i8> @llvm.aarch64.neon.pmull64(i64, i64)