[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits

[RKSimon]

Summary

For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types.

This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type.

[efriedma]

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

2233

Not sure why you're checking hasBitPreservingFPLogic; from its documentation, it specifically applies to the behavior of FABS/FNEG. BITCAST should always preserve all bits.

[RKSimon]

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

2233

@spatel might be able to give more details but I understood that hasBitPreservingFPLogic's scope was to cover all bit operations on floating point data? The FABS/FNEG are just explicitly mentioned as they're the typical uses of signbit masking/XORing.

I'm happy to simplify this if you think it safe.

[spatel]

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

2233

FABS/FNEG were the motivation for the hook (D19391), but it could apply to something other than that, so I should improve that comment.

I don't know anything about the targets that inspired the hook, so I'd feel safer keeping the check. (cc @escha @scanon
).

Bitcast works "as if the value was stored to memory with one type and loaded from the same address with the other type." That doesn't imply that bits must be preserved? Eg, I could imagine something like any NAN bit pattern getting set to a fixed NAN value by the hardware.

[efriedma]

In IR, we generally assume a store+load preserves the exact bit-pattern (for example, GVN depends on this). I guess that could be wrong if we lower a "load" instruction to an x87 "fld" (which transforms SNaN->QNaN), but as as far as I know no other architecture has that particular quirk.

The hasBitPreservingFPLogic hook is supposed to detect cases where fneg is lowered to a floating-point subtraction; it has nothing to do with any property of the underlying registers.

[RKSimon]

Drop hasBitPreservingFPLogic

[efriedma]

LGTM

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

void SelectionDAG::computeKnownBits(SDValue Op, KnownBits &Known,
   Known = KnownBits(BitWidth);   // Don't know anything.
 
   if (auto *C = dyn_cast<ConstantSDNode>(Op)) {
     // We know all of the bits for a constant!
     Known.One = C->getAPIntValue();
     Known.Zero = ~Known.One;
     return;
   }
+  if (auto *C = dyn_cast<ConstantFPSDNode>(Op)) {
+    // We know all of the bits for a constant fp!
+    Known.One = C->getValueAPF().bitcastToAPInt();
+    Known.Zero = ~Known.One;
+    return;
+  }
 
   if (Depth == 6)
     return;  // Limit search depth.
 
   KnownBits Known2;
   unsigned NumElts = DemandedElts.getBitWidth();
 
   if (!DemandedElts)
if (SubIdx && SubIdx->getAPIntValue().ule(NumSrcElts - NumElts)) {
       computeKnownBits(Src, Known, DemandedSrc, Depth + 1);
     } else {
       computeKnownBits(Src, Known, Depth + 1);
     }
     break;
   }
   case ISD::BITCAST: {
     SDValue N0 = Op.getOperand(0);
-    unsigned SubBitWidth = N0.getScalarValueSizeInBits();
+    EVT SubVT = N0.getValueType();
+    unsigned SubBitWidth = SubVT.getScalarSizeInBits();
 
-    // Ignore bitcasts from floating point.
-    if (!N0.getValueType().isInteger())
+    // Ignore bitcasts from unsupported types.
+    if (!(SubVT.isInteger() || SubVT.isFloatingPoint()))
       break;
 
     // Fast handling of 'identity' bitcasts.
     if (BitWidth == SubBitWidth) {
       computeKnownBits(N0, Known, DemandedElts, Depth + 1);
       break;
     }
 
     // Support big-endian targets when it becomes useful.
APInt DemandedElts = VT.isVector()
                            ? APInt::getAllOnesValue(VT.getVectorNumElements())
                            : APInt(1, 1);
   return ComputeNumSignBits(Op, DemandedElts, Depth);
 }
 
 unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
                                           unsigned Depth) const {
   EVT VT = Op.getValueType();
-  assert(VT.isInteger() && "Invalid VT!");
+  assert((VT.isInteger() || VT.isFloatingPoint()) && "Invalid VT!");
   unsigned VTBits = VT.getScalarSizeInBits();
   unsigned NumElts = DemandedElts.getBitWidth();
   unsigned Tmp, Tmp2;
   unsigned FirstAnswer = 1;
 
   if (auto *C = dyn_cast<ConstantSDNode>(Op)) {
     const APInt &Val = C->getAPIntValue();
     return Val.getNumSignBits();
case ISD::VECTOR_SHUFFLE: {
     if (Tmp == 1)
       break;
     assert(Tmp <= VTBits && "Failed to determine minimum sign bits");
     return Tmp;
   }
 
   case ISD::BITCAST: {
     SDValue N0 = Op.getOperand(0);
-    unsigned SrcBits = N0.getScalarValueSizeInBits();
+    EVT SrcVT = N0.getValueType();
+    unsigned SrcBits = SrcVT.getScalarSizeInBits();
 
-    // Ignore bitcasts from floating point.
-    if (!N0.getValueType().isInteger())
+    // Ignore bitcasts from unsupported types..
+    if (!(SrcVT.isInteger() || SrcVT.isFloatingPoint()))
       break;
 
     // Fast handling of 'identity' bitcasts.
     if (VTBits == SrcBits)
       return ComputeNumSignBits(N0, DemandedElts, Depth + 1);
 
     // Bitcast 'large element' scalar/vector to 'small element' vector.
     // TODO: Handle cases other than 'sign splat' when we have a use case.

test/CodeGen/X86/avx512-vec-cmp.ll

; SKX-NEXT:    retq
   ret <2 x i64> %1
 }
 
 define <2 x i64> @test46(<2 x float> %x, <2 x float> %y) #0 {
 ; KNL-LABEL: test46:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxdq %xmm0, %xmm0
-; KNL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; KNL-NEXT:    vpsrlq $63, %xmm0, %xmm0
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test46:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
 ; SKX-NEXT:    vmovdqa64 {{.*}}(%rip), %xmm0 {%k1} {z}
 ; SKX-NEXT:    retq
   %mask = fcmp oeq <2 x float> %x, %y
   %1 = zext <2 x i1> %mask to <2 x i64>
   ret <2 x i64> %1
 }

test/CodeGen/X86/bitcast-and-setcc-256.ll

; AVX512-NEXT:    ret{{[l|q]}}
   %x0 = icmp sgt <8 x i32> %a, %b
   %x1 = icmp sgt <8 x i32> %c, %d
   %y = and <8 x i1> %x0, %x1
   %res = bitcast <8 x i1> %y to i8
   ret i8 %res
 }
 
 define i8 @v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x float> %d) {
-; SSE2-LABEL: v8f32:
-; SSE2:       # BB#0:
-; SSE2-NEXT:    cmpltps %xmm1, %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    cmpltps %xmm0, %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE2-NEXT:    cmpltps %xmm5, %xmm7
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm7[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    cmpltps %xmm4, %xmm6
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm6[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
-; SSE2-NEXT:    pand %xmm0, %xmm2
-; SSE2-NEXT:    packsswb %xmm0, %xmm2
-; SSE2-NEXT:    pmovmskb %xmm2, %eax
-; SSE2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
-; SSE2-NEXT:    ret{{[l|q]}}
-;
-; SSSE3-LABEL: v8f32:
-; SSSE3:       # BB#0:
-; SSSE3-NEXT:    cmpltps %xmm1, %xmm3
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    cmpltps %xmm0, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm2
-; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSSE3-NEXT:    cmpltps %xmm5, %xmm7
-; SSSE3-NEXT:    pshufb %xmm1, %xmm7
-; SSSE3-NEXT:    cmpltps %xmm4, %xmm6
-; SSSE3-NEXT:    pshufb %xmm1, %xmm6
-; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm7[0]
-; SSSE3-NEXT:    pand %xmm2, %xmm6
-; SSSE3-NEXT:    packsswb %xmm0, %xmm6
-; SSSE3-NEXT:    pmovmskb %xmm6, %eax
-; SSSE3-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
-; SSSE3-NEXT:    ret{{[l|q]}}
+; SSE2-SSSE3-LABEL: v8f32:
+; SSE2-SSSE3:       # BB#0:
+; SSE2-SSSE3-NEXT:    cmpltps %xmm1, %xmm3
+; SSE2-SSSE3-NEXT:    cmpltps %xmm0, %xmm2
+; SSE2-SSSE3-NEXT:    packssdw %xmm3, %xmm2
+; SSE2-SSSE3-NEXT:    cmpltps %xmm5, %xmm7
+; SSE2-SSSE3-NEXT:    cmpltps %xmm4, %xmm6
+; SSE2-SSSE3-NEXT:    packssdw %xmm7, %xmm6
+; SSE2-SSSE3-NEXT:    pand %xmm2, %xmm6
+; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm6
+; SSE2-SSSE3-NEXT:    pmovmskb %xmm6, %eax
+; SSE2-SSSE3-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; SSE2-SSSE3-NEXT:    ret{{[l|q]}}
 ;
 ; AVX12-LABEL: v8f32:
 ; AVX12:       # BB#0:
 ; AVX12-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
 ; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX12-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX12-NEXT:    vcmpltps %ymm2, %ymm3, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm2

test/CodeGen/X86/bitcast-and-setcc-512.ll

; AVX512BW-NEXT:    ret{{[l|q]}}
   %y = and <8 x i1> %x0, %x1
   %res = bitcast <8 x i1> %y to i8
   ret i8 %res
 }
 
 define i8 @v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double> %d) {
 ; SSE-LABEL: v8f64:
 ; SSE:       # BB#0:
+; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm8
 ; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm8
 ; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm11
-; SSE-NEXT:    cmpltpd %xmm3, %xmm7
-; SSE-NEXT:    cmpltpd %xmm2, %xmm6
-; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2],xmm7[0,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,4,5,4,5,6,7,0,1,4,5,8,9,12,13]
-; SSE-NEXT:    pshufb %xmm2, %xmm6
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm5
 ; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm5[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm4
 ; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm6[4,5,6,7]
+; SSE-NEXT:    cmpltpd %xmm3, %xmm7
+; SSE-NEXT:    cmpltpd %xmm2, %xmm6
+; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2],xmm7[0,2]
+; SSE-NEXT:    packssdw %xmm6, %xmm6
+; SSE-NEXT:    pblendw {{.*#+}} xmm6 = xmm0[0,1,2,3],xmm6[4,5,6,7]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm11
-; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2],xmm11[0,2]
-; SSE-NEXT:    pshufb %xmm2, %xmm8
+; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm11[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,2,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm10[0,2,2,3,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm9
-; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm8[4,5,6,7]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    psllw $15, %xmm2
-; SSE-NEXT:    psraw $15, %xmm2
-; SSE-NEXT:    packsswb %xmm0, %xmm2
-; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm8
+; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2],xmm9[0,2]
+; SSE-NEXT:    packssdw %xmm8, %xmm8
+; SSE-NEXT:    pblendw {{.*#+}} xmm8 = xmm1[0,1,2,3],xmm8[4,5,6,7]
+; SSE-NEXT:    pand %xmm6, %xmm8
+; SSE-NEXT:    psllw $15, %xmm8
+; SSE-NEXT:    psraw $15, %xmm8
+; SSE-NEXT:    packsswb %xmm0, %xmm8
+; SSE-NEXT:    pmovmskb %xmm8, %eax
 ; SSE-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE-NEXT:    ret{{[l|q]}}
 ;
 ; AVX12-LABEL: v8f64:
 ; AVX12:       # BB#0:
 ; AVX12-NEXT:    vcmpltpd %ymm1, %ymm3, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX12-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
; AVX512BW-NEXT:    ret{{[l|q]}}
   %res = bitcast <16 x i1> %y to i16
   ret i16 %res
 }
 
 define i16 @v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x float> %d) {
 ; SSE-LABEL: v16f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm9
+; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    cmpltps %xmm3, %xmm7
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE-NEXT:    pshufb %xmm3, %xmm7
 ; SSE-NEXT:    cmpltps %xmm2, %xmm6
-; SSE-NEXT:    pshufb %xmm3, %xmm6
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm7[0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; SSE-NEXT:    pshufb %xmm2, %xmm6
+; SSE-NEXT:    packssdw %xmm7, %xmm6
 ; SSE-NEXT:    cmpltps %xmm1, %xmm5
-; SSE-NEXT:    pshufb %xmm3, %xmm5
 ; SSE-NEXT:    cmpltps %xmm0, %xmm4
-; SSE-NEXT:    pshufb %xmm3, %xmm4
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm5[0]
-; SSE-NEXT:    pshufb %xmm2, %xmm4
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm6[0]
+; SSE-NEXT:    packssdw %xmm5, %xmm4
+; SSE-NEXT:    packsswb %xmm6, %xmm4
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm11
-; SSE-NEXT:    pshufb %xmm3, %xmm11
-; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm9
-; SSE-NEXT:    pshufb %xmm3, %xmm9
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm11[0]
-; SSE-NEXT:    pshufb %xmm2, %xmm9
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    pshufb %xmm3, %xmm10
+; SSE-NEXT:    packssdw %xmm11, %xmm10
+; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    pshufb %xmm3, %xmm8
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm8 = xmm8[0],xmm10[0]
-; SSE-NEXT:    pshufb %xmm2, %xmm8
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm8 = xmm8[0],xmm9[0]
+; SSE-NEXT:    packssdw %xmm9, %xmm8
+; SSE-NEXT:    packsswb %xmm10, %xmm8
 ; SSE-NEXT:    pand %xmm4, %xmm8
 ; SSE-NEXT:    pmovmskb %xmm8, %eax
 ; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; SSE-NEXT:    ret{{[l|q]}}
 ;
 ; AVX12-LABEL: v16f32:
 ; AVX12:       # BB#0:
 ; AVX12-NEXT:    vcmpltps %ymm1, %ymm3, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX12-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; AVX12-NEXT:    vmovdqa {{.*#+}} xmm3 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; AVX12-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX12-NEXT:    vcmpltps %ymm0, %ymm2, %ymm0
 ; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX12-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX12-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX12-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX12-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX12-NEXT:    vcmpltps %ymm5, %ymm7, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX12-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX12-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX12-NEXT:    vcmpltps %ymm4, %ymm6, %ymm2
-; AVX12-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX12-NEXT:    vpackssdw %xmm4, %xmm2, %xmm2
-; AVX12-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX12-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX12-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX12-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; AVX12-NEXT:    vpacksswb %xmm1, %xmm2, %xmm1
 ; AVX12-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX12-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX12-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX12-NEXT:    vzeroupper
 ; AVX12-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512F-LABEL: v16f32:
 ; AVX512F:       # BB#0:

test/CodeGen/X86/cast-vsel.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
 
 ; If we have a cmp and a sel with different-sized operands followed by a size-changing cast,
 ; we may want to pull the cast ahead of the select operands to create a select with matching op sizes:
 ; ext (sel (cmp a, b), c, d) --> sel (cmp a, b), (ext c), (ext d)
 
 define <8 x i32> @sext(<8 x float> %a, <8 x float> %b, <8 x i16> %c, <8 x i16> %d) {
 ; SSE2-LABEL: sext:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    cmpltps %xmm3, %xmm1
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; SSE2-NEXT:    cmpltps %xmm2, %xmm0
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
-; SSE2-NEXT:    pand %xmm2, %xmm4
-; SSE2-NEXT:    pandn %xmm5, %xmm2
-; SSE2-NEXT:    por %xmm4, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT:    packssdw %xmm1, %xmm0
+; SSE2-NEXT:    pand %xmm0, %xmm4
+; SSE2-NEXT:    pandn %xmm5, %xmm0
+; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE2-NEXT:    psrad $16, %xmm2
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    psrad $16, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: sext:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    cmpltps %xmm3, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    pshufb %xmm3, %xmm1
 ; SSE41-NEXT:    cmpltps %xmm2, %xmm0
-; SSE41-NEXT:    pshufb %xmm3, %xmm0
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE41-NEXT:    pand %xmm0, %xmm4
-; SSE41-NEXT:    pandn %xmm5, %xmm0
-; SSE41-NEXT:    por %xmm4, %xmm0
-; SSE41-NEXT:    pmovsxwd %xmm0, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxwd %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    packssdw %xmm1, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm5
+; SSE41-NEXT:    pmovsxwd %xmm5, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[2,3,0,1]
+; SSE41-NEXT:    pmovsxwd %xmm1, %xmm1
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: sext:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vcmpltps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpmovsxwd %xmm2, %xmm1
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxwd %xmm2, %xmm2
; AVX2-NEXT:    retq
   %sel = select <8 x i1> %cmp, <8 x i16> %c, <8 x i16> %d
   %ext = sext <8 x i16> %sel to <8 x i32>
   ret <8 x i32> %ext
 }
 
 define <8 x i32> @zext(<8 x float> %a, <8 x float> %b, <8 x i16> %c, <8 x i16> %d) {
 ; SSE2-LABEL: zext:
 ; SSE2:       # BB#0:
+; SSE2-NEXT:    movaps %xmm0, %xmm6
 ; SSE2-NEXT:    cmpltps %xmm3, %xmm1
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,2,2,3]
-; SSE2-NEXT:    cmpltps %xmm2, %xmm0
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; SSE2-NEXT:    pand %xmm1, %xmm4
-; SSE2-NEXT:    pandn %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    xorps %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT:    cmpltps %xmm2, %xmm6
+; SSE2-NEXT:    packssdw %xmm1, %xmm6
+; SSE2-NEXT:    pand %xmm6, %xmm4
+; SSE2-NEXT:    pandn %xmm5, %xmm6
+; SSE2-NEXT:    por %xmm4, %xmm6
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm6, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm6 = xmm6[4],xmm1[4],xmm6[5],xmm1[5],xmm6[6],xmm1[6],xmm6[7],xmm1[7]
+; SSE2-NEXT:    movdqa %xmm6, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: zext:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    cmpltps %xmm3, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    pshufb %xmm3, %xmm1
 ; SSE41-NEXT:    cmpltps %xmm2, %xmm0
-; SSE41-NEXT:    pshufb %xmm3, %xmm0
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE41-NEXT:    pand %xmm0, %xmm4
-; SSE41-NEXT:    pandn %xmm5, %xmm0
-; SSE41-NEXT:    por %xmm4, %xmm0
-; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    packssdw %xmm1, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm5
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[2,3,0,1]
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: zext:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vcmpltps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero

 ; SSE2-NEXT:    movq $-4096, %rax # imm = 0xF000
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1]
 ; SSE2-NEXT:    .p2align 4, 0x90
 ; SSE2-NEXT:  .LBB5_1: # %vector.body
 ; SSE2-NEXT:    # =>This Inner Loop Header: Depth=1
 ; SSE2-NEXT:    movaps da+4096(%rax), %xmm1
 ; SSE2-NEXT:    movaps da+4112(%rax), %xmm2
 ; SSE2-NEXT:    cmpltps db+4112(%rax), %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
 ; SSE2-NEXT:    cmpltps db+4096(%rax), %xmm1
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE2-NEXT:    packssdw %xmm2, %xmm1
 ; SSE2-NEXT:    movaps dc+4096(%rax), %xmm2
 ; SSE2-NEXT:    movaps dc+4112(%rax), %xmm3
 ; SSE2-NEXT:    cmpltps dd+4112(%rax), %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
 ; SSE2-NEXT:    cmpltps dd+4096(%rax), %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE2-NEXT:    packssdw %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
 ; SSE2-NEXT:    pand %xmm0, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, dj+4112(%rax)
 ; SSE2-NEXT:    movdqa %xmm1, dj+4096(%rax)
 ; SSE2-NEXT:    addq $32, %rax
 ; SSE2-NEXT:    jne .LBB5_1
 ; SSE2-NEXT:  # BB#2: # %for.end
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: example25:
 ; SSE41:       # BB#0: # %vector.ph
 ; SSE41-NEXT:    movq $-4096, %rax # imm = 0xF000
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [1,1,1,1]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1]
 ; SSE41-NEXT:    .p2align 4, 0x90
 ; SSE41-NEXT:  .LBB5_1: # %vector.body
 ; SSE41-NEXT:    # =>This Inner Loop Header: Depth=1
-; SSE41-NEXT:    movaps da+4096(%rax), %xmm2
-; SSE41-NEXT:    movaps da+4112(%rax), %xmm3
-; SSE41-NEXT:    cmpltps db+4112(%rax), %xmm3
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    cmpltps db+4096(%rax), %xmm2
-; SSE41-NEXT:    pshufb %xmm0, %xmm2
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSE41-NEXT:    movaps dc+4096(%rax), %xmm3
-; SSE41-NEXT:    movaps dc+4112(%rax), %xmm4
-; SSE41-NEXT:    cmpltps dd+4112(%rax), %xmm4
-; SSE41-NEXT:    pshufb %xmm0, %xmm4
-; SSE41-NEXT:    cmpltps dd+4096(%rax), %xmm3
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; SSE41-NEXT:    movaps da+4096(%rax), %xmm1
+; SSE41-NEXT:    movaps da+4112(%rax), %xmm2
+; SSE41-NEXT:    cmpltps db+4112(%rax), %xmm2
+; SSE41-NEXT:    cmpltps db+4096(%rax), %xmm1
+; SSE41-NEXT:    packssdw %xmm2, %xmm1
+; SSE41-NEXT:    movaps dc+4096(%rax), %xmm2
+; SSE41-NEXT:    movaps dc+4112(%rax), %xmm3
+; SSE41-NEXT:    cmpltps dd+4112(%rax), %xmm3
+; SSE41-NEXT:    cmpltps dd+4096(%rax), %xmm2
+; SSE41-NEXT:    packssdw %xmm3, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE41-NEXT:    pand %xmm1, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, dj+4112(%rax)
-; SSE41-NEXT:    movdqa %xmm2, dj+4096(%rax)
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; SSE41-NEXT:    pand %xmm0, %xmm1
+; SSE41-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSE41-NEXT:    pand %xmm0, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, dj+4112(%rax)
+; SSE41-NEXT:    movdqa %xmm1, dj+4096(%rax)
 ; SSE41-NEXT:    addq $32, %rax
 ; SSE41-NEXT:    jne .LBB5_1
 ; SSE41-NEXT:  # BB#2: # %for.end
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: example25:
 ; AVX1:       # BB#0: # %vector.ph
 ; AVX1-NEXT:    movq $-4096, %rax # imm = 0xF000

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
 ; SSE2-NEXT:    movq $-4096, %rax # imm = 0xF000
 ; SSE2-NEXT:    .p2align 4, 0x90
 ; SSE2-NEXT:  .LBB6_1: # %vector.body
 ; SSE2-NEXT:    # =>This Inner Loop Header: Depth=1
 ; SSE2-NEXT:    movaps da+4096(%rax), %xmm2
 ; SSE2-NEXT:    movaps da+4112(%rax), %xmm3
 ; SSE2-NEXT:    cmpltps db+4112(%rax), %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
 ; SSE2-NEXT:    cmpltps db+4096(%rax), %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE2-NEXT:    packssdw %xmm3, %xmm2
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pand %xmm2, %xmm3
 ; SSE2-NEXT:    pandn %xmm1, %xmm2
 ; SSE2-NEXT:    por %xmm3, %xmm2
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
 ; SSE2-NEXT:    psrad $16, %xmm3
 ; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    psrad $16, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, dj+4112(%rax)
 ; SSE2-NEXT:    movdqa %xmm3, dj+4096(%rax)
 ; SSE2-NEXT:    addq $32, %rax
 ; SSE2-NEXT:    jne .LBB6_1
 ; SSE2-NEXT:  # BB#2: # %for.end
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: example24:
 ; SSE41:       # BB#0: # %vector.ph
 ; SSE41-NEXT:    movd %edi, %xmm0
 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
-; SSE41-NEXT:    movd %esi, %xmm1
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,0,1,1]
+; SSE41-NEXT:    movd %esi, %xmm0
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
 ; SSE41-NEXT:    movq $-4096, %rax # imm = 0xF000
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; SSE41-NEXT:    .p2align 4, 0x90
 ; SSE41-NEXT:  .LBB6_1: # %vector.body
 ; SSE41-NEXT:    # =>This Inner Loop Header: Depth=1
-; SSE41-NEXT:    movaps da+4096(%rax), %xmm3
-; SSE41-NEXT:    movaps da+4112(%rax), %xmm4
-; SSE41-NEXT:    cmpltps db+4112(%rax), %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    cmpltps db+4096(%rax), %xmm3
-; SSE41-NEXT:    pshufb %xmm2, %xmm3
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; SSE41-NEXT:    movdqa %xmm0, %xmm4
-; SSE41-NEXT:    pand %xmm3, %xmm4
-; SSE41-NEXT:    pandn %xmm1, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmovsxwd %xmm4, %xmm4
+; SSE41-NEXT:    movaps da+4096(%rax), %xmm0
+; SSE41-NEXT:    movaps da+4112(%rax), %xmm3
+; SSE41-NEXT:    cmpltps db+4112(%rax), %xmm3
+; SSE41-NEXT:    cmpltps db+4096(%rax), %xmm0
+; SSE41-NEXT:    packssdw %xmm3, %xmm0
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[2,3,0,1]
+; SSE41-NEXT:    pmovsxwd %xmm0, %xmm0
 ; SSE41-NEXT:    pmovsxwd %xmm3, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, dj+4096(%rax)
-; SSE41-NEXT:    movdqa %xmm4, dj+4112(%rax)
+; SSE41-NEXT:    movdqa %xmm0, dj+4112(%rax)
 ; SSE41-NEXT:    addq $32, %rax
 ; SSE41-NEXT:    jne .LBB6_1
 ; SSE41-NEXT:  # BB#2: # %for.end
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: example24:
 ; AVX1:       # BB#0: # %vector.ph
 ; AVX1-NEXT:    vmovd %edi, %xmm0

test/CodeGen/X86/combine-fcopysign.ll

 ; AVX-NEXT:    retq
   %1 = call <4 x float> @llvm.copysign.v4f32(<4 x float> %x, <4 x float> <float -0.0, float -2.0, float -4.0, float -8.0>)
   ret <4 x float> %1
 }
 
 define <4 x float> @combine_vec_fcopysign_fneg_fabs_sgn(<4 x float> %x, <4 x float> %y) {
 ; SSE-LABEL: combine_vec_fcopysign_fneg_fabs_sgn:
 ; SSE:       # BB#0:
-; SSE-NEXT:    movaps {{.*#+}} xmm2 = [-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00]
-; SSE-NEXT:    orps %xmm2, %xmm1
-; SSE-NEXT:    andps %xmm2, %xmm1
 ; SSE-NEXT:    andps {{.*}}(%rip), %xmm0
-; SSE-NEXT:    orps %xmm1, %xmm0
+; SSE-NEXT:    orps {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: combine_vec_fcopysign_fneg_fabs_sgn:
 ; AVX:       # BB#0:
+; AVX-NEXT:    vbroadcastss {{.*}}(%rip), %xmm1
 ; AVX-NEXT:    vbroadcastss {{.*}}(%rip), %xmm2
-; AVX-NEXT:    vorps %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vbroadcastss {{.*}}(%rip), %xmm3
-; AVX-NEXT:    vandps %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vandps %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vandps %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vorps %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %1 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %y)
   %2 = fsub <4 x float> <float -0.0, float -0.0, float -0.0, float -0.0>, %1
   %3 = call <4 x float> @llvm.copysign.v4f32(<4 x float> %x, <4 x float> %2)
   ret <4 x float> %3
 }
 

test/CodeGen/X86/machine-cp.ll

 
 ; Check that copy propagation does not kill thing like:
 ; dst = copy src <-- do not kill that.
 ; ... = op1 dst<undef>
 ; ... = op2 dst <-- this is used here.
 define <16 x float> @foo(<16 x float> %x) {
 ; CHECK-LABEL: foo:
 ; CHECK:       ## BB#0: ## %bb
-; CHECK-NEXT:    xorps %xmm4, %xmm4
+; CHECK-NEXT:    movaps %xmm3, %xmm8
+; CHECK-NEXT:    xorps %xmm3, %xmm3
 ; CHECK-NEXT:    pxor %xmm6, %xmm6
 ; CHECK-NEXT:    pcmpgtd %xmm0, %xmm6
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255]
 ; CHECK-NEXT:    pand %xmm6, %xmm5
+; CHECK-NEXT:    packuswb %xmm5, %xmm5
+; CHECK-NEXT:    packuswb %xmm5, %xmm5
 ; CHECK-NEXT:    cvttps2dq %xmm0, %xmm13
 ; CHECK-NEXT:    movdqa %xmm0, %xmm10
-; CHECK-NEXT:    cmpltps %xmm4, %xmm10
-; CHECK-NEXT:    movdqa %xmm6, %xmm8
-; CHECK-NEXT:    pxor %xmm10, %xmm8
+; CHECK-NEXT:    cmpltps %xmm3, %xmm10
+; CHECK-NEXT:    movdqa %xmm6, %xmm9
+; CHECK-NEXT:    pxor %xmm10, %xmm9
 ; CHECK-NEXT:    cvttps2dq %xmm1, %xmm14
 ; CHECK-NEXT:    movaps %xmm1, %xmm11
-; CHECK-NEXT:    cmpltps %xmm4, %xmm11
-; CHECK-NEXT:    movdqa %xmm6, %xmm9
-; CHECK-NEXT:    pxor %xmm11, %xmm9
-; CHECK-NEXT:    cvttps2dq %xmm2, %xmm1
-; CHECK-NEXT:    cmpltps %xmm4, %xmm2
+; CHECK-NEXT:    cmpltps %xmm3, %xmm11
 ; CHECK-NEXT:    movdqa %xmm6, %xmm7
-; CHECK-NEXT:    pxor %xmm2, %xmm7
-; CHECK-NEXT:    cvttps2dq %xmm3, %xmm12
-; CHECK-NEXT:    cmpltps %xmm4, %xmm3
-; CHECK-NEXT:    pxor %xmm3, %xmm6
+; CHECK-NEXT:    pxor %xmm11, %xmm7
+; CHECK-NEXT:    cvttps2dq %xmm2, %xmm1
+; CHECK-NEXT:    cmpltps %xmm3, %xmm2
+; CHECK-NEXT:    movdqa %xmm6, %xmm4
+; CHECK-NEXT:    pxor %xmm2, %xmm4
+; CHECK-NEXT:    cvttps2dq %xmm8, %xmm12
+; CHECK-NEXT:    cmpltps %xmm3, %xmm8
+; CHECK-NEXT:    pxor %xmm8, %xmm6
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1]
 ; CHECK-NEXT:    pand %xmm0, %xmm6
+; CHECK-NEXT:    pand %xmm0, %xmm4
 ; CHECK-NEXT:    pand %xmm0, %xmm7
 ; CHECK-NEXT:    pand %xmm0, %xmm9
-; CHECK-NEXT:    pand %xmm0, %xmm8
 ; CHECK-NEXT:    cvtdq2ps %xmm13, %xmm15
 ; CHECK-NEXT:    cvtdq2ps %xmm14, %xmm14
 ; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm13
 ; CHECK-NEXT:    cvtdq2ps %xmm12, %xmm12
 ; CHECK-NEXT:    pxor %xmm0, %xmm0
 ; CHECK-NEXT:    cmpltps %xmm12, %xmm0
-; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; CHECK-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; CHECK-NEXT:    xorps %xmm1, %xmm1
 ; CHECK-NEXT:    cmpltps %xmm13, %xmm1
-; CHECK-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; CHECK-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; CHECK-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; CHECK-NEXT:    pxor %xmm0, %xmm0
+; CHECK-NEXT:    packssdw %xmm0, %xmm1
+; CHECK-NEXT:    xorps %xmm0, %xmm0
 ; CHECK-NEXT:    cmpltps %xmm14, %xmm0
-; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; CHECK-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; CHECK-NEXT:    cmpltps %xmm15, %xmm4
-; CHECK-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; CHECK-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,6,7]
-; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm0[0]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,255,255,255,255]
-; CHECK-NEXT:    pand %xmm0, %xmm1
-; CHECK-NEXT:    pand %xmm0, %xmm4
-; CHECK-NEXT:    packuswb %xmm5, %xmm5
-; CHECK-NEXT:    packuswb %xmm5, %xmm5
-; CHECK-NEXT:    packuswb %xmm1, %xmm4
-; CHECK-NEXT:    pand %xmm5, %xmm4
-; CHECK-NEXT:    movdqa %xmm4, %xmm1
+; CHECK-NEXT:    cmpltps %xmm15, %xmm3
+; CHECK-NEXT:    packssdw %xmm0, %xmm3
+; CHECK-NEXT:    packsswb %xmm1, %xmm3
+; CHECK-NEXT:    pand %xmm5, %xmm3
+; CHECK-NEXT:    movdqa %xmm3, %xmm1
 ; CHECK-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
 ; CHECK-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
 ; CHECK-NEXT:    pslld $31, %xmm0
 ; CHECK-NEXT:    psrad $31, %xmm0
-; CHECK-NEXT:    pxor %xmm8, %xmm0
-; CHECK-NEXT:    pxor %xmm15, %xmm0
 ; CHECK-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; CHECK-NEXT:    pslld $31, %xmm1
 ; CHECK-NEXT:    psrad $31, %xmm1
-; CHECK-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm0[8],xmm4[9],xmm0[9],xmm4[10],xmm0[10],xmm4[11],xmm0[11],xmm4[12],xmm0[12],xmm4[13],xmm0[13],xmm4[14],xmm0[14],xmm4[15],xmm0[15]
-; CHECK-NEXT:    pxor %xmm9, %xmm1
-; CHECK-NEXT:    pxor %xmm14, %xmm1
-; CHECK-NEXT:    movdqa %xmm4, %xmm5
+; CHECK-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
+; CHECK-NEXT:    movdqa %xmm3, %xmm5
 ; CHECK-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
 ; CHECK-NEXT:    pslld $31, %xmm5
 ; CHECK-NEXT:    psrad $31, %xmm5
-; CHECK-NEXT:    pxor %xmm7, %xmm5
+; CHECK-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; CHECK-NEXT:    pslld $31, %xmm3
+; CHECK-NEXT:    psrad $31, %xmm3
+; CHECK-NEXT:    pxor %xmm9, %xmm0
+; CHECK-NEXT:    pxor %xmm15, %xmm0
+; CHECK-NEXT:    pxor %xmm7, %xmm1
+; CHECK-NEXT:    pxor %xmm14, %xmm1
+; CHECK-NEXT:    pxor %xmm4, %xmm5
 ; CHECK-NEXT:    pxor %xmm13, %xmm5
-; CHECK-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
-; CHECK-NEXT:    pslld $31, %xmm4
-; CHECK-NEXT:    psrad $31, %xmm4
-; CHECK-NEXT:    pxor %xmm6, %xmm4
-; CHECK-NEXT:    pxor %xmm12, %xmm4
-; CHECK-NEXT:    pand %xmm3, %xmm4
+; CHECK-NEXT:    pxor %xmm6, %xmm3
+; CHECK-NEXT:    pxor %xmm12, %xmm3
+; CHECK-NEXT:    pand %xmm8, %xmm3
 ; CHECK-NEXT:    pand %xmm2, %xmm5
 ; CHECK-NEXT:    pand %xmm11, %xmm1
 ; CHECK-NEXT:    pand %xmm10, %xmm0
-; CHECK-NEXT:    pxor %xmm8, %xmm0
-; CHECK-NEXT:    pxor %xmm9, %xmm1
-; CHECK-NEXT:    pxor %xmm7, %xmm5
-; CHECK-NEXT:    pxor %xmm6, %xmm4
+; CHECK-NEXT:    pxor %xmm9, %xmm0
+; CHECK-NEXT:    pxor %xmm7, %xmm1
+; CHECK-NEXT:    pxor %xmm4, %xmm5
+; CHECK-NEXT:    pxor %xmm6, %xmm3
 ; CHECK-NEXT:    movdqa %xmm5, %xmm2
-; CHECK-NEXT:    movdqa %xmm4, %xmm3
 ; CHECK-NEXT:    retq
 bb:
   %v3 = icmp slt <16 x i32> undef, zeroinitializer
   %v14 = zext <16 x i1> %v3 to <16 x i32>
   %v16 = fcmp olt <16 x float> %x, zeroinitializer
   %v17 = sext <16 x i1> %v16 to <16 x i32>
   %v18 = zext <16 x i1> %v16 to <16 x i32>
   %v19 = xor <16 x i32> %v14, %v18

test/CodeGen/X86/vector-compare-results.ll

 ; AVX512-NEXT:    retq
   %1 = fcmp ogt <16 x double> %a0, %a1
   ret <16 x i1> %1
 }
 
 define <32 x i1> @test_cmp_v32f32(<32 x float> %a0, <32 x float> %a1) nounwind {
 ; SSE2-LABEL: test_cmp_v32f32:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm8
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm9
-; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm10
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm11
+; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm10
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm12
+; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm8
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm13
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm14
 ; SSE2-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm15
 ; SSE2-NEXT:    cmpltps %xmm3, %xmm15
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm15[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
 ; SSE2-NEXT:    cmpltps %xmm2, %xmm14
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm14[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT:    pand %xmm3, %xmm2
+; SSE2-NEXT:    packssdw %xmm15, %xmm14
 ; SSE2-NEXT:    cmpltps %xmm1, %xmm13
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm13[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    cmpltps %xmm0, %xmm12
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm12[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    packuswb %xmm2, %xmm0
-; SSE2-NEXT:    cmpltps %xmm7, %xmm11
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm11[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    cmpltps %xmm0, %xmm8
+; SSE2-NEXT:    packssdw %xmm13, %xmm8
+; SSE2-NEXT:    packsswb %xmm14, %xmm8
+; SSE2-NEXT:    cmpltps %xmm7, %xmm12
 ; SSE2-NEXT:    cmpltps %xmm6, %xmm10
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm10[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    cmpltps %xmm5, %xmm9
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    cmpltps %xmm4, %xmm8
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm8[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm1[0]
-; SSE2-NEXT:    pand %xmm3, %xmm4
-; SSE2-NEXT:    packuswb %xmm2, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, -{{[0-9]+}}(%rsp)
+; SSE2-NEXT:    packssdw %xmm12, %xmm10
+; SSE2-NEXT:    cmpltps %xmm5, %xmm11
+; SSE2-NEXT:    cmpltps %xmm4, %xmm9
+; SSE2-NEXT:    packssdw %xmm11, %xmm9
+; SSE2-NEXT:    packsswb %xmm10, %xmm9
+; SSE2-NEXT:    movdqa %xmm9, -{{[0-9]+}}(%rsp)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al

 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
 ; SSE2-NEXT:    andb $1, %cl
 ; SSE2-NEXT:    movb %cl, 2(%rdi)
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
-; SSE2-NEXT:    movdqa %xmm0, -{{[0-9]+}}(%rsp)
+; SSE2-NEXT:    movdqa %xmm8, -{{[0-9]+}}(%rsp)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, (%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, (%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al

 ; SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
 ; SSE2-NEXT:    cmpltpd %xmm7, %xmm4
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
 ; SSE2-NEXT:    cmpltpd %xmm6, %xmm7
 ; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7]
 ; SSE2-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm7[0,1,0,2,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1]
-; SSE2-NEXT:    movsd {{.*#+}} xmm4 = xmm5[0],xmm4[1]
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm5
-; SSE2-NEXT:    cmpltpd %xmm1, %xmm5
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm6
-; SSE2-NEXT:    cmpltpd %xmm0, %xmm6
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm7[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm4[0],xmm6[1],xmm4[1]
+; SSE2-NEXT:    movsd {{.*#+}} xmm6 = xmm5[0],xmm6[1]
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE2-NEXT:    cmpltpd %xmm1, %xmm4
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm1
-; SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm6[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1]
+; SSE2-NEXT:    cmpltpd %xmm0, %xmm1
+; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1]
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm0
 ; SSE2-NEXT:    cmpltpd %xmm3, %xmm0
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
 ; SSE2-NEXT:    cmpltpd %xmm2, %xmm3
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm6
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm1
 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[0,1,0,2,4,5,6,7]
 ; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[0,1,0,2,4,5,6,7]
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm2
+; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm5[0],xmm0[1]
-; SSE2-NEXT:    movapd {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT:    andpd %xmm2, %xmm4
-; SSE2-NEXT:    andpd %xmm2, %xmm0
-; SSE2-NEXT:    packuswb %xmm4, %xmm0
+; SSE2-NEXT:    packsswb %xmm6, %xmm0
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
 ; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
 ; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
 ; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
 ; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
+; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm2
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
 ; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
 ; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
 ; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,1,0,2,4,5,6,7]
-; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm6
-; SSE2-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,1,0,2,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
-; SSE2-NEXT:    movsd {{.*#+}} xmm3 = xmm5[0],xmm3[1]
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm5
-; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm5
-; SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
 ; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
 ; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
-; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm5
-; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm5
-; SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; SSE2-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
+; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
+; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
 ; SSE2-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm1
 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
-; SSE2-NEXT:    andpd %xmm2, %xmm3
-; SSE2-NEXT:    andpd %xmm2, %xmm1
-; SSE2-NEXT:    packuswb %xmm3, %xmm1
+; SSE2-NEXT:    packsswb %xmm2, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, -{{[0-9]+}}(%rsp)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
 ; SSE2-NEXT:    andb $1, %al
 ; SSE2-NEXT:    movb %al, 2(%rdi)
 ; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al

 ; SSE42:       # BB#0:
 ; SSE42-NEXT:    pushq %rbp
 ; SSE42-NEXT:    pushq %r15
 ; SSE42-NEXT:    pushq %r14
 ; SSE42-NEXT:    pushq %r13
 ; SSE42-NEXT:    pushq %r12
 ; SSE42-NEXT:    pushq %rbx
 ; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm8
-; SSE42-NEXT:    cmpltpd %xmm0, %xmm8
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm0
-; SSE42-NEXT:    cmpltpd %xmm1, %xmm0
+; SSE42-NEXT:    cmpltpd %xmm7, %xmm8
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
+; SSE42-NEXT:    cmpltpd %xmm6, %xmm7
+; SSE42-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm6 = xmm8[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm6
+; SSE42-NEXT:    cmpltpd %xmm5, %xmm6
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm5
+; SSE42-NEXT:    cmpltpd %xmm4, %xmm5
+; SSE42-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm4 = xmm6[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
+; SSE42-NEXT:    pblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm7[4,5,6,7]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    cmpltpd %xmm3, %xmm4
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    cmpltpd %xmm2, %xmm3
+; SSE42-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm2
+; SSE42-NEXT:    cmpltpd %xmm1, %xmm2
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    cmpltpd %xmm0, %xmm4
 ; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm1
-; SSE42-NEXT:    cmpltpd %xmm2, %xmm1
+; SSE42-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm2
-; SSE42-NEXT:    cmpltpd %xmm3, %xmm2
+; SSE42-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
 ; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
-; SSE42-NEXT:    cmpltpd %xmm4, %xmm3
+; SSE42-NEXT:    packsswb %xmm5, %xmm0
 ; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
-; SSE42-NEXT:    cmpltpd %xmm5, %xmm4
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm5
-; SSE42-NEXT:    cmpltpd %xmm6, %xmm5
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm6
-; SSE42-NEXT:    cmpltpd %xmm7, %xmm6
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r8d
-; SSE42-NEXT:    pextrb $0, %xmm7, %r13d
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r14d
-; SSE42-NEXT:    pextrb $0, %xmm7, %ebx
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r15d
-; SSE42-NEXT:    pextrb $0, %xmm7, %ebp
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r12d
-; SSE42-NEXT:    pextrb $0, %xmm7, %ecx
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r11d
-; SSE42-NEXT:    pextrb $0, %xmm7, %edx
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %r9d
-; SSE42-NEXT:    pextrb $0, %xmm7, %esi
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    pextrb $8, %xmm7, %eax
-; SSE42-NEXT:    movl %eax, -{{[0-9]+}}(%rsp) # 4-byte Spill
-; SSE42-NEXT:    pextrb $0, %xmm7, %r10d
-; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm7
-; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm7
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm2
+; SSE42-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm4
+; SSE42-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; SSE42-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm3
+; SSE42-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm1
+; SSE42-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE42-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE42-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; SSE42-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm4[4,5,6,7]
+; SSE42-NEXT:    packsswb %xmm2, %xmm1
+; SSE42-NEXT:    pextrb $15, %xmm1, %eax
+; SSE42-NEXT:    andb $1, %al
+; SSE42-NEXT:    movb %al, 2(%rdi)
+; SSE42-NEXT:    pextrb $14, %xmm1, %eax
+; SSE42-NEXT:    andb $1, %al
+; SSE42-NEXT:    movb %al, 2(%rdi)
+; SSE42-NEXT:    pextrb $13, %xmm1, %r8d
+; SSE42-NEXT:    pextrb $12, %xmm1, %r9d
+; SSE42-NEXT:    pextrb $11, %xmm1, %r10d
+; SSE42-NEXT:    pextrb $10, %xmm1, %r11d
+; SSE42-NEXT:    pextrb $9, %xmm1, %r14d
+; SSE42-NEXT:    pextrb $8, %xmm1, %r15d
+; SSE42-NEXT:    pextrb $7, %xmm1, %r12d
+; SSE42-NEXT:    pextrb $6, %xmm1, %r13d
+; SSE42-NEXT:    pextrb $5, %xmm1, %ebx
+; SSE42-NEXT:    pextrb $4, %xmm1, %ebp
+; SSE42-NEXT:    pextrb $3, %xmm1, %eax
+; SSE42-NEXT:    pextrb $2, %xmm1, %ecx
+; SSE42-NEXT:    pextrb $1, %xmm1, %edx
+; SSE42-NEXT:    pextrb $0, %xmm1, %esi
 ; SSE42-NEXT:    andb $1, %r8b
 ; SSE42-NEXT:    movb %r8b, 2(%rdi)
-; SSE42-NEXT:    andb $1, %r13b
-; SSE42-NEXT:    movb %r13b, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm7, %eax
-; SSE42-NEXT:    movl %eax, -{{[0-9]+}}(%rsp) # 4-byte Spill
-; SSE42-NEXT:    pextrb $0, %xmm7, %r13d
+; SSE42-NEXT:    andb $1, %r9b
+; SSE42-NEXT:    movb %r9b, 2(%rdi)
+; SSE42-NEXT:    andb $1, %r10b
+; SSE42-NEXT:    movb %r10b, 2(%rdi)
+; SSE42-NEXT:    andb $1, %r11b
+; SSE42-NEXT:    movb %r11b, 2(%rdi)
 ; SSE42-NEXT:    andb $1, %r14b
 ; SSE42-NEXT:    movb %r14b, 2(%rdi)
-; SSE42-NEXT:    andb $1, %bl
-; SSE42-NEXT:    movb %bl, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm6, %r14d
-; SSE42-NEXT:    pextrb $0, %xmm6, %r8d
 ; SSE42-NEXT:    andb $1, %r15b
 ; SSE42-NEXT:    movb %r15b, 2(%rdi)
-; SSE42-NEXT:    andb $1, %bpl
-; SSE42-NEXT:    movb %bpl, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm5, %r15d
-; SSE42-NEXT:    pextrb $0, %xmm5, %ebp
 ; SSE42-NEXT:    andb $1, %r12b
 ; SSE42-NEXT:    movb %r12b, 2(%rdi)
+; SSE42-NEXT:    andb $1, %r13b
+; SSE42-NEXT:    movb %r13b, 2(%rdi)
+; SSE42-NEXT:    andb $1, %bl
+; SSE42-NEXT:    movb %bl, 2(%rdi)
+; SSE42-NEXT:    andb $1, %bpl
+; SSE42-NEXT:    movb %bpl, 2(%rdi)
+; SSE42-NEXT:    andb $1, %al
+; SSE42-NEXT:    movb %al, 2(%rdi)
 ; SSE42-NEXT:    andb $1, %cl
 ; SSE42-NEXT:    movb %cl, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm4, %r12d
-; SSE42-NEXT:    pextrb $0, %xmm4, %ebx
-; SSE42-NEXT:    andb $1, %r11b
-; SSE42-NEXT:    movb %r11b, 2(%rdi)
 ; SSE42-NEXT:    andb $1, %dl
 ; SSE42-NEXT:    movb %dl, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm3, %r11d
-; SSE42-NEXT:    pextrb $0, %xmm3, %ecx
-; SSE42-NEXT:    andb $1, %r9b
-; SSE42-NEXT:    movb %r9b, 2(%rdi)
 ; SSE42-NEXT:    andb $1, %sil
 ; SSE42-NEXT:    movb %sil, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm2, %r9d
-; SSE42-NEXT:    pextrb $0, %xmm2, %edx
-; SSE42-NEXT:    movl -{{[0-9]+}}(%rsp), %eax # 4-byte Reload
+; SSE42-NEXT:    pextrb $15, %xmm0, %eax
 ; SSE42-NEXT:    andb $1, %al
-; SSE42-NEXT:    movb %al, 2(%rdi)
-; SSE42-NEXT:    andb $1, %r10b
-; SSE42-NEXT:    movb %r10b, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm1, %r10d
-; SSE42-NEXT:    pextrb $0, %xmm1, %esi
-; SSE42-NEXT:    movl -{{[0-9]+}}(%rsp), %eax # 4-byte Reload
+; SSE42-NEXT:    movb %al, (%rdi)
+; SSE42-NEXT:    pextrb $14, %xmm0, %eax
 ; SSE42-NEXT:    andb $1, %al
-; SSE42-NEXT:    movb %al, 2(%rdi)
-; SSE42-NEXT:    andb $1, %r13b
-; SSE42-NEXT:    movb %r13b, 2(%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm0, %r13d
-; SSE42-NEXT:    pextrb $0, %xmm0, %eax
-; SSE42-NEXT:    andb $1, %r14b
-; SSE42-NEXT:    movb %r14b, (%rdi)
+; SSE42-NEXT:    movb %al, (%rdi)
+; SSE42-NEXT:    pextrb $13, %xmm0, %r8d
+; SSE42-NEXT:    pextrb $12, %xmm0, %r9d
+; SSE42-NEXT:    pextrb $11, %xmm0, %r10d
+; SSE42-NEXT:    pextrb $10, %xmm0, %r11d
+; SSE42-NEXT:    pextrb $9, %xmm0, %r14d
+; SSE42-NEXT:    pextrb $8, %xmm0, %r15d
+; SSE42-NEXT:    pextrb $7, %xmm0, %r12d
+; SSE42-NEXT:    pextrb $6, %xmm0, %r13d
+; SSE42-NEXT:    pextrb $5, %xmm0, %ebx
+; SSE42-NEXT:    pextrb $4, %xmm0, %ebp
+; SSE42-NEXT:    pextrb $3, %xmm0, %eax
+; SSE42-NEXT:    pextrb $2, %xmm0, %ecx
+; SSE42-NEXT:    pextrb $1, %xmm0, %edx
+; SSE42-NEXT:    pextrb $0, %xmm0, %esi
 ; SSE42-NEXT:    andb $1, %r8b
 ; SSE42-NEXT:    movb %r8b, (%rdi)
-; SSE42-NEXT:    pextrb $8, %xmm8, %r8d
-; SSE42-NEXT:    pextrb $0, %xmm8, %r14d
+; SSE42-NEXT:    andb $1, %r9b
+; SSE42-NEXT:    movb %r9b, (%rdi)
+; SSE42-NEXT:    andb $1, %r10b
+; SSE42-NEXT:    movb %r10b, (%rdi)
+; SSE42-NEXT:    andb $1, %r11b
+; SSE42-NEXT:    movb %r11b, (%rdi)
+; SSE42-NEXT:    andb $1, %r14b
+; SSE42-NEXT:    movb %r14b, (%rdi)
 ; SSE42-NEXT:    andb $1, %r15b
 ; SSE42-NEXT:    movb %r15b, (%rdi)
-; SSE42-NEXT:    andb $1, %bpl
-; SSE42-NEXT:    movb %bpl, (%rdi)
 ; SSE42-NEXT:    andb $1, %r12b
 ; SSE42-NEXT:    movb %r12b, (%rdi)
+; SSE42-NEXT:    andb $1, %r13b
+; SSE42-NEXT:    movb %r13b, (%rdi)
 ; SSE42-NEXT:    andb $1, %bl
 ; SSE42-NEXT:    movb %bl, (%rdi)
-; SSE42-NEXT:    andb $1, %r11b
-; SSE42-NEXT:    movb %r11b, (%rdi)
+; SSE42-NEXT:    andb $1, %bpl
+; SSE42-NEXT:    movb %bpl, (%rdi)
+; SSE42-NEXT:    andb $1, %al
+; SSE42-NEXT:    movb %al, (%rdi)
 ; SSE42-NEXT:    andb $1, %cl
 ; SSE42-NEXT:    movb %cl, (%rdi)
-; SSE42-NEXT:    andb $1, %r9b
-; SSE42-NEXT:    movb %r9b, (%rdi)
 ; SSE42-NEXT:    andb $1, %dl
 ; SSE42-NEXT:    movb %dl, (%rdi)
-; SSE42-NEXT:    andb $1, %r10b
-; SSE42-NEXT:    movb %r10b, (%rdi)
 ; SSE42-NEXT:    andb $1, %sil
 ; SSE42-NEXT:    movb %sil, (%rdi)
-; SSE42-NEXT:    andb $1, %r13b
-; SSE42-NEXT:    movb %r13b, (%rdi)
-; SSE42-NEXT:    andb $1, %al
-; SSE42-NEXT:    movb %al, (%rdi)
-; SSE42-NEXT:    andb $1, %r8b
-; SSE42-NEXT:    movb %r8b, (%rdi)
-; SSE42-NEXT:    andb $1, %r14b
-; SSE42-NEXT:    movb %r14b, (%rdi)
 ; SSE42-NEXT:    movq %rdi, %rax
 ; SSE42-NEXT:    popq %rbx
 ; SSE42-NEXT:    popq %r12
 ; SSE42-NEXT:    popq %r13
 ; SSE42-NEXT:    popq %r14
 ; SSE42-NEXT:    popq %r15
 ; SSE42-NEXT:    popq %rbp
 ; SSE42-NEXT:    retq