Index: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2895,6 +2895,16 @@
     return DAG.getNode(ISD::SUB, DL, VT, N0.getOperand(0), NewC);
   }
 
+  // fold (c1-A)-c2 -> (c1-c2)-A
+  if (N0.getOpcode() == ISD::SUB &&
+      isConstantOrConstantVector(N1, /* NoOpaques */ true) &&
+      isConstantOrConstantVector(N0.getOperand(0), /* NoOpaques */ true)) {
+    SDValue NewC = DAG.FoldConstantArithmetic(
+        ISD::SUB, DL, VT, N0.getOperand(0).getNode(), N1.getNode());
+    assert(NewC && "Constant folding failed");
+    return DAG.getNode(ISD::SUB, DL, VT, NewC, N0.getOperand(1));
+  }
+
   // fold ((A+(B+or-C))-B) -> A+or-C
   if (N0.getOpcode() == ISD::ADD &&
       (N0.getOperand(1).getOpcode() == ISD::SUB ||
Index: llvm/trunk/test/CodeGen/AArch64/addsub-constant-folding.ll
===================================================================
--- llvm/trunk/test/CodeGen/AArch64/addsub-constant-folding.ll
+++ llvm/trunk/test/CodeGen/AArch64/addsub-constant-folding.ll
@@ -344,10 +344,8 @@
 define <4 x i32> @const_sub_sub_const(<4 x i32> %arg) {
 ; CHECK-LABEL: const_sub_sub_const:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v1.4s, #8
+; CHECK-NEXT:    movi v1.4s, #6
 ; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
-; CHECK-NEXT:    movi v1.4s, #2
-; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    ret
   %t0 = sub <4 x i32> <i32 8, i32 8, i32 8, i32 8>, %arg
   %t1 = sub <4 x i32> %t0, <i32 2, i32 2, i32 2, i32 2>
@@ -362,13 +360,13 @@
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset w30, -16
 ; CHECK-NEXT:    movi v1.4s, #8
-; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
 ; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
+; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
 ; 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:    movi v0.4s, #2
-; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
+; CHECK-NEXT:    movi v0.4s, #6
+; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    add sp, sp, #32 // =32
 ; CHECK-NEXT:    ret
   %t0 = sub <4 x i32> <i32 8, i32 8, i32 8, i32 8>, %arg
@@ -382,10 +380,7 @@
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI23_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI23_0]
-; CHECK-NEXT:    adrp x8, .LCPI23_1
-; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI23_1]
 ; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
-; CHECK-NEXT:    sub v0.4s, v0.4s, v2.4s
 ; CHECK-NEXT:    ret
   %t0 = sub <4 x i32> <i32 21, i32 undef, i32 8, i32 8>, %arg
   %t1 = sub <4 x i32> %t0, <i32 2, i32 3, i32 undef, i32 2>
Index: llvm/trunk/test/CodeGen/X86/addsub-constant-folding.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/addsub-constant-folding.ll
+++ llvm/trunk/test/CodeGen/X86/addsub-constant-folding.ll
@@ -500,17 +500,15 @@
 define <4 x i32> @const_sub_sub_const(<4 x i32> %arg) {
 ; X86-LABEL: const_sub_sub_const:
 ; X86:       # %bb.0:
-; X86-NEXT:    movdqa {{.*#+}} xmm1 = [8,8,8,8]
+; X86-NEXT:    movdqa {{.*#+}} xmm1 = [6,6,6,6]
 ; X86-NEXT:    psubd %xmm0, %xmm1
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: const_sub_sub_const:
 ; X64:       # %bb.0:
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [8,8,8,8]
+; X64-NEXT:    movdqa {{.*#+}} xmm1 = [6,6,6,6]
 ; X64-NEXT:    psubd %xmm0, %xmm1
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
   %t0 = sub <4 x i32> <i32 8, i32 8, i32 8, i32 8>, %arg
@@ -523,13 +521,14 @@
 ; X86:       # %bb.0:
 ; X86-NEXT:    subl $28, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 32
-; X86-NEXT:    movdqa {{.*#+}} xmm1 = [8,8,8,8]
-; X86-NEXT:    psubd %xmm0, %xmm1
-; X86-NEXT:    movdqu %xmm1, (%esp) # 16-byte Spill
-; X86-NEXT:    movdqa %xmm1, %xmm0
+; X86-NEXT:    movdqa %xmm0, %xmm1
+; X86-NEXT:    movdqu %xmm0, (%esp) # 16-byte Spill
+; X86-NEXT:    movdqa {{.*#+}} xmm0 = [8,8,8,8]
+; X86-NEXT:    psubd %xmm1, %xmm0
 ; X86-NEXT:    calll use
-; X86-NEXT:    movdqu (%esp), %xmm0 # 16-byte Reload
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
+; X86-NEXT:    movdqa {{.*#+}} xmm0 = [6,6,6,6]
+; X86-NEXT:    movdqu (%esp), %xmm1 # 16-byte Reload
+; X86-NEXT:    psubd %xmm1, %xmm0
 ; X86-NEXT:    addl $28, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NEXT:    retl
@@ -538,13 +537,13 @@
 ; X64:       # %bb.0:
 ; X64-NEXT:    subq $24, %rsp
 ; X64-NEXT:    .cfi_def_cfa_offset 32
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [8,8,8,8]
-; X64-NEXT:    psubd %xmm0, %xmm1
-; X64-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
-; X64-NEXT:    movdqa %xmm1, %xmm0
+; X64-NEXT:    movdqa %xmm0, %xmm1
+; X64-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
+; X64-NEXT:    movdqa {{.*#+}} xmm0 = [8,8,8,8]
+; X64-NEXT:    psubd %xmm1, %xmm0
 ; X64-NEXT:    callq use
-; X64-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm0 = [6,6,6,6]
+; X64-NEXT:    psubd (%rsp), %xmm0 # 16-byte Folded Reload
 ; X64-NEXT:    addq $24, %rsp
 ; X64-NEXT:    .cfi_def_cfa_offset 8
 ; X64-NEXT:    retq
@@ -557,17 +556,15 @@
 define <4 x i32> @const_sub_sub_const_nonsplat(<4 x i32> %arg) {
 ; X86-LABEL: const_sub_sub_const_nonsplat:
 ; X86:       # %bb.0:
-; X86-NEXT:    movdqa {{.*#+}} xmm1 = <21,u,8,8>
+; X86-NEXT:    movdqa {{.*#+}} xmm1 = <19,u,u,6>
 ; X86-NEXT:    psubd %xmm0, %xmm1
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: const_sub_sub_const_nonsplat:
 ; X64:       # %bb.0:
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = <21,u,8,8>
+; X64-NEXT:    movdqa {{.*#+}} xmm1 = <19,u,u,6>
 ; X64-NEXT:    psubd %xmm0, %xmm1
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
   %t0 = sub <4 x i32> <i32 21, i32 undef, i32 8, i32 8>, %arg