diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -17640,6 +17640,53 @@
   return SDValue();
 }
 
+/// Transform a vector operation on a splatted vector into a scalar operation on
+/// the splat value.
+static SDValue scalarizeSplatValue(SDValue Vec, EVT ResVT, SelectionDAG &DAG) {
+  // Don't scalarize if we need the full vector anyway.
+  // TODO: Handle the case where there are multiple extract_elt users
+  if (!Vec.hasOneUse() || Vec.getNode()->getNumValues() != 1)
+    return SDValue();
+
+  SDLoc DL(Vec);
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  EVT ScalarVT = Vec.getValueType().getScalarType();
+  unsigned Opc = Vec.getOpcode();
+
+  // Don't scalarize if the target does not support the scalar operation
+  if (!TLI.isOperationLegalOrCustomOrPromote(Opc, ResVT))
+    return SDValue();
+
+  // TODO: Check for all the structures recognized by SelectionDAG::isSplatValue
+  switch (Opc) {
+  default:
+    break;
+  case ISD::SPLAT_VECTOR: {
+    SDValue SplatVal = Vec.getOperand(0);
+    return ResVT == ScalarVT ? SplatVal
+                             : DAG.getAnyExtOrTrunc(SplatVal, DL, ResVT);
+  }
+  case ISD::BUILD_VECTOR: {
+    int SplatIdx;
+    if (!DAG.getSplatSourceVector(Vec, SplatIdx))
+      break;
+    SDValue SplatVal = Vec.getOperand(SplatIdx);
+    return ResVT == ScalarVT ? SplatVal
+                             : DAG.getAnyExtOrTrunc(SplatVal, DL, ResVT);
+  }
+  case ISD::ADD:
+  case ISD::SUB:
+  case ISD::AND: {
+    SDValue LHS, RHS;
+    if ((LHS = scalarizeSplatValue(Vec.getOperand(0), ResVT, DAG)) &&
+        (RHS = scalarizeSplatValue(Vec.getOperand(1), ResVT, DAG)))
+      return DAG.getNode(Opc, DL, ResVT, LHS, RHS);
+    break;
+  }
+  }
+  return SDValue();
+}
+
 SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
   SDValue VecOp = N->getOperand(0);
   SDValue Index = N->getOperand(1);
@@ -17751,6 +17798,9 @@
   if (SDValue BO = scalarizeExtractedBinop(N, DAG, LegalOperations))
     return BO;
 
+  if (SDValue SO = scalarizeSplatValue(VecOp, ScalarVT, DAG))
+    return SO;
+
   // Transform: (EXTRACT_VECTOR_ELT( VECTOR_SHUFFLE )) -> EXTRACT_VECTOR_ELT.
   // We only perform this optimization before the op legalization phase because
   // we may introduce new vector instructions which are not backed by TD
diff --git a/llvm/test/CodeGen/AArch64/arm64-nvcast.ll b/llvm/test/CodeGen/AArch64/arm64-nvcast.ll
--- a/llvm/test/CodeGen/AArch64/arm64-nvcast.ll
+++ b/llvm/test/CodeGen/AArch64/arm64-nvcast.ll
@@ -24,17 +24,8 @@
 define void @test2(float * %p1, i32 %v1) {
 ; CHECK-LABEL: test2:
 ; CHECK:       ; %bb.0: ; %entry
-; CHECK-NEXT:    sub sp, sp, #16 ; =16
-; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    ; kill: def $w1 killed $w1 def $x1
-; CHECK-NEXT:    movi.16b v0, #63
-; CHECK-NEXT:    and x8, x1, #0x3
-; CHECK-NEXT:    mov x9, sp
-; CHECK-NEXT:    str q0, [sp]
-; CHECK-NEXT:    bfi x9, x8, #2, #2
-; CHECK-NEXT:    ldr s0, [x9]
-; CHECK-NEXT:    str s0, [x0]
-; CHECK-NEXT:    add sp, sp, #16 ; =16
+; CHECK-NEXT:    mov w8, #1061109567
+; CHECK-NEXT:    str w8, [x0]
 ; CHECK-NEXT:    ret
 entry:
   %v2 = extractelement <3 x float> <float 0.7470588088035583, float 0.7470588088035583, float 0.7470588088035583>, i32 %v1
diff --git a/llvm/test/CodeGen/SystemZ/vec-trunc-to-i1.ll b/llvm/test/CodeGen/SystemZ/vec-trunc-to-i1.ll
--- a/llvm/test/CodeGen/SystemZ/vec-trunc-to-i1.ll
+++ b/llvm/test/CodeGen/SystemZ/vec-trunc-to-i1.ll
@@ -7,13 +7,10 @@
 ; CHECK-LABEL: pr32275:
 ; CHECK:       # %bb.0: # %BB
 ; CHECK-NEXT:    vlgvb %r0, %v24, 3
-; CHECK-NEXT:    vlvgp %v0, %r0, %r0
-; CHECK-NEXT:    vrepif %v1, 1
-; CHECK-NEXT:    vn %v0, %v0, %v1
-; CHECK-NEXT:    vlgvf %r0, %v0, 3
 ; CHECK-NEXT:  .LBB0_1: # %CF34
 ; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    cijlh %r0, 0, .LBB0_1
+; CHECK-NEXT:    tmll %r0, 1
+; CHECK-NEXT:    jne .LBB0_1
 ; CHECK-NEXT:  # %bb.2: # %CF36
 ; CHECK-NEXT:    br %r14
 BB:
diff --git a/llvm/test/CodeGen/WebAssembly/simd-shift-complex-splats.ll b/llvm/test/CodeGen/WebAssembly/simd-shift-complex-splats.ll
--- a/llvm/test/CodeGen/WebAssembly/simd-shift-complex-splats.ll
+++ b/llvm/test/CodeGen/WebAssembly/simd-shift-complex-splats.ll
@@ -10,12 +10,9 @@
 
 ; CHECK-LABEL: shl_add:
 ; CHECK-NEXT: .functype shl_add (v128, i32, i32) -> (v128)
-; CHECK-NEXT: i8x16.splat $push1=, $1
-; CHECK-NEXT: i8x16.splat $push0=, $2
-; CHECK-NEXT: i8x16.add $push2=, $pop1, $pop0
-; CHECK-NEXT: i8x16.extract_lane_u $push3=, $pop2, 0
-; CHECK-NEXT: i8x16.shl $push4=, $0, $pop3
-; CHECK-NEXT: return $pop4
+; CHECK-NEXT: i32.add $push0=, $1, $2
+; CHECK-NEXT: i8x16.shl $push1=, $0, $pop0
+; CHECK-NEXT: return $pop1
 define <16 x i8> @shl_add(<16 x i8> %v, i8 %a, i8 %b) {
   %t1 = insertelement <16 x i8> undef, i8 %a, i32 0
   %va = shufflevector <16 x i8> %t1, <16 x i8> undef, <16 x i32> zeroinitializer
diff --git a/llvm/test/CodeGen/X86/avx512-calling-conv.ll b/llvm/test/CodeGen/X86/avx512-calling-conv.ll
--- a/llvm/test/CodeGen/X86/avx512-calling-conv.ll
+++ b/llvm/test/CodeGen/X86/avx512-calling-conv.ll
@@ -785,18 +785,18 @@
 ; KNL-NEXT:    kmovw %edi, %k1
 ; KNL-NEXT:    kshiftlw $15, %k1, %k1
 ; KNL-NEXT:    kshiftrw $3, %k1, %k1
-; KNL-NEXT:    korw %k1, %k0, %k1
+; KNL-NEXT:    korw %k1, %k0, %k0
 ; KNL-NEXT:    movw $-8193, %di ## imm = 0xDFFF
-; KNL-NEXT:    kmovw %edi, %k0
-; KNL-NEXT:    kandw %k0, %k1, %k1
+; KNL-NEXT:    kmovw %edi, %k1
+; KNL-NEXT:    kandw %k1, %k0, %k0
 ; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %dil
 ; KNL-NEXT:    kmovw %edi, %k6
 ; KNL-NEXT:    kshiftlw $15, %k6, %k6
 ; KNL-NEXT:    kshiftrw $2, %k6, %k6
-; KNL-NEXT:    korw %k6, %k1, %k6
+; KNL-NEXT:    korw %k6, %k0, %k6
 ; KNL-NEXT:    movw $-16385, %di ## imm = 0xBFFF
-; KNL-NEXT:    kmovw %edi, %k1
-; KNL-NEXT:    kandw %k1, %k6, %k6
+; KNL-NEXT:    kmovw %edi, %k0
+; KNL-NEXT:    kandw %k0, %k6, %k6
 ; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %dil
 ; KNL-NEXT:    kmovw %edi, %k7
 ; KNL-NEXT:    kshiftlw $14, %k7, %k7
@@ -885,36 +885,26 @@
 ; KNL-NEXT:    kshiftlw $15, %k3, %k3
 ; KNL-NEXT:    kshiftrw $3, %k3, %k3
 ; KNL-NEXT:    korw %k3, %k2, %k2
-; KNL-NEXT:    kandw %k0, %k2, %k0
+; KNL-NEXT:    kandw %k1, %k2, %k1
 ; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %cl
 ; KNL-NEXT:    kmovw %ecx, %k2
 ; KNL-NEXT:    kshiftlw $15, %k2, %k2
 ; KNL-NEXT:    kshiftrw $2, %k2, %k2
-; KNL-NEXT:    korw %k2, %k0, %k0
-; KNL-NEXT:    xorl %ecx, %ecx
-; KNL-NEXT:    testb $1, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    movl $65535, %edx ## imm = 0xFFFF
-; KNL-NEXT:    movl $0, %esi
-; KNL-NEXT:    cmovnel %edx, %esi
-; KNL-NEXT:    testb $1, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    cmovnel %edx, %ecx
-; KNL-NEXT:    kandw %k1, %k0, %k0
-; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %dl
-; KNL-NEXT:    kmovw %edx, %k1
+; KNL-NEXT:    korw %k2, %k1, %k1
+; KNL-NEXT:    kandw %k0, %k1, %k0
+; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %cl
+; KNL-NEXT:    kmovw %ecx, %k1
 ; KNL-NEXT:    kshiftlw $14, %k1, %k1
 ; KNL-NEXT:    korw %k1, %k0, %k0
 ; KNL-NEXT:    kshiftlw $1, %k0, %k0
 ; KNL-NEXT:    kshiftrw $1, %k0, %k0
-; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %dl
-; KNL-NEXT:    kmovw %edx, %k1
+; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %cl
+; KNL-NEXT:    kmovw %ecx, %k1
 ; KNL-NEXT:    kshiftlw $15, %k1, %k1
 ; KNL-NEXT:    korw %k1, %k0, %k0
-; KNL-NEXT:    kmovw %esi, %k1
-; KNL-NEXT:    kmovw {{[-0-9]+}}(%r{{[sb]}}p), %k2 ## 2-byte Reload
-; KNL-NEXT:    kandw %k2, %k0, %k0
-; KNL-NEXT:    kmovw %ecx, %k2
-; KNL-NEXT:    kandw %k1, %k2, %k1
-; KNL-NEXT:    kmovw %k1, %r8d
+; KNL-NEXT:    kmovw {{[-0-9]+}}(%r{{[sb]}}p), %k1 ## 2-byte Reload
+; KNL-NEXT:    kandw %k1, %k0, %k0
+; KNL-NEXT:    movb {{[0-9]+}}(%rsp), %cl
 ; KNL-NEXT:    kshiftrw $1, %k0, %k1
 ; KNL-NEXT:    kmovw %k1, %r9d
 ; KNL-NEXT:    kshiftrw $2, %k0, %k1
@@ -936,28 +926,30 @@
 ; KNL-NEXT:    kshiftrw $10, %k0, %k1
 ; KNL-NEXT:    kmovw %k1, %ebp
 ; KNL-NEXT:    kshiftrw $11, %k0, %k1
-; KNL-NEXT:    kmovw %k1, %ecx
+; KNL-NEXT:    kmovw %k1, %r8d
 ; KNL-NEXT:    kshiftrw $12, %k0, %k1
 ; KNL-NEXT:    kmovw %k1, %edx
 ; KNL-NEXT:    kshiftrw $13, %k0, %k1
 ; KNL-NEXT:    kmovw %k1, %edi
 ; KNL-NEXT:    kshiftrw $14, %k0, %k1
-; KNL-NEXT:    andl $1, %r8d
-; KNL-NEXT:    movb %r8b, 2(%rax)
-; KNL-NEXT:    kmovw %k0, %r8d
-; KNL-NEXT:    andl $1, %r8d
+; KNL-NEXT:    andb {{[0-9]+}}(%rsp), %cl
+; KNL-NEXT:    movzbl %cl, %ecx
+; KNL-NEXT:    andl $1, %ecx
+; KNL-NEXT:    movb %cl, 2(%rax)
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    andl $1, %ecx
 ; KNL-NEXT:    andl $1, %r9d
-; KNL-NEXT:    leal (%r8,%r9,2), %r8d
-; KNL-NEXT:    kmovw %k1, %r9d
+; KNL-NEXT:    leal (%rcx,%r9,2), %r9d
+; KNL-NEXT:    kmovw %k1, %ecx
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    andl $1, %r10d
-; KNL-NEXT:    leal (%r8,%r10,4), %r8d
+; KNL-NEXT:    leal (%r9,%r10,4), %r9d
 ; KNL-NEXT:    kmovw %k0, %r10d
 ; KNL-NEXT:    andl $1, %r11d
-; KNL-NEXT:    leal (%r8,%r11,8), %r8d
+; KNL-NEXT:    leal (%r9,%r11,8), %r9d
 ; KNL-NEXT:    andl $1, %r12d
 ; KNL-NEXT:    shll $4, %r12d
-; KNL-NEXT:    orl %r8d, %r12d
+; KNL-NEXT:    orl %r9d, %r12d
 ; KNL-NEXT:    andl $1, %r15d
 ; KNL-NEXT:    shll $5, %r15d
 ; KNL-NEXT:    orl %r12d, %r15d
@@ -976,20 +968,20 @@
 ; KNL-NEXT:    shll $10, %ebp
 ; KNL-NEXT:    orl %esi, %ebp
 ; KNL-NEXT:    orl %r15d, %ebp
-; KNL-NEXT:    andl $1, %ecx
-; KNL-NEXT:    shll $11, %ecx
+; KNL-NEXT:    andl $1, %r8d
+; KNL-NEXT:    shll $11, %r8d
 ; KNL-NEXT:    andl $1, %edx
 ; KNL-NEXT:    shll $12, %edx
-; KNL-NEXT:    orl %ecx, %edx
+; KNL-NEXT:    orl %r8d, %edx
 ; KNL-NEXT:    andl $1, %edi
 ; KNL-NEXT:    shll $13, %edi
 ; KNL-NEXT:    orl %edx, %edi
-; KNL-NEXT:    andl $1, %r9d
-; KNL-NEXT:    shll $14, %r9d
-; KNL-NEXT:    orl %edi, %r9d
+; KNL-NEXT:    andl $1, %ecx
+; KNL-NEXT:    shll $14, %ecx
+; KNL-NEXT:    orl %edi, %ecx
 ; KNL-NEXT:    andl $1, %r10d
 ; KNL-NEXT:    shll $15, %r10d
-; KNL-NEXT:    orl %r9d, %r10d
+; KNL-NEXT:    orl %ecx, %r10d
 ; KNL-NEXT:    orl %ebp, %r10d
 ; KNL-NEXT:    movw %r10w, (%rax)
 ; KNL-NEXT:    popq %rbx
@@ -1538,107 +1530,99 @@
 ; KNL_X32-NEXT:    kshiftlw $15, %k2, %k2
 ; KNL_X32-NEXT:    kshiftrw $2, %k2, %k2
 ; KNL_X32-NEXT:    korw %k2, %k1, %k1
-; KNL_X32-NEXT:    xorl %eax, %eax
-; KNL_X32-NEXT:    testb $1, {{[0-9]+}}(%esp)
-; KNL_X32-NEXT:    movl $65535, %ecx ## imm = 0xFFFF
-; KNL_X32-NEXT:    movl $0, %edx
-; KNL_X32-NEXT:    cmovnel %ecx, %edx
 ; KNL_X32-NEXT:    kandw %k0, %k1, %k0
-; KNL_X32-NEXT:    movb {{[0-9]+}}(%esp), %bl
-; KNL_X32-NEXT:    kmovw %ebx, %k1
+; KNL_X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; KNL_X32-NEXT:    kmovw %eax, %k1
 ; KNL_X32-NEXT:    kshiftlw $14, %k1, %k1
 ; KNL_X32-NEXT:    korw %k1, %k0, %k0
 ; KNL_X32-NEXT:    kshiftlw $1, %k0, %k0
 ; KNL_X32-NEXT:    kshiftrw $1, %k0, %k0
-; KNL_X32-NEXT:    movb {{[0-9]+}}(%esp), %bl
-; KNL_X32-NEXT:    kmovw %ebx, %k1
+; KNL_X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; KNL_X32-NEXT:    kmovw %eax, %k1
 ; KNL_X32-NEXT:    kshiftlw $15, %k1, %k1
 ; KNL_X32-NEXT:    korw %k1, %k0, %k0
-; KNL_X32-NEXT:    kmovw %edx, %k1
-; KNL_X32-NEXT:    testb $1, {{[0-9]+}}(%esp)
-; KNL_X32-NEXT:    cmovnel %ecx, %eax
-; KNL_X32-NEXT:    kmovw {{[-0-9]+}}(%e{{[sb]}}p), %k2 ## 2-byte Reload
-; KNL_X32-NEXT:    kandw %k2, %k0, %k0
-; KNL_X32-NEXT:    kmovw %eax, %k2
-; KNL_X32-NEXT:    kandw %k1, %k2, %k1
+; KNL_X32-NEXT:    kmovw {{[-0-9]+}}(%e{{[sb]}}p), %k1 ## 2-byte Reload
+; KNL_X32-NEXT:    kandw %k1, %k0, %k0
 ; KNL_X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; KNL_X32-NEXT:    kmovw %k1, %ebx
+; KNL_X32-NEXT:    movb {{[0-9]+}}(%esp), %dl
 ; KNL_X32-NEXT:    kshiftrw $1, %k0, %k1
-; KNL_X32-NEXT:    kmovw %k1, %esi
-; KNL_X32-NEXT:    kshiftrw $2, %k0, %k1
 ; KNL_X32-NEXT:    kmovw %k1, %edi
+; KNL_X32-NEXT:    kshiftrw $2, %k0, %k1
+; KNL_X32-NEXT:    kmovw %k1, %ebx
 ; KNL_X32-NEXT:    kshiftrw $3, %k0, %k1
 ; KNL_X32-NEXT:    kmovw %k1, %ebp
 ; KNL_X32-NEXT:    kshiftrw $4, %k0, %k1
-; KNL_X32-NEXT:    kmovw %k1, %edx
+; KNL_X32-NEXT:    kmovw %k1, %esi
 ; KNL_X32-NEXT:    kshiftrw $5, %k0, %k1
 ; KNL_X32-NEXT:    kmovw %k1, %ecx
 ; KNL_X32-NEXT:    kshiftrw $6, %k0, %k1
-; KNL_X32-NEXT:    andl $1, %ebx
-; KNL_X32-NEXT:    movb %bl, 2(%eax)
-; KNL_X32-NEXT:    kmovw %k0, %ebx
-; KNL_X32-NEXT:    andl $1, %ebx
-; KNL_X32-NEXT:    andl $1, %esi
-; KNL_X32-NEXT:    leal (%ebx,%esi,2), %esi
-; KNL_X32-NEXT:    kmovw %k1, %ebx
-; KNL_X32-NEXT:    kshiftrw $7, %k0, %k1
+; KNL_X32-NEXT:    andb {{[0-9]+}}(%esp), %dl
+; KNL_X32-NEXT:    movzbl %dl, %edx
+; KNL_X32-NEXT:    andl $1, %edx
+; KNL_X32-NEXT:    movb %dl, 2(%eax)
+; KNL_X32-NEXT:    kmovw %k0, %edx
+; KNL_X32-NEXT:    andl $1, %edx
 ; KNL_X32-NEXT:    andl $1, %edi
-; KNL_X32-NEXT:    leal (%esi,%edi,4), %esi
+; KNL_X32-NEXT:    leal (%edx,%edi,2), %edx
 ; KNL_X32-NEXT:    kmovw %k1, %edi
+; KNL_X32-NEXT:    kshiftrw $7, %k0, %k1
+; KNL_X32-NEXT:    andl $1, %ebx
+; KNL_X32-NEXT:    leal (%edx,%ebx,4), %edx
+; KNL_X32-NEXT:    kmovw %k1, %ebx
 ; KNL_X32-NEXT:    kshiftrw $8, %k0, %k1
 ; KNL_X32-NEXT:    andl $1, %ebp
-; KNL_X32-NEXT:    leal (%esi,%ebp,8), %esi
+; KNL_X32-NEXT:    leal (%edx,%ebp,8), %edx
 ; KNL_X32-NEXT:    kmovw %k1, %ebp
 ; KNL_X32-NEXT:    kshiftrw $9, %k0, %k1
-; KNL_X32-NEXT:    andl $1, %edx
-; KNL_X32-NEXT:    shll $4, %edx
-; KNL_X32-NEXT:    orl %esi, %edx
-; KNL_X32-NEXT:    kmovw %k1, %esi
+; KNL_X32-NEXT:    andl $1, %esi
+; KNL_X32-NEXT:    shll $4, %esi
+; KNL_X32-NEXT:    orl %edx, %esi
+; KNL_X32-NEXT:    kmovw %k1, %edx
 ; KNL_X32-NEXT:    kshiftrw $10, %k0, %k1
 ; KNL_X32-NEXT:    andl $1, %ecx
 ; KNL_X32-NEXT:    shll $5, %ecx
-; KNL_X32-NEXT:    orl %edx, %ecx
-; KNL_X32-NEXT:    kmovw %k1, %edx
+; KNL_X32-NEXT:    orl %esi, %ecx
+; KNL_X32-NEXT:    kmovw %k1, %esi
 ; KNL_X32-NEXT:    kshiftrw $11, %k0, %k1
-; KNL_X32-NEXT:    andl $1, %ebx
-; KNL_X32-NEXT:    shll $6, %ebx
 ; KNL_X32-NEXT:    andl $1, %edi
-; KNL_X32-NEXT:    shll $7, %edi
-; KNL_X32-NEXT:    orl %ebx, %edi
-; KNL_X32-NEXT:    kmovw %k1, %ebx
+; KNL_X32-NEXT:    shll $6, %edi
+; KNL_X32-NEXT:    andl $1, %ebx
+; KNL_X32-NEXT:    shll $7, %ebx
+; KNL_X32-NEXT:    orl %edi, %ebx
+; KNL_X32-NEXT:    kmovw %k1, %edi
 ; KNL_X32-NEXT:    kshiftrw $12, %k0, %k1
 ; KNL_X32-NEXT:    andl $1, %ebp
 ; KNL_X32-NEXT:    shll $8, %ebp
-; KNL_X32-NEXT:    orl %edi, %ebp
-; KNL_X32-NEXT:    kmovw %k1, %edi
+; KNL_X32-NEXT:    orl %ebx, %ebp
+; KNL_X32-NEXT:    kmovw %k1, %ebx
 ; KNL_X32-NEXT:    kshiftrw $13, %k0, %k1
-; KNL_X32-NEXT:    andl $1, %esi
-; KNL_X32-NEXT:    shll $9, %esi
-; KNL_X32-NEXT:    orl %ebp, %esi
+; KNL_X32-NEXT:    andl $1, %edx
+; KNL_X32-NEXT:    shll $9, %edx
+; KNL_X32-NEXT:    orl %ebp, %edx
 ; KNL_X32-NEXT:    kmovw %k1, %ebp
 ; KNL_X32-NEXT:    kshiftrw $14, %k0, %k1
-; KNL_X32-NEXT:    andl $1, %edx
-; KNL_X32-NEXT:    shll $10, %edx
-; KNL_X32-NEXT:    orl %esi, %edx
-; KNL_X32-NEXT:    kmovw %k1, %esi
+; KNL_X32-NEXT:    andl $1, %esi
+; KNL_X32-NEXT:    shll $10, %esi
+; KNL_X32-NEXT:    orl %edx, %esi
+; KNL_X32-NEXT:    kmovw %k1, %edx
 ; KNL_X32-NEXT:    kshiftrw $15, %k0, %k0
-; KNL_X32-NEXT:    orl %ecx, %edx
+; KNL_X32-NEXT:    orl %ecx, %esi
 ; KNL_X32-NEXT:    kmovw %k0, %ecx
-; KNL_X32-NEXT:    andl $1, %ebx
-; KNL_X32-NEXT:    shll $11, %ebx
 ; KNL_X32-NEXT:    andl $1, %edi
-; KNL_X32-NEXT:    shll $12, %edi
-; KNL_X32-NEXT:    orl %ebx, %edi
+; KNL_X32-NEXT:    shll $11, %edi
+; KNL_X32-NEXT:    andl $1, %ebx
+; KNL_X32-NEXT:    shll $12, %ebx
+; KNL_X32-NEXT:    orl %edi, %ebx
 ; KNL_X32-NEXT:    andl $1, %ebp
 ; KNL_X32-NEXT:    shll $13, %ebp
-; KNL_X32-NEXT:    orl %edi, %ebp
-; KNL_X32-NEXT:    andl $1, %esi
-; KNL_X32-NEXT:    shll $14, %esi
-; KNL_X32-NEXT:    orl %ebp, %esi
+; KNL_X32-NEXT:    orl %ebx, %ebp
+; KNL_X32-NEXT:    andl $1, %edx
+; KNL_X32-NEXT:    shll $14, %edx
+; KNL_X32-NEXT:    orl %ebp, %edx
 ; KNL_X32-NEXT:    andl $1, %ecx
 ; KNL_X32-NEXT:    shll $15, %ecx
-; KNL_X32-NEXT:    orl %esi, %ecx
 ; KNL_X32-NEXT:    orl %edx, %ecx
+; KNL_X32-NEXT:    orl %esi, %ecx
 ; KNL_X32-NEXT:    movw %cx, (%eax)
 ; KNL_X32-NEXT:    addl $20, %esp
 ; KNL_X32-NEXT:    popl %esi