Index: lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2870,6 +2870,16 @@
   if (N0.getOpcode() == ISD::ADD && N0.getOperand(1) == N1)
     return N0.getOperand(0);
 
+  // fold (A+C1)-C2 -> A+(C1-C2)
+  if (N0.getOpcode() == ISD::ADD &&
+      isConstantOrConstantVector(N1, /* NoOpaques */ true) &&
+      isConstantOrConstantVector(N0.getOperand(1), /* NoOpaques */ true)) {
+    SDValue NewC = DAG.FoldConstantArithmetic(
+        ISD::SUB, DL, VT, N0.getOperand(1).getNode(), N1.getNode());
+    assert(NewC && "Constant folding failed");
+    return DAG.getNode(ISD::ADD, DL, VT, N0.getOperand(0), NewC);
+  }
+
   // fold C2-(A+C1) -> (C2-C1)-A
   if (N1.getOpcode() == ISD::ADD) {
     SDValue N11 = N1.getOperand(1);
Index: test/CodeGen/AArch64/vec_add.ll
===================================================================
--- test/CodeGen/AArch64/vec_add.ll
+++ test/CodeGen/AArch64/vec_add.ll
@@ -42,13 +42,13 @@
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset w30, -16
 ; CHECK-NEXT:    movi v1.4s, #8
-; CHECK-NEXT:    add v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
+; CHECK-NEXT:    add v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    bl use
 ; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
 ; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
-; CHECK-NEXT:    mvni v0.4s, #1
-; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
+; CHECK-NEXT:    movi v0.4s, #10
+; CHECK-NEXT:    add v0.4s, v1.4s, v0.4s
 ; CHECK-NEXT:    add sp, sp, #32 // =32
 ; CHECK-NEXT:    ret
   %t0 = add <4 x i32> %arg, <i32 8, i32 8, i32 8, i32 8>
Index: test/CodeGen/X86/vec_add.ll
===================================================================
--- test/CodeGen/X86/vec_add.ll
+++ test/CodeGen/X86/vec_add.ll
@@ -53,11 +53,13 @@
 ; X86:       # %bb.0:
 ; X86-NEXT:    subl $28, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 32
-; X86-NEXT:    paddd {{\.LCPI.*}}, %xmm0
+; X86-NEXT:    movdqa %xmm0, %xmm1
 ; X86-NEXT:    movdqu %xmm0, (%esp) # 16-byte Spill
+; X86-NEXT:    movdqa {{.*#+}} xmm0 = [8,8,8,8]
+; X86-NEXT:    paddd %xmm1, %xmm0
 ; X86-NEXT:    calll use
 ; X86-NEXT:    movdqu (%esp), %xmm0 # 16-byte Reload
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
+; X86-NEXT:    paddd {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    addl $28, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NEXT:    retl
@@ -66,11 +68,13 @@
 ; X64:       # %bb.0:
 ; X64-NEXT:    subq $24, %rsp
 ; X64-NEXT:    .cfi_def_cfa_offset 32
-; X64-NEXT:    paddd {{.*}}(%rip), %xmm0
+; X64-NEXT:    movdqa %xmm0, %xmm1
 ; X64-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
+; X64-NEXT:    movdqa {{.*#+}} xmm0 = [8,8,8,8]
+; X64-NEXT:    paddd %xmm1, %xmm0
 ; X64-NEXT:    callq use
 ; X64-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
+; X64-NEXT:    paddd {{.*}}(%rip), %xmm0
 ; X64-NEXT:    addq $24, %rsp
 ; X64-NEXT:    .cfi_def_cfa_offset 8
 ; X64-NEXT:    retq