[X86] Try to avoid casts around logical vector ops recursively.

Currently PromoteMaskArithemtic only looks at a single operation to
skip casts. This means we miss cases where we combine multiple masks.

This patch updates PromoteMaskArithemtic to try to recursively promote
AND/XOR/AND nodes that terminate in truncates of the right size or
constant vectors.

Reviewers: craig.topper, RKSimon, spatel

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D72524

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 5a1dd9f..e24e221 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -39898,6 +39898,65 @@ static SDValue combineANDXORWithAllOnesIntoANDNP(SDNode *N, SelectionDAG &DAG) {
   return DAG.getNode(X86ISD::ANDNP, SDLoc(N), VT, X, Y);
 }
 
+// Try to widen AND, OR and XOR nodes to VT in order to remove casts around
+// logical operations, like in the example below.
+//   or (and (truncate x, truncate y)),
+//      (xor (truncate z, build_vector (constants)))
+// Given a target type \p VT, we generate
+//   or (and x, y), (xor z, zext(build_vector (constants)))
+// given x, y and z are of type \p VT. We can do so, if operands are either
+// truncates from VT types, the second operand is a vector of constants or can
+// be recursively promoted.
+static SDValue PromoteMaskArithmetic(SDNode *N, EVT VT, SelectionDAG &DAG,
+                                     unsigned Depth) {
+  // Limit recursion to avoid excessive compile times.
+  if (Depth >= SelectionDAG::MaxRecursionDepth)
+    return SDValue();
+
+  if (N->getOpcode() != ISD::XOR && N->getOpcode() != ISD::AND &&
+      N->getOpcode() != ISD::OR)
+    return SDValue();
+
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+  SDLoc DL(N);
+
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  if (!TLI.isOperationLegalOrPromote(N->getOpcode(), VT))
+    return SDValue();
+
+  if (SDValue NN0 = PromoteMaskArithmetic(N0.getNode(), VT, DAG, Depth + 1))
+    N0 = NN0;
+  else {
+    // The Left side has to be a trunc.
+    if (N0.getOpcode() != ISD::TRUNCATE)
+      return SDValue();
+
+    // The type of the truncated inputs.
+    if (N0.getOperand(0).getValueType() != VT)
+      return SDValue();
+
+    N0 = N0.getOperand(0);
+  }
+
+  if (SDValue NN1 = PromoteMaskArithmetic(N1.getNode(), VT, DAG, Depth + 1))
+    N1 = NN1;
+  else {
+    // The right side has to be a 'trunc' or a constant vector.
+    bool RHSTrunc = N1.getOpcode() == ISD::TRUNCATE &&
+                    N1.getOperand(0).getValueType() == VT;
+    if (!RHSTrunc && !ISD::isBuildVectorOfConstantSDNodes(N1.getNode()))
+      return SDValue();
+
+    if (RHSTrunc)
+      N1 = N1.getOperand(0);
+    else
+      N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, VT, N1);
+  }
+
+  return DAG.getNode(N->getOpcode(), DL, VT, N0, N1);
+}
+
 // On AVX/AVX2 the type v8i1 is legalized to v8i16, which is an XMM sized
 // register. In most cases we actually compare or select YMM-sized registers
 // and mixing the two types creates horrible code. This method optimizes
@@ -39909,6 +39968,7 @@ static SDValue PromoteMaskArithmetic(SDNode *N, SelectionDAG &DAG,
   EVT VT = N->getValueType(0);
   assert(VT.isVector() && "Expected vector type");
 
+  SDLoc DL(N);
   assert((N->getOpcode() == ISD::ANY_EXTEND ||
           N->getOpcode() == ISD::ZERO_EXTEND ||
           N->getOpcode() == ISD::SIGN_EXTEND) && "Invalid Node");
@@ -39916,46 +39976,11 @@ static SDValue PromoteMaskArithmetic(SDNode *N, SelectionDAG &DAG,
   SDValue Narrow = N->getOperand(0);
   EVT NarrowVT = Narrow.getValueType();
 
-  if (Narrow->getOpcode() != ISD::XOR &&
-      Narrow->getOpcode() != ISD::AND &&
-      Narrow->getOpcode() != ISD::OR)
-    return SDValue();
-
-  SDValue N0  = Narrow->getOperand(0);
-  SDValue N1  = Narrow->getOperand(1);
-  SDLoc DL(Narrow);
-
-  // The Left side has to be a trunc.
-  if (N0.getOpcode() != ISD::TRUNCATE)
-    return SDValue();
-
-  // The type of the truncated inputs.
-  if (N0.getOperand(0).getValueType() != VT)
-    return SDValue();
-
-  // The right side has to be a 'trunc' or a constant vector.
-  bool RHSTrunc = N1.getOpcode() == ISD::TRUNCATE &&
-                  N1.getOperand(0).getValueType() == VT;
-  if (!RHSTrunc &&
-      !ISD::isBuildVectorOfConstantSDNodes(N1.getNode()))
-    return SDValue();
-
-  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
-
-  if (!TLI.isOperationLegalOrPromote(Narrow->getOpcode(), VT))
-    return SDValue();
-
-  // Set N0 and N1 to hold the inputs to the new wide operation.
-  N0 = N0.getOperand(0);
-  if (RHSTrunc)
-    N1 = N1.getOperand(0);
-  else
-    N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, VT, N1);
-
   // Generate the wide operation.
-  SDValue Op = DAG.getNode(Narrow->getOpcode(), DL, VT, N0, N1);
-  unsigned Opcode = N->getOpcode();
-  switch (Opcode) {
+  SDValue Op = PromoteMaskArithmetic(Narrow.getNode(), VT, DAG, 0);
+  if (!Op)
+    return SDValue();
+  switch (N->getOpcode()) {
   default: llvm_unreachable("Unexpected opcode");
   case ISD::ANY_EXTEND:
     return Op;

diff --git a/llvm/test/CodeGen/X86/v8i1-masks.ll b/llvm/test/CodeGen/X86/v8i1-masks.ll
index 82ff357..a661e77 100644 (file)
--- a/llvm/test/CodeGen/X86/v8i1-masks.ll
+++ b/llvm/test/CodeGen/X86/v8i1-masks.ll
@@ -212,18 +212,10 @@ define <8 x i32> @three_ands(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI4_1, %ymm0, %ymm2
-; X32-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: three_ands:
@@ -231,18 +223,10 @@ define <8 x i32> @three_ands(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: three_ands:
@@ -251,15 +235,10 @@ define <8 x i32> @three_ands(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: three_ands:
@@ -268,15 +247,10 @@ define <8 x i32> @three_ands(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -294,22 +268,12 @@ define <8 x i32> @four_ands(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI5_1, %ymm0, %ymm2
-; X32-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI5_2, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X32-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: four_ands:
@@ -317,22 +281,12 @@ define <8 x i32> @four_ands(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X64-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: four_ands:
@@ -341,20 +295,13 @@ define <8 x i32> @four_ands(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X32-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: four_ands:
@@ -363,20 +310,13 @@ define <8 x i32> @four_ands(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X64-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -396,26 +336,14 @@ define <8 x i32> @five_ands(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI6_1, %ymm0, %ymm2
-; X32-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI6_2, %ymm0, %ymm3
-; X32-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI6_3, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X32-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: five_ands:
@@ -423,26 +351,14 @@ define <8 x i32> @five_ands(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
-; X64-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X64-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: five_ands:
@@ -451,25 +367,16 @@ define <8 x i32> @five_ands(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X32-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: five_ands:
@@ -478,25 +385,16 @@ define <8 x i32> @five_ands(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
-; X64-AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -560,18 +458,10 @@ define <8 x i32> @three_or(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI8_1, %ymm0, %ymm2
-; X32-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: three_or:
@@ -579,18 +469,10 @@ define <8 x i32> @three_or(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: three_or:
@@ -599,15 +481,10 @@ define <8 x i32> @three_or(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: three_or:
@@ -616,15 +493,10 @@ define <8 x i32> @three_or(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -642,22 +514,12 @@ define <8 x i32> @four_or(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI9_1, %ymm0, %ymm2
-; X32-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI9_2, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: four_or:
@@ -665,22 +527,12 @@ define <8 x i32> @four_or(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: four_or:
@@ -689,20 +541,13 @@ define <8 x i32> @four_or(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: four_or:
@@ -711,20 +556,13 @@ define <8 x i32> @four_or(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -744,26 +582,14 @@ define <8 x i32> @five_or(<8 x float> %x) {
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI10_1, %ymm0, %ymm2
-; X32-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI10_2, %ymm0, %ymm3
-; X32-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; X32-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI10_3, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: five_or:
@@ -771,26 +597,14 @@ define <8 x i32> @five_or(<8 x float> %x) {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
-; X64-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
-; X64-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; X64-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: five_or:
@@ -799,25 +613,16 @@ define <8 x i32> @five_or(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-AVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; X32-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: five_or:
@@ -826,25 +631,16 @@ define <8 x i32> @five_or(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vorps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-AVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; X64-AVX2-NEXT:    vorps %ymm3, %ymm2, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -923,19 +719,11 @@ define <8 x i32> @four_or_and(<8 x float> %x) {
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI12_1, %ymm0, %ymm2
 ; X32-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI12_2, %ymm0, %ymm0
 ; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: four_or_and:
@@ -944,19 +732,11 @@ define <8 x i32> @four_or_and(<8 x float> %x) {
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
 ; X64-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
 ; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: four_or_and:
@@ -966,17 +746,12 @@ define <8 x i32> @four_or_and(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: four_or_and:
@@ -986,17 +761,12 @@ define <8 x i32> @four_or_and(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -1019,19 +789,11 @@ define <8 x i32> @five_or_and(<8 x float> %x) {
 ; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
 ; X32-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X32-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vcmpneqps LCPI13_2, %ymm0, %ymm2
+; X32-NEXT:    vcmpneqps LCPI13_2, %ymm0, %ymm3
 ; X32-NEXT:    vcmpneqps LCPI13_3, %ymm0, %ymm0
-; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm3, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm2, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: five_or_and:
@@ -1042,19 +804,11 @@ define <8 x i32> @five_or_and(<8 x float> %x) {
 ; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
 ; X64-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X64-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm2
+; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm3, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm2, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: five_or_and:
@@ -1066,18 +820,13 @@ define <8 x i32> @five_or_and(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
 ; X32-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm3, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: five_or_and:
@@ -1089,18 +838,13 @@ define <8 x i32> @five_or_and(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
 ; X64-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X64-AVX2-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm3, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm2, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -1123,19 +867,11 @@ define <8 x i32> @four_or_and_xor(<8 x float> %x) {
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X32-NEXT:    vcmpltps LCPI14_1, %ymm0, %ymm2
 ; X32-NEXT:    vxorps %ymm2, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; X32-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X32-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X32-NEXT:    vcmpneqps LCPI14_2, %ymm0, %ymm0
 ; X32-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: four_or_and_xor:
@@ -1144,19 +880,11 @@ define <8 x i32> @four_or_and_xor(<8 x float> %x) {
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
 ; X64-NEXT:    vxorps %ymm2, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; X64-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X64-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
 ; X64-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: four_or_and_xor:
@@ -1166,17 +894,12 @@ define <8 x i32> @four_or_and_xor(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vxorps %ymm2, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X32-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: four_or_and_xor:
@@ -1186,17 +909,12 @@ define <8 x i32> @four_or_and_xor(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vxorps %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -1215,100 +933,66 @@ define <8 x i32> @five_or_and_xor(<8 x float> %x) {
 ; X32:       ## %bb.0: ## %entry
 ; X32-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X32-NEXT:    vcmpltps LCPI15_1, %ymm0, %ymm2
 ; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-NEXT:    vxorps %ymm3, %ymm2, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps LCPI15_2, %ymm0, %ymm3
+; X32-NEXT:    vcmpneqps LCPI15_2, %ymm0, %ymm4
 ; X32-NEXT:    vcmpneqps LCPI15_3, %ymm0, %ymm0
-; X32-NEXT:    vandps %ymm0, %ymm3, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-NEXT:    vpxor %xmm0, %xmm2, %xmm0
-; X32-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; X32-NEXT:    vxorps %ymm0, %ymm3, %ymm0
+; X32-NEXT:    vxorps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: five_or_and_xor:
 ; X64:       ## %bb.0: ## %entry
 ; X64-NEXT:    vmovaps {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
 ; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-NEXT:    vxorps %ymm3, %ymm2, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
+; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm4
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vandps %ymm0, %ymm3, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-NEXT:    vpxor %xmm0, %xmm2, %xmm0
-; X64-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; X64-NEXT:    vxorps %ymm0, %ymm3, %ymm0
+; X64-NEXT:    vxorps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: five_or_and_xor:
 ; X32-AVX2:       ## %bb.0: ## %entry
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X32-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vxorps %ymm3, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vandps %ymm0, %ymm3, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpxor %xmm0, %xmm2, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm4
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm5 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm5, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; X32-AVX2-NEXT:    vxorps %ymm0, %ymm3, %ymm0
+; X32-AVX2-NEXT:    vxorps %ymm0, %ymm2, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: five_or_and_xor:
 ; X64-AVX2:       ## %bb.0: ## %entry
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
 ; X64-AVX2-NEXT:    vcmpleps %ymm0, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vxorps %ymm3, %ymm2, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vandps %ymm0, %ymm3, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm0, %xmm2, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm4
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm5 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm5, %ymm0, %ymm0
+; X64-AVX2-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; X64-AVX2-NEXT:    vxorps %ymm0, %ymm3, %ymm0
+; X64-AVX2-NEXT:    vxorps %ymm0, %ymm2, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
@@ -1331,26 +1015,14 @@ define <8 x i32> @six_or_and_xor(<8 x float> %x) {
 ; X32-NEXT:    vcmpltps LCPI16_1, %ymm0, %ymm2
 ; X32-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-NEXT:    vcmpneqps LCPI16_2, %ymm0, %ymm4
+; X32-NEXT:    vandps %ymm4, %ymm3, %ymm3
 ; X32-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X32-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps LCPI16_2, %ymm0, %ymm3
-; X32-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X32-NEXT:    vcmpneqps LCPI16_3, %ymm0, %ymm3
-; X32-NEXT:    vxorps %ymm1, %ymm3, %ymm1
-; X32-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; X32-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; X32-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; X32-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; X32-NEXT:    vcmpneqps LCPI16_3, %ymm0, %ymm2
+; X32-NEXT:    vxorps %ymm1, %ymm2, %ymm1
 ; X32-NEXT:    vcmpneqps LCPI16_4, %ymm0, %ymm0
-; X32-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X32-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X32-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: six_or_and_xor:
@@ -1360,26 +1032,14 @@ define <8 x i32> @six_or_and_xor(<8 x float> %x) {
 ; X64-NEXT:    vcmpltps {{.*}}(%rip), %ymm0, %ymm2
 ; X64-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm4
+; X64-NEXT:    vandps %ymm4, %ymm3, %ymm3
 ; X64-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X64-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
-; X64-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm3
-; X64-NEXT:    vxorps %ymm1, %ymm3, %ymm1
-; X64-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; X64-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; X64-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm2
+; X64-NEXT:    vxorps %ymm1, %ymm2, %ymm1
 ; X64-NEXT:    vcmpneqps {{.*}}(%rip), %ymm0, %ymm0
-; X64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X32-AVX2-LABEL: six_or_and_xor:
@@ -1390,26 +1050,17 @@ define <8 x i32> @six_or_and_xor(<8 x float> %x) {
 ; X32-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X32-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm4
+; X32-AVX2-NEXT:    vandps %ymm4, %ymm3, %ymm3
 ; X32-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X32-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X32-AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X32-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vxorps %ymm1, %ymm3, %ymm1
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; X32-AVX2-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; X32-AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; X32-AVX2-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
+; X32-AVX2-NEXT:    vxorps %ymm1, %ymm2, %ymm1
 ; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1]
 ; X32-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X32-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X32-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X32-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X32-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: six_or_and_xor:
@@ -1420,26 +1071,17 @@ define <8 x i32> @six_or_and_xor(<8 x float> %x) {
 ; X64-AVX2-NEXT:    vcmpltps %ymm2, %ymm0, %ymm2
 ; X64-AVX2-NEXT:    vxorps %xmm3, %xmm3, %xmm3
 ; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm4, %ymm0, %ymm4
+; X64-AVX2-NEXT:    vandps %ymm4, %ymm3, %ymm3
 ; X64-AVX2-NEXT:    vandps %ymm3, %ymm2, %ymm2
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; X64-AVX2-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
-; X64-AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
-; X64-AVX2-NEXT:    vcmpneqps %ymm3, %ymm0, %ymm3
-; X64-AVX2-NEXT:    vxorps %ymm1, %ymm3, %ymm1
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; X64-AVX2-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; X64-AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; X64-AVX2-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
+; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm2
+; X64-AVX2-NEXT:    vxorps %ymm1, %ymm2, %ymm1
 ; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1]
 ; X64-AVX2-NEXT:    vcmpneqps %ymm2, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; X64-AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; X64-AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
+; X64-AVX2-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; X64-AVX2-NEXT:    retq
 entry:
   %cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>