[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8.

[craig.topper]

Summary

For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register.

I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes.

There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined.

I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc))

Overall this removes something like 13000 CHECK lines from lit tests.

[craig.topper]

Rebase after improving vXi16/vXi8 select combines

[RKSimon]

There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined.

I've been putting off looking at this but I've been thinking about ways to improve shuffle combining - 2 things come to mind: some form of DemandedVectorElts (preferably in DAGCombine but just in x86 if necessary) and the ability for the x86 shuffle combines to combine through subvector insertion/extraction/concat (and avx512 compress/expand?). Supporting extend/trucate could fit in with the second.

[craig.topper]

Rebase

[craig.topper]

While the v32i1 is technically a regression, its not a fair test of what happens with v32i1 shuffles. Previously we were getting lucky with type promotion working favorably with the types used for argument passing. I think a shuffle sandwiched between say an icmp and a select condition would be very different.

I'll add more test cases to show what happens when vXi1 shuffles are used near k-registers.

[craig.topper]

Add diffs for more v32i1 shuffle test case. I haven't commited the current versions of the tests to the repo yet, but you can see the before and after here.

Looks like we still have a regression when we have a shuffle between a v32i8 icmp and a v32i8 select. We could add a DAG combine to detect sign extends of i1 shuffles or i1 shuffles of truncates and adjust the type to match instead of somewhat arbitrarily picking a type to extend due to during lowering.

[RKSimon]

test/CodeGen/X86/avx512-insert-extract.ll

1068

Is the andb necessary? Are we missing some known bits handling here?

[craig.topper]

test/CodeGen/X86/avx512-insert-extract.ll

1068

Yeah its unnecessary based on how it gets isel. But I'm not sure the way the DAG is written that we can guarantee it.

t0: ch = EntryToken
                t2: v64i8,ch = CopyFromReg t0, Register:v64i8 %0
                t4: v64i8,ch = CopyFromReg t0, Register:v64i8 %1
              t24: v64i1 = X86ISD::CMPMU t2, t4, Constant:i8<6>
            t28: v64i1 = X86ISD::KSHIFTR t24, Constant:i8<63>
          t30: i32 = extract_vector_elt t28, Constant:i64<0>
        t26: i8 = truncate t30
      t22: i8 = and t26, Constant:i8<1>
    t19: i8 = sub Constant:i8<4>, t22
  t13: i32 = zero_extend t19
t16: ch,glue = CopyToReg t0, Register:i32 %eax, t13
t17: ch = X86ISD::RET_FLAG t16, TargetConstant:i32<0>, Register:i32 %eax, t16:1

I don't think we can make any assumptions about the upper bits of the input to the extract_vector_elt being passed through.

We could maybe pattern match this out during isel?

Or we could legalize to a DAG that explicitly contains a KMOVW node instead of extract_vector_elt. Then we could probably put out an AssertZExt.

[RKSimon]

test/CodeGen/X86/avx512-insert-extract.ll

1068

Couldn't we handle this by adding X86ISD::KSHIFTR handling to X86TargetLowering::computeKnownBitsForTargetNode (with suitable DemandedElts twiddling)?

[craig.topper]

test/CodeGen/X86/avx512-insert-extract.ll

1068

But computeKnownBits for extract_vector_elt won't pass the info through. It doesn't know the bits from the upper elements are going to make it through to the result. (they don't for any other vector type). It would just take the known bits info about element 0 and any extend that to i32.

[craig.topper]

Rebase to change vmovdqa32 to vmovdqa64 in some tests.

[RKSimon]

test/CodeGen/X86/avx512-insert-extract.ll

1068

OK - sorry I was thinking that the extract_vector_elt disappeared when it became a KSHIFT+KMOVW.

[craig.topper]

Is this patch good enough to go in as is? And we file bugs and work on any problems after?

test/CodeGen/X86/avx512-insert-extract.ll

1068

Oh sorry. If we use KMOVW explicitly in the DAG then yes the extract_vector_elt goes away. But it still looks sort of like a vector to scalar bitcast.

The KMOVW will demand all elts from the KSHIFT. And the KSHIFT won't be able to say that some elts are 0 and the lowest element is unknown will it?

[craig.topper]

Rebase after prefer vector width.

[RKSimon]

LGTM

test/CodeGen/X86/avx512-insert-extract.ll

1068

OK - it sounds like we need some better computeknownbits / demandedbits support for target nodes - that was what D38832 was trying to do yes?

[craig.topper]

test/CodeGen/X86/avx512-insert-extract.ll

1068

I don't know if that helps here. For vectors, compute known bits tells the common known bits of all elements. So for the KSHIFT it would only return a KnownBits object with width 1. That won't be able to convey anything about the upper elements separately from the lower element.

[craig.topper]

test/CodeGen/X86/avx512-insert-extract.ll

1068

Maybe it makes more sense to do the KMOV and then a GPR shift if assuming we have a wide enough KMOV available? KSHIFT has a 3 cycle latency on SKX while a GPR shift is only 1 cyc.

Known bits would handle that well.

lib/Target/X86/X86ISelLowering.h

public:
     SDValue BuildFILD(SDValue Op, EVT SrcVT, SDValue Chain, SDValue StackSlot,
                       SelectionDAG &DAG) const;
 
     bool isNoopAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const override;
 
     /// \brief Customize the preferred legalization strategy for certain types.
     LegalizeTypeAction getPreferredVectorAction(EVT VT) const override;
 
+    MVT getRegisterTypeForCallingConv(MVT VT) const override;
+
+    MVT getRegisterTypeForCallingConv(LLVMContext &Context,
+                                      EVT VT) const override;
+
+    unsigned getNumRegistersForCallingConv(LLVMContext &Context,
+                                           EVT VT) const override;
+
     bool isIntDivCheap(EVT VT, AttributeList Attr) const override;
 
     bool supportSwiftError() const override;
 
     StringRef getStackProbeSymbolName(MachineFunction &MF) const override;
 
     unsigned getMaxSupportedInterleaveFactor() const override { return 4; }
 

lib/Target/X86/X86ISelLowering.cpp

SDValue X86TargetLowering::emitStackGuardXorFP(SelectionDAG &DAG, SDValue Val,
   EVT PtrTy = getPointerTy(DAG.getDataLayout());
   unsigned XorOp = Subtarget.is64Bit() ? X86::XOR64_FP : X86::XOR32_FP;
   MachineSDNode *Node = DAG.getMachineNode(XorOp, DL, PtrTy, Val);
   return SDValue(Node, 0);
 }
 
 TargetLoweringBase::LegalizeTypeAction
 X86TargetLowering::getPreferredVectorAction(EVT VT) const {
+  if (VT == MVT::v32i1 && Subtarget.hasAVX512() && !Subtarget.hasBWI())
+    return TypeSplitVector;
+
   if (ExperimentalVectorWideningLegalization &&
       VT.getVectorNumElements() != 1 &&
       VT.getVectorElementType().getSimpleVT() != MVT::i1)
     return TypeWidenVector;
 
   return TargetLoweringBase::getPreferredVectorAction(VT);
 }
 
+MVT X86TargetLowering::getRegisterTypeForCallingConv(MVT VT) const {
+  if (VT == MVT::v32i1 && Subtarget.hasAVX512() && !Subtarget.hasBWI())
+    return MVT::v32i8;
+  return TargetLowering::getRegisterTypeForCallingConv(VT);
+}
+
+MVT X86TargetLowering::getRegisterTypeForCallingConv(LLVMContext &Context,
+                                                     EVT VT) const {
+  if (VT == MVT::v32i1 && Subtarget.hasAVX512() && !Subtarget.hasBWI())
+    return MVT::v32i8;
+  return TargetLowering::getRegisterTypeForCallingConv(Context, VT);
+}
+
+unsigned X86TargetLowering::getNumRegistersForCallingConv(LLVMContext &Context,
+                                                          EVT VT) const {
+  if (VT == MVT::v32i1 && Subtarget.hasAVX512() && !Subtarget.hasBWI())
+    return 1;
+  return TargetLowering::getNumRegistersForCallingConv(Context, VT);
+}
+
 EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL,
                                           LLVMContext& Context,
                                           EVT VT) const {
   if (!VT.isVector())
     return MVT::i8;
 
   if (Subtarget.hasAVX512()) {
     const unsigned NumElts = VT.getVectorNumElements();

test/CodeGen/X86/avg-mask.ll

; AVX512BWVL-NEXT:    retq
   %mask1 = bitcast i16 %mask to <16 x i1>
   %res = select <16 x i1> %mask1, <16 x i8> %trunc, <16 x i8> zeroinitializer
   ret <16 x i8> %res
 }
 
 define <32 x i8> @avg_v32i8_mask(<32 x i8> %a, <32 x i8> %b, <32 x i8> %src, i32 %mask) nounwind {
 ; AVX512F-LABEL: avg_v32i8_mask:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
 ; AVX512F-NEXT:    vpavgb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
+; AVX512F-NEXT:    kmovw %edi, %k2
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
 ; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v32i8_mask:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovd %edi, %k1
 ; AVX512BWVL-NEXT:    vpavgb %ymm1, %ymm0, %ymm2 {%k1}
 ; AVX512BWVL-NEXT:    vmovdqa %ymm2, %ymm0
 ; AVX512BWVL-NEXT:    retq
   %za = zext <32 x i8> %a to <32 x i16>
   %zb = zext <32 x i8> %b to <32 x i16>
   %add = add nuw nsw <32 x i16> %za, %zb
   %add1 = add nuw nsw <32 x i16> %add, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %lshr = lshr <32 x i16> %add1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %trunc = trunc <32 x i16> %lshr to <32 x i8>
   %mask1 = bitcast i32 %mask to <32 x i1>
   %res = select <32 x i1> %mask1, <32 x i8> %trunc, <32 x i8> %src
   ret <32 x i8> %res
 }
 
 define <32 x i8> @avg_v32i8_maskz(<32 x i8> %a, <32 x i8> %b, i32 %mask) nounwind {
 ; AVX512F-LABEL: avg_v32i8_maskz:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
 ; AVX512F-NEXT:    vpavgb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
+; AVX512F-NEXT:    kmovw %edi, %k2
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
 ; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v32i8_maskz:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovd %edi, %k1
 ; AVX512BWVL-NEXT:    vpavgb %ymm1, %ymm0, %ymm0 {%k1} {z}
 ; AVX512BWVL-NEXT:    retq
   %za = zext <32 x i8> %a to <32 x i16>
   %zb = zext <32 x i8> %b to <32 x i16>
   %add = add nuw nsw <32 x i16> %za, %zb
   %add1 = add nuw nsw <32 x i16> %add, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %lshr = lshr <32 x i16> %add1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %trunc = trunc <32 x i16> %lshr to <32 x i8>
   %mask1 = bitcast i32 %mask to <32 x i1>
   %res = select <32 x i1> %mask1, <32 x i8> %trunc, <32 x i8> zeroinitializer
   ret <32 x i8> %res
 }
 
 define <64 x i8> @avg_v64i8_mask(<64 x i8> %a, <64 x i8> %b, <64 x i8> %src, i64 %mask) nounwind {
 ; AVX512F-LABEL: avg_v64i8_mask:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $64, %rsp
 ; AVX512F-NEXT:    movq %rdi, %rax
-; AVX512F-NEXT:    shrq $32, %rax
-; AVX512F-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    movl %edi, (%rsp)
+; AVX512F-NEXT:    movq %rdi, %rcx
+; AVX512F-NEXT:    kmovw %edi, %k2
+; AVX512F-NEXT:    movl %edi, %edx
+; AVX512F-NEXT:    shrl $16, %edx
+; AVX512F-NEXT:    shrq $48, %rax
+; AVX512F-NEXT:    shrq $32, %rcx
 ; AVX512F-NEXT:    vpavgb %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpavgb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
+; AVX512F-NEXT:    kmovw %ecx, %k1
+; AVX512F-NEXT:    kmovw %eax, %k3
+; AVX512F-NEXT:    kmovw %edx, %k4
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k4} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
-; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm5, %ymm1
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm6
+; AVX512F-NEXT:    vpblendvb %xmm3, %xmm2, %xmm6, %xmm2
 ; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
-; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    vpblendvb %xmm3, %xmm0, %xmm4, %xmm0
+; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k3} {z}
+; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512F-NEXT:    vextracti128 $1, %ymm5, %xmm4
+; AVX512F-NEXT:    vpblendvb %xmm3, %xmm2, %xmm4, %xmm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512F-NEXT:    vpblendvb %xmm3, %xmm1, %xmm5, %xmm1
+; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v64i8_mask:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovq %rdi, %k1
 ; AVX512BWVL-NEXT:    vpavgb %zmm1, %zmm0, %zmm2 {%k1}
 ; AVX512BWVL-NEXT:    vmovdqa64 %zmm2, %zmm0
 ; AVX512BWVL-NEXT:    retq
   %za = zext <64 x i8> %a to <64 x i16>
   %zb = zext <64 x i8> %b to <64 x i16>
   %add = add nuw nsw <64 x i16> %za, %zb
   %add1 = add nuw nsw <64 x i16> %add, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %lshr = lshr <64 x i16> %add1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %trunc = trunc <64 x i16> %lshr to <64 x i8>
   %mask1 = bitcast i64 %mask to <64 x i1>
   %res = select <64 x i1> %mask1, <64 x i8> %trunc, <64 x i8> %src
   ret <64 x i8> %res
 }
 
 define <64 x i8> @avg_v64i8_maskz(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind {
 ; AVX512F-LABEL: avg_v64i8_maskz:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $64, %rsp
 ; AVX512F-NEXT:    movq %rdi, %rax
-; AVX512F-NEXT:    shrq $32, %rax
-; AVX512F-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    movl %edi, (%rsp)
+; AVX512F-NEXT:    movq %rdi, %rcx
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    movl %edi, %edx
+; AVX512F-NEXT:    shrl $16, %edx
+; AVX512F-NEXT:    shrq $48, %rax
+; AVX512F-NEXT:    shrq $32, %rcx
 ; AVX512F-NEXT:    vpavgb %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpavgb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
+; AVX512F-NEXT:    kmovw %ecx, %k2
+; AVX512F-NEXT:    kmovw %eax, %k3
+; AVX512F-NEXT:    kmovw %edx, %k4
 ; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
+; AVX512F-NEXT:    vpand %xmm0, %xmm2, %xmm2
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k4} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
-; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512F-NEXT:    vpand %xmm0, %xmm3, %xmm0
+; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k3} {z}
+; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512F-NEXT:    vpand %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
-; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpand %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    vpand %xmm1, %xmm3, %xmm1
+; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v64i8_maskz:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovq %rdi, %k1
 ; AVX512BWVL-NEXT:    vpavgb %zmm1, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BWVL-NEXT:    retq
   %za = zext <64 x i8> %a to <64 x i16>
; AVX512BWVL-NEXT:    retq
   %mask1 = bitcast i16 %mask to <16 x i1>
   %res = select <16 x i1> %mask1, <16 x i16> %trunc, <16 x i16> zeroinitializer
   ret <16 x i16> %res
 }
 
 define <32 x i16> @avg_v32i16_mask(<32 x i16> %a, <32 x i16> %b, <32 x i16> %src, i32 %mask) nounwind {
 ; AVX512F-LABEL: avg_v32i16_mask:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm6, %zmm6, %zmm6 {%k2} {z}
-; AVX512F-NEXT:    vpmovdb %zmm6, %xmm6
-; AVX512F-NEXT:    vpternlogd $255, %zmm7, %zmm7, %zmm7 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm7, %xmm7
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
 ; AVX512F-NEXT:    vpavgw %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpavgw %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero,xmm7[8],zero,xmm7[9],zero,xmm7[10],zero,xmm7[11],zero,xmm7[12],zero,xmm7[13],zero,xmm7[14],zero,xmm7[15],zero
-; AVX512F-NEXT:    vpsllw $15, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsraw $15, %ymm2, %ymm2
+; AVX512F-NEXT:    kmovw %edi, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero,xmm6[8],zero,xmm6[9],zero,xmm6[10],zero,xmm6[11],zero,xmm6[12],zero,xmm6[13],zero,xmm6[14],zero,xmm6[15],zero
-; AVX512F-NEXT:    vpsllw $15, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsraw $15, %ymm2, %ymm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm5, %ymm1
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v32i16_mask:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovd %edi, %k1
 ; AVX512BWVL-NEXT:    vpavgw %zmm1, %zmm0, %zmm2 {%k1}
 ; AVX512BWVL-NEXT:    vmovdqa64 %zmm2, %zmm0
 ; AVX512BWVL-NEXT:    retq
   %za = zext <32 x i16> %a to <32 x i32>
   %zb = zext <32 x i16> %b to <32 x i32>
   %add = add nuw nsw <32 x i32> %za, %zb
   %add1 = add nuw nsw <32 x i32> %add, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   %lshr = lshr <32 x i32> %add1, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   %trunc = trunc <32 x i32> %lshr to <32 x i16>
   %mask1 = bitcast i32 %mask to <32 x i1>
   %res = select <32 x i1> %mask1, <32 x i16> %trunc, <32 x i16> %src
   ret <32 x i16> %res
 }
 
 define <32 x i16> @avg_v32i16_maskz(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind {
 ; AVX512F-LABEL: avg_v32i16_maskz:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k2} {z}
-; AVX512F-NEXT:    vpmovdb %zmm4, %xmm4
-; AVX512F-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm5, %xmm5
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
 ; AVX512F-NEXT:    vpavgw %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpavgw %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero,xmm5[8],zero,xmm5[9],zero,xmm5[10],zero,xmm5[11],zero,xmm5[12],zero,xmm5[13],zero,xmm5[14],zero,xmm5[15],zero
-; AVX512F-NEXT:    vpsllw $15, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsraw $15, %ymm2, %ymm2
+; AVX512F-NEXT:    kmovw %edi, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512F-NEXT:    vpand %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512F-NEXT:    vpsllw $15, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsraw $15, %ymm2, %ymm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512F-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: avg_v32i16_maskz:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    kmovd %edi, %k1
 ; AVX512BWVL-NEXT:    vpavgw %zmm1, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BWVL-NEXT:    retq
   %za = zext <32 x i16> %a to <32 x i32>

test/CodeGen/X86/avx512-ext.ll

; SKX-NEXT:    retq
   %x   = sext <16 x i8> %a to <16 x i16>
   %ret = select <16 x i1> %mask, <16 x i16> %x, <16 x i16> zeroinitializer
   ret <16 x i16> %ret
 }
 
 define <32 x i16> @zext_32x8mem_to_32x16(<32 x i8> *%i , <32 x i1> %mask) nounwind readnone {
 ; KNL-LABEL: zext_32x8mem_to_32x16:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; KNL-NEXT:    vpsllw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpsraw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpand %ymm2, %ymm3, %ymm2
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
 ; KNL-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; KNL-NEXT:    vpsraw $15, %ymm0, %ymm0
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm1
-; KNL-NEXT:    vmovdqa %ymm2, %ymm0
+; KNL-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: zext_32x8mem_to_32x16:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovb2m %ymm0, %k1
 ; SKX-NEXT:    vpmovzxbw {{.*#+}} zmm0 {%k1} {z} = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero,mem[16],zero,mem[17],zero,mem[18],zero,mem[19],zero,mem[20],zero,mem[21],zero,mem[22],zero,mem[23],zero,mem[24],zero,mem[25],zero,mem[26],zero,mem[27],zero,mem[28],zero,mem[29],zero,mem[30],zero,mem[31],zero
 ; SKX-NEXT:    retq
   %a   = load <32 x i8>,<32 x i8> *%i,align 1
   %x   = zext <32 x i8> %a to <32 x i16>
   %ret = select <32 x i1> %mask, <32 x i16> %x, <32 x i16> zeroinitializer
   ret <32 x i16> %ret
 }
 
 define <32 x i16> @sext_32x8mem_to_32x16(<32 x i8> *%i , <32 x i1> %mask) nounwind readnone {
 ; KNL-LABEL: sext_32x8mem_to_32x16:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpmovsxbw 16(%rdi), %ymm1
-; KNL-NEXT:    vpmovsxbw (%rdi), %ymm2
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; KNL-NEXT:    vpsllw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpsraw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpand %ymm2, %ymm3, %ymm2
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; KNL-NEXT:    vpmovsxbw 16(%rdi), %ymm2
+; KNL-NEXT:    vpmovsxbw (%rdi), %ymm3
 ; KNL-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; KNL-NEXT:    vpsraw $15, %ymm0, %ymm0
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm1
-; KNL-NEXT:    vmovdqa %ymm2, %ymm0
+; KNL-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: sext_32x8mem_to_32x16:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovb2m %ymm0, %k1
 ; SKX-NEXT:    vpmovsxbw (%rdi), %zmm0 {%k1} {z}
 ; SKX-NEXT:    retq

 ; SKX-NEXT:    retq
   %x   = zext <32 x i8> %a to <32 x i16>
   ret <32 x i16> %x
 }
 
 define <32 x i16> @zext_32x8_to_32x16_mask(<32 x i8> %a ,<32 x i1> %mask) nounwind readnone {
 ; KNL-LABEL: zext_32x8_to_32x16_mask:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; KNL-NEXT:    vpsllw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpsraw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpand %ymm0, %ymm3, %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; KNL-NEXT:    vpsllw $15, %ymm1, %ymm1
 ; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; KNL-NEXT:    vpsllw $15, %ymm2, %ymm1
+; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm3, %ymm1, %ymm1
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: zext_32x8_to_32x16_mask:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm1, %ymm1
 ; SKX-NEXT:    vpmovb2m %ymm1, %k1
 ; SKX-NEXT:    vpmovzxbw {{.*#+}} zmm0 {%k1} {z} = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; SKX-NEXT:    retq

 ; SKX-NEXT:    retq
   %x   = sext <32 x i8> %a to <32 x i16>
   ret <32 x i16> %x
 }
 
 define <32 x i16> @sext_32x8_to_32x16_mask(<32 x i8> %a ,<32 x i1> %mask) nounwind readnone {
 ; KNL-LABEL: sext_32x8_to_32x16_mask:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vpmovsxbw %xmm2, %ymm2
-; KNL-NEXT:    vpmovsxbw %xmm0, %ymm0
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; KNL-NEXT:    vpsllw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpsraw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpand %ymm0, %ymm3, %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; KNL-NEXT:    vpmovsxbw %xmm3, %ymm3
+; KNL-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; KNL-NEXT:    vpsllw $15, %ymm1, %ymm1
 ; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; KNL-NEXT:    vpsllw $15, %ymm2, %ymm1
+; KNL-NEXT:    vpsraw $15, %ymm1, %ymm1
+; KNL-NEXT:    vpand %ymm3, %ymm1, %ymm1
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: sext_32x8_to_32x16_mask:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm1, %ymm1
 ; SKX-NEXT:    vpmovb2m %ymm1, %k1
 ; SKX-NEXT:    vpmovsxbw %ymm0, %zmm0 {%k1} {z}
 ; SKX-NEXT:    retq

test/CodeGen/X86/avx512-insert-extract.ll

 ; CHECK-NEXT:    retq
   %r = insertelement <32 x i8> %x, i8 %y, i32 20
   ret <32 x i8> %r
 }
 
 define i32 @test_insertelement_v32i1(i32 %a, i32 %b, <32 x i32> %x , <32 x i32> %y) {
 ; KNL-LABEL: test_insertelement_v32i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    .cfi_def_cfa_offset 16
-; KNL-NEXT:    .cfi_offset %rbp, -16
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    .cfi_def_cfa_register %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $32, %rsp
-; KNL-NEXT:    xorl %eax, %eax
 ; KNL-NEXT:    cmpl %esi, %edi
 ; KNL-NEXT:    setb %al
-; KNL-NEXT:    vpcmpltud %zmm2, %zmm0, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpcmpltud %zmm3, %zmm1, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
-; KNL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    movl (%rsp), %eax
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
+; KNL-NEXT:    vpcmpltud %zmm3, %zmm1, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    vpcmpltud %zmm2, %zmm0, %k0
+; KNL-NEXT:    kshiftrw $4, %k0, %k1
+; KNL-NEXT:    kmovw %eax, %k2
+; KNL-NEXT:    kxorw %k2, %k1, %k1
+; KNL-NEXT:    kshiftlw $15, %k1, %k1
+; KNL-NEXT:    kshiftrw $11, %k1, %k1
+; KNL-NEXT:    kxorw %k0, %k1, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    orl %ecx, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_insertelement_v32i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    cmpl %esi, %edi
 ; SKX-NEXT:    setb %al
 ; SKX-NEXT:    vpcmpltud %zmm2, %zmm0, %k0

 
 define zeroext i8 @test_extractelement_v32i1(<32 x i8> %a, <32 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v32i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpextrb $2, %xmm0, %eax
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    kshiftrw $2, %k0, %k0
+; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andl $1, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_extractelement_v32i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpcmpnleub %ymm1, %ymm0, %k0
 ; SKX-NEXT:    kshiftrd $2, %k0, %k0

 define zeroext i8 @test_extractelement_v64i1(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v64i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm0, %ymm3, %ymm2
 ; KNL-NEXT:    vpxor %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpextrb $15, %xmm0, %eax
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_extractelement_v64i1:

 define zeroext i8 @extractelement_v64i1_alt(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: extractelement_v64i1_alt:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm0, %ymm3, %ymm2
 ; KNL-NEXT:    vpxor %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpextrb $15, %xmm0, %eax
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: extractelement_v64i1_alt:
 ; SKX:       ## %bb.0:

 ; KNL-LABEL: test_insertelement_variable_v32i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $96, %rsp
+; KNL-NEXT:    subq $64, %rsp
 ; KNL-NEXT:    ## kill: def %esi killed %esi def %rsi
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; KNL-NEXT:    andl $31, %esi
 ; KNL-NEXT:    testb %dil, %dil
-; KNL-NEXT:    vmovdqa %ymm0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    setne 32(%rsp,%rsi)
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
+; KNL-NEXT:    setne (%rsp,%rsi)
+; KNL-NEXT:    vpmovsxbd (%rsp), %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vpmovsxbd {{[0-9]+}}(%rsp), %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    movl (%rsp), %eax
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %ecx, %eax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_insertelement_variable_v32i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    pushq %rbp

 ; KNL-LABEL: test_insertelement_variable_v64i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-64, %rsp
-; KNL-NEXT:    subq $192, %rsp
+; KNL-NEXT:    subq $128, %rsp
 ; KNL-NEXT:    ## kill: def %esi killed %esi def %rsi
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
 ; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; KNL-NEXT:    vpcmpgtb %ymm2, %ymm1, %ymm1
 ; KNL-NEXT:    andl $63, %esi
 ; KNL-NEXT:    testb %dil, %dil
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vmovdqa %ymm0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    setne 64(%rsp,%rsi)
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm0
+; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
+; KNL-NEXT:    setne (%rsp,%rsi)
+; KNL-NEXT:    vmovdqa (%rsp), %ymm0
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm2
 ; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
 ; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpmovsxbd %xmm1, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm0
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm0
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; KNL-NEXT:    movl (%rsp), %eax
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %edx, %eax
 ; KNL-NEXT:    shlq $32, %rax
 ; KNL-NEXT:    orq %rcx, %rax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_insertelement_variable_v64i1:

 ; KNL-LABEL: test_insertelement_variable_v96i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-128, %rsp
-; KNL-NEXT:    subq $384, %rsp ## imm = 0x180
+; KNL-NEXT:    subq $256, %rsp ## imm = 0x100
 ; KNL-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; KNL-NEXT:    vpinsrb $1, 488(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $2, 496(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $3, 504(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $4, 512(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $5, 520(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $6, 528(%rbp), %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $7, 536(%rbp), %xmm0, %xmm0

 ; KNL-NEXT:    vpcmpgtb %ymm3, %ymm0, %ymm0
 ; KNL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; KNL-NEXT:    movl 744(%rbp), %eax
 ; KNL-NEXT:    andl $127, %eax
 ; KNL-NEXT:    cmpb $0, 736(%rbp)
 ; KNL-NEXT:    vmovdqa %ymm3, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vmovdqa %ymm2, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    setne 128(%rsp,%rax)
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm2
+; KNL-NEXT:    vmovdqa %ymm2, (%rsp)
+; KNL-NEXT:    setne (%rsp,%rax)
+; KNL-NEXT:    vmovdqa (%rsp), %ymm2
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm3
+; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm4
-; KNL-NEXT:    vpmovsxbd %xmm4, %zmm4
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm4
 ; KNL-NEXT:    vpslld $31, %zmm4, %zmm4
 ; KNL-NEXT:    vptestmd %zmm4, %zmm4, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm2, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm3, %xmm1
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm2
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpmovsxbd %xmm3, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    vextracti128 $1, %ymm3, %xmm2
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %edx, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    orq %rcx, %rax
+; KNL-NEXT:    vpmovsxbd %xmm1, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm3, %zmm1
+; KNL-NEXT:    kmovw %k0, %esi
+; KNL-NEXT:    shll $16, %esi
+; KNL-NEXT:    orl %ecx, %esi
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; KNL-NEXT:    shlq $32, %rax
-; KNL-NEXT:    orq %rcx, %rax
-; KNL-NEXT:    movl (%rsp), %ecx
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %edx
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    shll $16, %edx
+; KNL-NEXT:    orl %ecx, %edx
 ; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
+; KNL-NEXT:    orq %rsi, %rdx
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_insertelement_variable_v96i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    pushq %rbp

 ; KNL-LABEL: test_insertelement_variable_v128i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-128, %rsp
-; KNL-NEXT:    subq $384, %rsp ## imm = 0x180
+; KNL-NEXT:    subq $256, %rsp ## imm = 0x100
 ; KNL-NEXT:    ## kill: def %esi killed %esi def %rsi
 ; KNL-NEXT:    vmovdqa {{.*#+}} ymm4 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; KNL-NEXT:    vpxor %ymm4, %ymm0, %ymm0
 ; KNL-NEXT:    vpcmpgtb %ymm4, %ymm0, %ymm0
 ; KNL-NEXT:    vpxor %ymm4, %ymm1, %ymm1
 ; KNL-NEXT:    vpcmpgtb %ymm4, %ymm1, %ymm1
 ; KNL-NEXT:    vpxor %ymm4, %ymm2, %ymm2
 ; KNL-NEXT:    vpcmpgtb %ymm4, %ymm2, %ymm2
 ; KNL-NEXT:    vpxor %ymm4, %ymm3, %ymm3
 ; KNL-NEXT:    vpcmpgtb %ymm4, %ymm3, %ymm3
 ; KNL-NEXT:    andl $127, %esi
 ; KNL-NEXT:    testb %dil, %dil
 ; KNL-NEXT:    vmovdqa %ymm3, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm2, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vmovdqa %ymm0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    setne 128(%rsp,%rsi)
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
-; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm2
+; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
+; KNL-NEXT:    setne (%rsp,%rsi)
+; KNL-NEXT:    vmovdqa (%rsp), %ymm2
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm3
+; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm4
-; KNL-NEXT:    vpmovsxbd %xmm4, %zmm4
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm4
 ; KNL-NEXT:    vpslld $31, %zmm4, %zmm4
 ; KNL-NEXT:    vptestmd %zmm4, %zmm4, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm2, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm3, %xmm1
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm2
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpmovsxbd %xmm3, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    vextracti128 $1, %ymm3, %xmm2
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %edx, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    orq %rcx, %rax
+; KNL-NEXT:    vpmovsxbd %xmm1, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm3, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
+; KNL-NEXT:    kmovw %k0, %esi
+; KNL-NEXT:    shll $16, %esi
+; KNL-NEXT:    orl %ecx, %esi
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; KNL-NEXT:    shlq $32, %rax
-; KNL-NEXT:    orq %rcx, %rax
-; KNL-NEXT:    movl (%rsp), %ecx
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %edx
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    shll $16, %edx
+; KNL-NEXT:    orl %ecx, %edx
 ; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
+; KNL-NEXT:    orq %rsi, %rdx
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_insertelement_variable_v128i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    pushq %rbp

test/CodeGen/X86/avx512-mask-op.ll

; AVX512DQ-NEXT:    retq
   %c = select i1 %mask, <16 x i1> %a, <16 x i1> %b
   ret <16 x i1> %c
 }
 
 define <64 x i8> @test16(i64 %x) {
 ;
 ; KNL-LABEL: test16:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    .cfi_def_cfa_offset 16
-; KNL-NEXT:    .cfi_offset %rbp, -16
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    .cfi_def_cfa_register %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $64, %rsp
-; KNL-NEXT:    movl %edi, (%rsp)
+; KNL-NEXT:    movq %rdi, %rax
+; KNL-NEXT:    movl %edi, %ecx
+; KNL-NEXT:    kmovw %edi, %k0
 ; KNL-NEXT:    shrq $32, %rdi
-; KNL-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    kmovw (%rsp), %k1
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; KNL-NEXT:    shrq $48, %rax
+; KNL-NEXT:    shrl $16, %ecx
+; KNL-NEXT:    kmovw %ecx, %k1
+; KNL-NEXT:    kmovw %eax, %k2
+; KNL-NEXT:    kmovw %edi, %k3
+; KNL-NEXT:    movb $1, %al
+; KNL-NEXT:    kmovw %eax, %k4
+; KNL-NEXT:    kshiftrw $5, %k0, %k5
+; KNL-NEXT:    kxorw %k4, %k5, %k4
+; KNL-NEXT:    kshiftlw $15, %k4, %k4
+; KNL-NEXT:    kshiftrw $10, %k4, %k4
+; KNL-NEXT:    kxorw %k0, %k4, %k4
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k3} {z}
 ; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    movl $1, %eax
-; KNL-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; KNL-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k2} {z}
 ; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
+; KNL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k4} {z}
+; KNL-NEXT:    vpmovdb %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; KNL-NEXT:    vpmovdb %zmm2, %xmm2
-; KNL-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
-; KNL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
+; KNL-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test16:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    kmovq %rdi, %k0
 ; SKX-NEXT:    movb $1, %al
 ; SKX-NEXT:    kmovd %eax, %k1
 ; SKX-NEXT:    kshiftrq $5, %k0, %k2

 ; AVX512BW-NEXT:    kshiftlq $63, %k1, %k1
 ; AVX512BW-NEXT:    kshiftrq $58, %k1, %k1
 ; AVX512BW-NEXT:    kxorq %k0, %k1, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test16:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    pushq %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_offset 16
-; AVX512DQ-NEXT:    .cfi_offset %rbp, -16
-; AVX512DQ-NEXT:    movq %rsp, %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512DQ-NEXT:    andq $-32, %rsp
-; AVX512DQ-NEXT:    subq $64, %rsp
-; AVX512DQ-NEXT:    movl %edi, (%rsp)
+; AVX512DQ-NEXT:    movq %rdi, %rax
+; AVX512DQ-NEXT:    movl %edi, %ecx
+; AVX512DQ-NEXT:    kmovw %edi, %k0
 ; AVX512DQ-NEXT:    shrq $32, %rdi
-; AVX512DQ-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; AVX512DQ-NEXT:    kmovw (%rsp), %k0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
+; AVX512DQ-NEXT:    shrq $48, %rax
+; AVX512DQ-NEXT:    shrl $16, %ecx
+; AVX512DQ-NEXT:    kmovw %ecx, %k1
+; AVX512DQ-NEXT:    kmovw %eax, %k2
+; AVX512DQ-NEXT:    kmovw %edi, %k3
+; AVX512DQ-NEXT:    movb $1, %al
+; AVX512DQ-NEXT:    kmovw %eax, %k4
+; AVX512DQ-NEXT:    kshiftrw $5, %k0, %k5
+; AVX512DQ-NEXT:    kxorw %k4, %k5, %k4
+; AVX512DQ-NEXT:    kshiftlw $15, %k4, %k4
+; AVX512DQ-NEXT:    kshiftrw $10, %k4, %k4
+; AVX512DQ-NEXT:    kxorw %k0, %k4, %k0
+; AVX512DQ-NEXT:    vpmovm2d %k3, %zmm0
 ; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
-; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512DQ-NEXT:    movl $1, %eax
-; AVX512DQ-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
+; AVX512DQ-NEXT:    vpmovm2d %k2, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm2
 ; AVX512DQ-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512DQ-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; AVX512DQ-NEXT:    movq %rbp, %rsp
-; AVX512DQ-NEXT:    popq %rbp
+; AVX512DQ-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    retq
   %a = bitcast i64 %x to <64 x i1>
   %b = insertelement <64 x i1>%a, i1 true, i32 5
   %c = sext <64 x i1>%b to <64 x i8>
   ret <64 x i8>%c
 }
 
 define <64 x i8> @test17(i64 %x, i32 %y, i32 %z) {
 ;
 ; KNL-LABEL: test17:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    .cfi_def_cfa_offset 16
-; KNL-NEXT:    .cfi_offset %rbp, -16
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    .cfi_def_cfa_register %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $64, %rsp
-; KNL-NEXT:    movl %edi, (%rsp)
+; KNL-NEXT:    movq %rdi, %rax
+; KNL-NEXT:    movl %edi, %ecx
+; KNL-NEXT:    kmovw %edi, %k0
 ; KNL-NEXT:    shrq $32, %rdi
-; KNL-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    kmovw (%rsp), %k1
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    xorl %eax, %eax
+; KNL-NEXT:    shrq $48, %rax
+; KNL-NEXT:    shrl $16, %ecx
+; KNL-NEXT:    kmovw %ecx, %k1
+; KNL-NEXT:    kmovw %eax, %k2
+; KNL-NEXT:    kmovw %edi, %k3
 ; KNL-NEXT:    cmpl %edx, %esi
 ; KNL-NEXT:    setg %al
-; KNL-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; KNL-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; KNL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; KNL-NEXT:    kmovw %eax, %k4
+; KNL-NEXT:    kshiftrw $5, %k0, %k5
+; KNL-NEXT:    kxorw %k4, %k5, %k4
+; KNL-NEXT:    kshiftlw $15, %k4, %k4
+; KNL-NEXT:    kshiftrw $10, %k4, %k4
+; KNL-NEXT:    kxorw %k0, %k4, %k4
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k3} {z}
+; KNL-NEXT:    vpmovdb %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k2} {z}
 ; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
+; KNL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k4} {z}
+; KNL-NEXT:    vpmovdb %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; KNL-NEXT:    vpmovdb %zmm2, %xmm2
-; KNL-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
-; KNL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
+; KNL-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test17:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    kmovq %rdi, %k0
 ; SKX-NEXT:    cmpl %edx, %esi
 ; SKX-NEXT:    setg %al
 ; SKX-NEXT:    kmovd %eax, %k1

 ; AVX512BW-NEXT:    kshiftlq $63, %k1, %k1
 ; AVX512BW-NEXT:    kshiftrq $58, %k1, %k1
 ; AVX512BW-NEXT:    kxorq %k0, %k1, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test17:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    pushq %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_offset 16
-; AVX512DQ-NEXT:    .cfi_offset %rbp, -16
-; AVX512DQ-NEXT:    movq %rsp, %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512DQ-NEXT:    andq $-32, %rsp
-; AVX512DQ-NEXT:    subq $64, %rsp
-; AVX512DQ-NEXT:    movl %edi, (%rsp)
+; AVX512DQ-NEXT:    movq %rdi, %rax
+; AVX512DQ-NEXT:    movl %edi, %ecx
+; AVX512DQ-NEXT:    kmovw %edi, %k0
 ; AVX512DQ-NEXT:    shrq $32, %rdi
-; AVX512DQ-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; AVX512DQ-NEXT:    kmovw (%rsp), %k0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
-; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512DQ-NEXT:    xorl %eax, %eax
+; AVX512DQ-NEXT:    shrq $48, %rax
+; AVX512DQ-NEXT:    shrl $16, %ecx
+; AVX512DQ-NEXT:    kmovw %ecx, %k1
+; AVX512DQ-NEXT:    kmovw %eax, %k2
+; AVX512DQ-NEXT:    kmovw %edi, %k3
 ; AVX512DQ-NEXT:    cmpl %edx, %esi
 ; AVX512DQ-NEXT:    setg %al
-; AVX512DQ-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k0
-; AVX512DQ-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
+; AVX512DQ-NEXT:    kmovw %eax, %k4
+; AVX512DQ-NEXT:    kshiftrw $5, %k0, %k5
+; AVX512DQ-NEXT:    kxorw %k4, %k5, %k4
+; AVX512DQ-NEXT:    kshiftlw $15, %k4, %k4
+; AVX512DQ-NEXT:    kshiftrw $10, %k4, %k4
+; AVX512DQ-NEXT:    kxorw %k0, %k4, %k0
+; AVX512DQ-NEXT:    vpmovm2d %k3, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovm2d %k2, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm2
 ; AVX512DQ-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512DQ-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; AVX512DQ-NEXT:    movq %rbp, %rsp
-; AVX512DQ-NEXT:    popq %rbp
+; AVX512DQ-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    retq
   %a = bitcast i64 %x to <64 x i1>
   %b = icmp sgt i32 %y, %z
   %c = insertelement <64 x i1>%a, i1 %b, i32 5
   %d = sext <64 x i1>%c to <64 x i8>
   ret <64 x i8>%d
 }
 
; AVX512DQ-NEXT:    retq
   %el2 = extractelement <16 x i1>%b1, i32 9
   %c = insertelement <8 x i1>%b, i1 %el1, i32 7
   %d = insertelement <8 x i1>%c, i1 %el2, i32 6
   ret <8 x i1>%d
 }
 define <32 x i16> @test21(<32 x i16> %x , <32 x i1> %mask) nounwind readnone {
 ; KNL-LABEL: test21:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; KNL-NEXT:    vpsllw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpsraw $15, %ymm3, %ymm3
-; KNL-NEXT:    vpand %ymm0, %ymm3, %ymm0
-; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm2
+; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; KNL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; KNL-NEXT:    vpsllw $15, %ymm2, %ymm2
 ; KNL-NEXT:    vpsraw $15, %ymm2, %ymm2
+; KNL-NEXT:    vpand %ymm0, %ymm2, %ymm0
+; KNL-NEXT:    vpsllw $15, %ymm3, %ymm2
+; KNL-NEXT:    vpsraw $15, %ymm2, %ymm2
 ; KNL-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test21:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm1, %ymm1
 ; SKX-NEXT:    vpmovb2m %ymm1, %k1
 ; SKX-NEXT:    vmovdqu16 %zmm0, %zmm0 {%k1} {z}
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test21:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm0 {%k1} {z}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test21:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512DQ-NEXT:    vpsllw $15, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpsraw $15, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpand %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm2, %xmm2
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512DQ-NEXT:    vpsllw $15, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpsraw $15, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpand %ymm0, %ymm2, %ymm0
+; AVX512DQ-NEXT:    vpsllw $15, %ymm3, %ymm2
+; AVX512DQ-NEXT:    vpsraw $15, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; AVX512DQ-NEXT:    retq
   %ret = select <32 x i1> %mask, <32 x i16> %x, <32 x i16> zeroinitializer
   ret <32 x i16> %ret
 }
 
 define void @test22(<4 x i1> %a, <4 x i1>* %addr) {
 ; KNL-LABEL: test22:

 ; AVX512DQ-NEXT:    retq
   %ret = select <32 x i1> <i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 true, i1 false>, <32 x i16> %x, <32 x i16> zeroinitializer
   ret <32 x i16> %ret
 }
 
 define <64 x i8> @test_build_vec_v64i1(<64 x i8> %x) {
 ; KNL-LABEL: test_build_vec_v64i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm0
-; KNL-NEXT:    vandps {{.*}}(%rip), %ymm1, %ymm1
+; KNL-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; KNL-NEXT:    vmovaps {{.*#+}} xmm3 = [255,0,255,0,0,0,255,0,255,0,0,255,255,0,255,0]
+; KNL-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; KNL-NEXT:    vandps %xmm3, %xmm1, %xmm1
+; KNL-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; KNL-NEXT:    vandps {{.*}}(%rip), %xmm0, %xmm2
+; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; KNL-NEXT:    vandps {{.*}}(%rip), %xmm0, %xmm0
+; KNL-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_build_vec_v64i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpshufb {{.*#+}} zmm0 = zero,zero,zmm0[2],zero,zero,zero,zmm0[6],zero,zmm0[8],zero,zmm0[10],zero,zmm0[12],zero,zero,zmm0[15],zero,zero,zmm0[18],zero,zmm0[20],zero,zmm0[22],zero,zmm0[24],zero,zero,zmm0[27],zero,zero,zmm0[30],zero,zmm0[32],zero,zmm0[34],zero,zero,zero,zmm0[38],zero,zmm0[40],zero,zero,zmm0[43,44],zero,zmm0[46],zero,zmm0[48],zero,zmm0[50],zero,zero,zero,zmm0[54],zero,zmm0[56],zero,zero,zmm0[59,60],zero,zmm0[62],zero
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_build_vec_v64i1:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm0 = zero,zero,zmm0[2],zero,zero,zero,zmm0[6],zero,zmm0[8],zero,zmm0[10],zero,zmm0[12],zero,zero,zmm0[15],zero,zero,zmm0[18],zero,zmm0[20],zero,zmm0[22],zero,zmm0[24],zero,zero,zmm0[27],zero,zero,zmm0[30],zero,zmm0[32],zero,zmm0[34],zero,zero,zero,zmm0[38],zero,zmm0[40],zero,zero,zmm0[43,44],zero,zmm0[46],zero,zmm0[48],zero,zmm0[50],zero,zero,zero,zmm0[54],zero,zmm0[56],zero,zero,zmm0[59,60],zero,zmm0[62],zero
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_build_vec_v64i1:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %ymm1, %ymm1
+; AVX512DQ-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX512DQ-NEXT:    vmovaps {{.*#+}} xmm3 = [255,0,255,0,0,0,255,0,255,0,0,255,255,0,255,0]
+; AVX512DQ-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX512DQ-NEXT:    vandps %xmm3, %xmm1, %xmm1
+; AVX512DQ-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %xmm0, %xmm2
+; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %xmm0, %xmm0
+; AVX512DQ-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
 ; AVX512DQ-NEXT:    retq
   %ret = select <64 x i1> <i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 true, i1 false>, <64 x i8> %x, <64 x i8> zeroinitializer
   ret <64 x i8> %ret
 }
 
 define void @ktest_1(<8 x double> %in, double * %base) {
 ; KNL-LABEL: ktest_1:
 ; KNL:       ## %bb.0:

 End:
   ret void
 }
 
 define void @ktest_2(<32 x float> %in, float * %base) {
 ;
 ; KNL-LABEL: ktest_2:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    .cfi_def_cfa_offset 16
-; KNL-NEXT:    .cfi_offset %rbp, -16
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    .cfi_def_cfa_register %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $32, %rsp
 ; KNL-NEXT:    vmovups (%rdi), %zmm2
 ; KNL-NEXT:    vmovups 64(%rdi), %zmm3
-; KNL-NEXT:    vcmpltps %zmm0, %zmm2, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm2, %xmm2
-; KNL-NEXT:    vcmpltps %zmm1, %zmm3, %k2
-; KNL-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; KNL-NEXT:    vpmovdb %zmm3, %xmm3
-; KNL-NEXT:    vmovups 68(%rdi), %zmm4 {%k2} {z}
-; KNL-NEXT:    vmovups 4(%rdi), %zmm5 {%k1} {z}
-; KNL-NEXT:    vcmpltps %zmm5, %zmm0, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm5, %xmm5
-; KNL-NEXT:    vpor %xmm5, %xmm2, %xmm2
-; KNL-NEXT:    vcmpltps %zmm4, %zmm1, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm4, %xmm4
-; KNL-NEXT:    vpor %xmm4, %xmm3, %xmm3
-; KNL-NEXT:    vpmovsxbd %xmm3, %zmm3
-; KNL-NEXT:    vpslld $31, %zmm3, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
-; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
-; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    cmpl $0, (%rsp)
+; KNL-NEXT:    vcmpltps %zmm1, %zmm3, %k1
+; KNL-NEXT:    vcmpltps %zmm0, %zmm2, %k2
+; KNL-NEXT:    vmovups 4(%rdi), %zmm2 {%k2} {z}
+; KNL-NEXT:    vmovups 68(%rdi), %zmm3 {%k1} {z}
+; KNL-NEXT:    vcmpltps %zmm3, %zmm1, %k0
+; KNL-NEXT:    vcmpltps %zmm2, %zmm0, %k3
+; KNL-NEXT:    korw %k3, %k2, %k2
+; KNL-NEXT:    kmovw %k2, %eax
+; KNL-NEXT:    korw %k0, %k1, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    orl %eax, %ecx
 ; KNL-NEXT:    je LBB42_2
 ; KNL-NEXT:  ## %bb.1: ## %L1
 ; KNL-NEXT:    vmovaps %zmm0, (%rdi)
 ; KNL-NEXT:    vmovaps %zmm1, 64(%rdi)
-; KNL-NEXT:    jmp LBB42_3
+; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    retq
 ; KNL-NEXT:  LBB42_2: ## %L2
 ; KNL-NEXT:    vmovaps %zmm0, 4(%rdi)
 ; KNL-NEXT:    vmovaps %zmm1, 68(%rdi)
-; KNL-NEXT:  LBB42_3: ## %End
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: ktest_2:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vmovups (%rdi), %zmm2
 ; SKX-NEXT:    vmovups 64(%rdi), %zmm3
 ; SKX-NEXT:    vcmpltps %zmm0, %zmm2, %k1

 ; AVX512BW-NEXT:  LBB42_2: ## %L2
 ; AVX512BW-NEXT:    vmovaps %zmm0, 4(%rdi)
 ; AVX512BW-NEXT:    vmovaps %zmm1, 68(%rdi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: ktest_2:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    pushq %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_offset 16
-; AVX512DQ-NEXT:    .cfi_offset %rbp, -16
-; AVX512DQ-NEXT:    movq %rsp, %rbp
-; AVX512DQ-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512DQ-NEXT:    andq $-32, %rsp
-; AVX512DQ-NEXT:    subq $32, %rsp
 ; AVX512DQ-NEXT:    vmovups (%rdi), %zmm2
 ; AVX512DQ-NEXT:    vmovups 64(%rdi), %zmm3
-; AVX512DQ-NEXT:    vcmpltps %zmm0, %zmm2, %k1
-; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm2
-; AVX512DQ-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512DQ-NEXT:    vcmpltps %zmm1, %zmm3, %k2
-; AVX512DQ-NEXT:    vpmovm2d %k2, %zmm3
-; AVX512DQ-NEXT:    vpmovdb %zmm3, %xmm3
-; AVX512DQ-NEXT:    vmovups 68(%rdi), %zmm4 {%k2} {z}
-; AVX512DQ-NEXT:    vmovups 4(%rdi), %zmm5 {%k1} {z}
-; AVX512DQ-NEXT:    vcmpltps %zmm5, %zmm0, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm5
-; AVX512DQ-NEXT:    vpmovdb %zmm5, %xmm5
-; AVX512DQ-NEXT:    vpor %xmm5, %xmm2, %xmm2
-; AVX512DQ-NEXT:    vcmpltps %zmm4, %zmm1, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm4
-; AVX512DQ-NEXT:    vpmovdb %zmm4, %xmm4
-; AVX512DQ-NEXT:    vpor %xmm4, %xmm3, %xmm3
-; AVX512DQ-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512DQ-NEXT:    vpslld $31, %zmm3, %zmm3
-; AVX512DQ-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512DQ-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512DQ-NEXT:    vpslld $31, %zmm2, %zmm2
-; AVX512DQ-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512DQ-NEXT:    kmovw %k0, (%rsp)
-; AVX512DQ-NEXT:    cmpl $0, (%rsp)
+; AVX512DQ-NEXT:    vcmpltps %zmm1, %zmm3, %k1
+; AVX512DQ-NEXT:    vcmpltps %zmm0, %zmm2, %k2
+; AVX512DQ-NEXT:    vmovups 4(%rdi), %zmm2 {%k2} {z}
+; AVX512DQ-NEXT:    vmovups 68(%rdi), %zmm3 {%k1} {z}
+; AVX512DQ-NEXT:    vcmpltps %zmm3, %zmm1, %k0
+; AVX512DQ-NEXT:    vcmpltps %zmm2, %zmm0, %k3
+; AVX512DQ-NEXT:    korw %k3, %k2, %k2
+; AVX512DQ-NEXT:    kmovw %k2, %eax
+; AVX512DQ-NEXT:    korw %k0, %k1, %k0
+; AVX512DQ-NEXT:    kmovw %k0, %ecx
+; AVX512DQ-NEXT:    shll $16, %ecx
+; AVX512DQ-NEXT:    orl %eax, %ecx
 ; AVX512DQ-NEXT:    je LBB42_2
 ; AVX512DQ-NEXT:  ## %bb.1: ## %L1
 ; AVX512DQ-NEXT:    vmovaps %zmm0, (%rdi)
 ; AVX512DQ-NEXT:    vmovaps %zmm1, 64(%rdi)
-; AVX512DQ-NEXT:    jmp LBB42_3
+; AVX512DQ-NEXT:    vzeroupper
+; AVX512DQ-NEXT:    retq
 ; AVX512DQ-NEXT:  LBB42_2: ## %L2
 ; AVX512DQ-NEXT:    vmovaps %zmm0, 4(%rdi)
 ; AVX512DQ-NEXT:    vmovaps %zmm1, 68(%rdi)
-; AVX512DQ-NEXT:  LBB42_3: ## %End
-; AVX512DQ-NEXT:    movq %rbp, %rsp
-; AVX512DQ-NEXT:    popq %rbp
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
   %addr1 = getelementptr float, float * %base, i64 0
   %addr2 = getelementptr float, float * %base, i64 1
 
   %vaddr1 = bitcast float* %addr1 to <32 x float>*
   %vaddr2 = bitcast float* %addr2 to <32 x float>*
 

 ; AVX512DQ-NEXT:    retq
   store <16 x i1> %v, <16 x i1>* %a
   ret void
 }
 
 define void @store_32i1(<32 x i1>* %a, <32 x i1> %v) {
 ; KNL-LABEL: store_32i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
+; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, 2(%rdi)
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; KNL-NEXT:    kmovw %k1, 2(%rdi)
 ; KNL-NEXT:    kmovw %k0, (%rdi)
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: store_32i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovb2m %ymm0, %k0
 ; SKX-NEXT:    kmovd %k0, (%rdi)
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_32i1:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, (%rdi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: store_32i1:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; AVX512DQ-NEXT:    vpslld $31, %zmm1, %zmm1
 ; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 2(%rdi)
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512DQ-NEXT:    kmovw %k1, 2(%rdi)
 ; AVX512DQ-NEXT:    kmovw %k0, (%rdi)
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
   store <32 x i1> %v, <32 x i1>* %a
   ret void
 }
 
 define void @store_32i1_1(<32 x i1>* %a, <32 x i16> %v) {
 ; KNL-LABEL: store_32i1_1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vpmovsxwd %ymm0, %zmm0
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpmovsxwd %ymm1, %zmm1
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, 2(%rdi)
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    vpmovsxwd %ymm1, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; KNL-NEXT:    kmovw %k1, 2(%rdi)
 ; KNL-NEXT:    kmovw %k0, (%rdi)
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: store_32i1_1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpsllw $15, %zmm0, %zmm0
 ; SKX-NEXT:    vpmovw2m %zmm0, %k0
 ; SKX-NEXT:    kmovd %k0, (%rdi)
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_32i1_1:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpsllw $15, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, (%rdi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: store_32i1_1:
 ; AVX512DQ:       ## %bb.0:
 ; AVX512DQ-NEXT:    vpmovsxwd %ymm0, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512DQ-NEXT:    vpmovsxwd %ymm1, %zmm1
-; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512DQ-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 2(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512DQ-NEXT:    vpmovsxwd %ymm1, %zmm0
+; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512DQ-NEXT:    kmovw %k1, 2(%rdi)
 ; AVX512DQ-NEXT:    kmovw %k0, (%rdi)
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
   %v1 = trunc <32 x i16> %v to <32 x i1>
   store <32 x i1> %v1, <32 x i1>* %a
   ret void
 }
 
 
 define void @store_64i1(<64 x i1>* %a, <64 x i1> %v) {
 ;
 ; KNL-LABEL: store_64i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vpmovsxbd %xmm3, %zmm3
-; KNL-NEXT:    vpslld $31, %zmm3, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; KNL-NEXT:    kmovw %k0, 6(%rdi)
-; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
-; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
-; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; KNL-NEXT:    kmovw %k0, 4(%rdi)
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, 2(%rdi)
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    vpmovsxbd %xmm1, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; KNL-NEXT:    vpmovsxbd %xmm3, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k3
+; KNL-NEXT:    kmovw %k3, 6(%rdi)
+; KNL-NEXT:    kmovw %k2, 4(%rdi)
+; KNL-NEXT:    kmovw %k1, 2(%rdi)
 ; KNL-NEXT:    kmovw %k0, (%rdi)
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: store_64i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; SKX-NEXT:    vpmovb2m %zmm0, %k0
 ; SKX-NEXT:    kmovq %k0, (%rdi)
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_64i1:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
 ; AVX512BW-NEXT:    kmovq %k0, (%rdi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: store_64i1:
 ; AVX512DQ:       ## %bb.0:
-; AVX512DQ-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512DQ-NEXT:    vpslld $31, %zmm3, %zmm3
-; AVX512DQ-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 6(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512DQ-NEXT:    vpslld $31, %zmm2, %zmm2
-; AVX512DQ-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 4(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512DQ-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 2(%rdi)
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm0
+; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512DQ-NEXT:    vpmovsxbd %xmm2, %zmm0
+; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; AVX512DQ-NEXT:    vpmovsxbd %xmm3, %zmm0
+; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k3
+; AVX512DQ-NEXT:    kmovw %k3, 6(%rdi)
+; AVX512DQ-NEXT:    kmovw %k2, 4(%rdi)
+; AVX512DQ-NEXT:    kmovw %k1, 2(%rdi)
 ; AVX512DQ-NEXT:    kmovw %k0, (%rdi)
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
   store <64 x i1> %v, <64 x i1>* %a
   ret void
 }
 
 define i32 @test_bitcast_v8i1_zext(<16 x i32> %a) {

test/CodeGen/X86/avx512-masked-memop-64-32.ll

; SKX-NEXT:    retq
   %res = call <16 x double> @llvm.masked.load.v16f64.p0v16f64(<16 x double>* %ptrs, i32 4, <16 x i1> %mask, <16 x double> %src0)
   ret <16 x double> %res
 }
 declare <16 x double> @llvm.masked.load.v16f64.p0v16f64(<16 x double>* %ptrs, i32, <16 x i1> %mask, <16 x double> %src0)
 
 define <32 x double> @test_load_32f64(<32 x double>* %ptrs, <32 x i1> %mask, <32 x double> %src0)  {
 ; AVX512F-LABEL: test_load_32f64:
 ; AVX512F:       ## %bb.0:
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm5
-; AVX512F-NEXT:    vpmovsxbd %xmm5, %zmm5
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm5
 ; AVX512F-NEXT:    vpslld $31, %zmm5, %zmm5
 ; AVX512F-NEXT:    vptestmd %zmm5, %zmm5, %k1
-; AVX512F-NEXT:    vblendmpd 128(%rdi), %zmm3, %zmm5 {%k1}
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k2
-; AVX512F-NEXT:    vblendmpd (%rdi), %zmm1, %zmm0 {%k2}
+; AVX512F-NEXT:    vblendmpd 128(%rdi), %zmm3, %zmm5 {%k2}
+; AVX512F-NEXT:    vblendmpd (%rdi), %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT:    kshiftrw $8, %k2, %k2
+; AVX512F-NEXT:    vblendmpd 192(%rdi), %zmm4, %zmm3 {%k2}
 ; AVX512F-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512F-NEXT:    vblendmpd 192(%rdi), %zmm4, %zmm3 {%k1}
-; AVX512F-NEXT:    kshiftrw $8, %k2, %k1
 ; AVX512F-NEXT:    vblendmpd 64(%rdi), %zmm2, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovapd %zmm5, %zmm2
 ; AVX512F-NEXT:    retq
 ;
 ; SKX-LABEL: test_load_32f64:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovb2m %ymm0, %k1

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

; SKX-NEXT:    retq
   %res = icmp eq <16 x i64> %a, %b
   %res1 = bitcast <16 x i1> %res to i16
   ret i16 %res1
 }
 
 define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
 ; KNL-LABEL: test12_v32i32:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $32, %rsp
-; KNL-NEXT:    vpcmpeqd %zmm3, %zmm1, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    vpcmpeqd %zmm2, %zmm0, %k1
-; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    movl (%rsp), %eax
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
+; KNL-NEXT:    vpcmpeqd %zmm2, %zmm0, %k0
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    vpcmpeqd %zmm3, %zmm1, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %ecx, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test12_v32i32:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpcmpeqd %zmm2, %zmm0, %k0
 ; SKX-NEXT:    vpcmpeqd %zmm3, %zmm1, %k1
 ; SKX-NEXT:    kunpckwd %k0, %k1, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
   %res = icmp eq <32 x i32> %a, %b
   %res1 = bitcast <32 x i1> %res to i32
   ret i32 %res1
 }
 
 define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind {
 ; KNL-LABEL: test12_v64i16:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    pushq %rbp
-; KNL-NEXT:    movq %rsp, %rbp
-; KNL-NEXT:    andq $-32, %rsp
-; KNL-NEXT:    subq $64, %rsp
-; KNL-NEXT:    vpcmpeqw %ymm5, %ymm1, %ymm1
-; KNL-NEXT:    vpmovsxwd %ymm1, %zmm1
-; KNL-NEXT:    vpmovdb %zmm1, %xmm1
-; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; KNL-NEXT:    vpmovsxwd %ymm0, %zmm0
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, (%rsp)
-; KNL-NEXT:    vpcmpeqw %ymm7, %ymm3, %ymm0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    vpcmpeqw %ymm5, %ymm1, %ymm0
 ; KNL-NEXT:    vpmovsxwd %ymm0, %zmm0
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    shll $16, %ecx
+; KNL-NEXT:    orl %eax, %ecx
 ; KNL-NEXT:    vpcmpeqw %ymm6, %ymm2, %ymm0
 ; KNL-NEXT:    vpmovsxwd %ymm0, %zmm0
-; KNL-NEXT:    vpmovdb %zmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; KNL-NEXT:    movl (%rsp), %ecx
-; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    vpcmpeqw %ymm7, %ymm3, %ymm0
+; KNL-NEXT:    vpmovsxwd %ymm0, %zmm0
+; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    shll $16, %eax
+; KNL-NEXT:    orl %edx, %eax
 ; KNL-NEXT:    shlq $32, %rax
 ; KNL-NEXT:    orq %rcx, %rax
-; KNL-NEXT:    movq %rbp, %rsp
-; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test12_v64i16:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpcmpeqw %zmm2, %zmm0, %k0
 ; SKX-NEXT:    vpcmpeqw %zmm3, %zmm1, %k1
 ; SKX-NEXT:    kunpckdq %k0, %k1, %k0

test/CodeGen/X86/avx512vl-vec-masked-cmp.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -disable-peephole -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq | FileCheck %s --check-prefix=CHECK --check-prefix=VLX
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -disable-peephole -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=NoVLX
 
 define zeroext i32 @test_vpcmpeqb_v16i1_v32i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqb_v16i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqb (%rdi), %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqb_v16i1_v32i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqb_v16i1_v32i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqb (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpeqb_v16i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqb_v16i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb (%rdi), %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqb_v16i1_v64i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqb_v16i1_v64i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp eq <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp eq <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqb_v32i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqb (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqb (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp eq <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqb_v32i1_v64i1_mask(i32 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp eq <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqb_v32i1_v64i1_mask_mem(i32 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqb (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vpcmpeqb (%rsi), %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; NoVLX-NEXT:    vpmovsxbd %xmm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
-; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp eq <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpeqw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqw_v8i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rdi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqw_v8i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqw_v8i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpeqw_v8i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqw_v8i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rdi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v8i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v8i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp eq <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqw_v16i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqw_v16i1_v32i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqw_v16i1_v32i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqw_v16i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v16i1_v64i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v16i1_v64i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp eq <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vmovq %xmm3, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm3
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm2, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm5, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
+; NoVLX-NEXT:    vpcmpeqw %ymm3, %ymm1, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm3, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp eq <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqw_v32i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqw (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NoVLX-NEXT:    vmovq %xmm1, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm1
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpeqw 32(%rdi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpcmpeqw (%rdi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpcmpeqw 32(%rdi), %ymm1, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp eq <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v32i1_v64i1_mask(i32 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vmovq %xmm3, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm3
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm5, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vpcmpeqw %ymm3, %ymm1, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp eq <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqw_v32i1_v64i1_mask_mem(i32 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqw (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; NoVLX-NEXT:    vmovq %xmm1, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm1
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw 32(%rsi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqw (%rsi), %ymm0, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp eq <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqd_v4i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v4i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v4i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqd_v4i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v4i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v4i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqd_v8i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v8i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v8i1_v32i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqd_v8i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v8i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v8i1_v64i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqd_v16i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v16i1_v32i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v16i1_v32i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqd_v16i1_v32i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <16 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqd_v16i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v16i1_v64i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v16i1_v64i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpeqd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqd_v16i1_v64i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpeqd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <16 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqq_v2i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v2i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v2i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v2i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp eq <2 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqq_v2i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v2i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v2i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp eq <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v2i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp eq <2 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqq_v4i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v4i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqq_v4i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v4i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpeqq_v8i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v8i1_v32i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v8i1_v32i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpeqq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpeqq_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpeqq_v8i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpeqq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v8i1_v64i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v8i1_v64i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpeqq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpeqq_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpeqq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp eq <8 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpsgtb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtb_v16i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb (%rdi), %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtb_v16i1_v32i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtb_v16i1_v32i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpsgtb_v16i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtb_v16i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb (%rdi), %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtb_v16i1_v64i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtb_v16i1_v64i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sgt <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp sgt <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtb_v32i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtb (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp sgt <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtb_v32i1_v64i1_mask(i32 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp sgt <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtb_v32i1_v64i1_mask_mem(i32 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtb (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vpcmpgtb (%rsi), %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; NoVLX-NEXT:    vpmovsxbd %xmm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
-; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp sgt <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpsgtw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtw_v8i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rdi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtw_v8i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtw_v8i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpsgtw_v8i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtw_v8i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rdi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v8i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v8i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sgt <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtw_v16i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtw_v16i1_v32i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtw_v16i1_v32i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtw_v16i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rdi), %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v16i1_v64i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v16i1_v64i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sgt <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vmovq %xmm3, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm3
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm2, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm5, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
+; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp sgt <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtw_v32i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtw (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NoVLX-NEXT:    vmovq %xmm1, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm1
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw 32(%rdi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpcmpgtw (%rdi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpcmpgtw 32(%rdi), %ymm1, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp sgt <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v32i1_v64i1_mask(i32 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vmovq %xmm3, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm3
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm5, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp sgt <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtw_v32i1_v64i1_mask_mem(i32 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtw (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; NoVLX-NEXT:    vmovq %xmm1, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm1
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw 32(%rsi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtw (%rsi), %ymm0, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp sgt <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtd_v4i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v4i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtd_v4i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v4i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtd_v8i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v8i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtd_v8i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v8i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtd_v16i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v16i1_v32i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <16 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtd_v16i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v16i1_v64i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpgtd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <16 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtq_v2i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v2i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sgt <2 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtq_v2i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v2i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sgt <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sgt <2 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtq_v4i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v4i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtq_v4i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v4i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgtq_v8i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v8i1_v32i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgtq_v8i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgtq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v8i1_v64i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpgtq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sgt <8 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleb %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgeb_v16i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeb_v16i1_v32i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleb %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgeb_v16i1_v32i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpsgeb_v16i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleb %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgeb_v16i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltb (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeb_v16i1_v64i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleb %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgeb_v16i1_v64i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltb (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleb %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp sge <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgeb_v32i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltb (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp sge <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeb_v32i1_v64i1_mask(i32 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleb %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp sge <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgeb_v32i1_v64i1_mask_mem(i32 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltb (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vmovdqa (%rsi), %ymm3
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm3, %ymm0
-; NoVLX-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
-; NoVLX-NEXT:    vpxor %ymm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; NoVLX-NEXT:    vpmovsxbd %xmm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
-; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
+; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp sge <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmplew %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgew_v8i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgew_v8i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmplew %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgew_v8i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpsgew_v8i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmplew %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgew_v8i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v8i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmplew %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v8i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmplew %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgew_v16i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgew_v16i1_v32i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmplew %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgew_v16i1_v32i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmplew %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgew_v16i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v16i1_v64i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmplew %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v16i1_v64i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmplew %zmm0, %zmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vmovq %xmm3, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm3
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm2, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm5, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
+; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
 ; NoVLX-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp sge <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgew_v32i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgew_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltw (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NoVLX-NEXT:    vmovq %xmm1, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm1
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm2
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
 ; NoVLX-NEXT:    vmovdqa 32(%rdi), %ymm2
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm2, %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp sge <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v32i1_v64i1_mask(i32 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmplew %zmm0, %zmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vmovq %xmm3, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm3
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm5, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp sge <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgew_v32i1_v64i1_mask_mem(i32 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltw (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; NoVLX-NEXT:    vmovq %xmm1, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm1
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vmovdqa (%rsi), %ymm3
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm3, %ymm0
-; NoVLX-NEXT:    vmovdqa 32(%rsi), %ymm3
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
-; NoVLX-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
-; NoVLX-NEXT:    vpxor %ymm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %ymm3, %ymm1, %ymm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; NoVLX-NEXT:    vmovdqa (%rsi), %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    vmovdqa 32(%rsi), %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm2, %ymm1
+; NoVLX-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
+; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp sge <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsged_v4i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v4i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v4i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsged_v4i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v4i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v4i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsged_v8i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v8i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v8i1_v32i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsged_v8i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v8i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v8i1_v64i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsged_v16i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v16i1_v32i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v16i1_v32i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsged_v16i1_v32i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <16 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsged_v16i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsged_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpnltd (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v16i1_v64i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpled %zmm0, %zmm1, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v16i1_v64i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpnltd (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsged_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpnltd (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsged_v16i1_v64i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsged_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpnltd (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <16 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgeq_v2i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v2i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sge <2 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgeq_v2i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v2i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sge <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp sge <2 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgeq_v4i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v4i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %ymm0, %ymm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgeq_v4i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v4i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %ymm0, %ymm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpsgeq_v8i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpnltq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v8i1_v32i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpnltq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpnltq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpnltq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpsgeq_v8i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpsgeq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpnltq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v8i1_v64i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpleq %zmm0, %zmm1, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpnltq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpnltq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpnltq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp sge <8 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultb_v16i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultb_v16i1_v32i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultb_v16i1_v32i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpultb_v16i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultb_v16i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultb_v16i1_v64i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultb_v16i1_v64i1_mask_mem(i16 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp ult <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultb_v32i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp ult <32 x i8> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultb_v32i1_v64i1_mask(i32 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm3, %xmm3
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm4 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm4, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm4, %ymm1, %ymm1
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; NoVLX-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %1 = bitcast <4 x i64> %__b to <32 x i8>
   %2 = icmp ult <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultb_v32i1_v64i1_mask_mem(i32 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm3, %ymm3
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm3, %ymm0
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; NoVLX-NEXT:    vpmovsxbd %xmm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
-; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <32 x i8>
   %2 = icmp ult <32 x i8> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultw_v8i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultw_v8i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultw_v8i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 
 define zeroext i64 @test_vpcmpultw_v8i1_v64i1_mask(<2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultw_v8i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v8i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v8i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultw_v16i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultw_v16i1_v32i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultw_v16i1_v32i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultw_v16i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v16i1_v64i1_mask(i16 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v16i1_v64i1_mask_mem(i16 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
 ; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vmovq %xmm3, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vmovq %xmm2, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm2
+; NoVLX-NEXT:    vmovd %eax, %xmm3
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm2, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm5, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm4
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm4, %xmm4
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm4, %ymm2
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm4 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm4, %ymm3, %ymm3
-; NoVLX-NEXT:    vpxor %ymm4, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm2, %ymm2
-; NoVLX-NEXT:    vpmovsxwd %ymm2, %zmm2
-; NoVLX-NEXT:    vpmovdb %zmm2, %xmm2
-; NoVLX-NEXT:    vpmovsxbd %xmm2, %zmm2
-; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpxor %ymm4, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm4, %ymm1, %ymm1
+; NoVLX-NEXT:    vpxor %ymm4, %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpxor %ymm4, %ymm1, %ymm0
+; NoVLX-NEXT:    vpxor %ymm4, %ymm3, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp ult <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultw_v32i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NoVLX-NEXT:    vmovq %xmm1, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm1
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm3, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpxor 32(%rdi), %ymm2, %ymm3
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rdi), %ymm2, %ymm1
+; NoVLX-NEXT:    vpxor (%rdi), %ymm2, %ymm3
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm3, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm0
+; NoVLX-NEXT:    vpxor 32(%rdi), %ymm2, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp ult <32 x i16> %0, %1
   %3 = shufflevector <32 x i1> %2, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v32i1_v64i1_mask(i32 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
-; NoVLX-NEXT:    vmovq %xmm2, %rax
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; NoVLX-NEXT:    vmovq %xmm3, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
-; NoVLX-NEXT:    vmovd %eax, %xmm3
+; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm4
-; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm8
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm5
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm8
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
+; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm6
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm7
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm2
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
-; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
-; NoVLX-NEXT:    movl %eax, %ecx
-; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm9
-; NoVLX-NEXT:    movl %ecx, %eax
-; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    movq %rcx, %rax
-; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm6, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm4
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm6
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vmovq %xmm7, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm5
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpextrq $1, %xmm7, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm5, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm7
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm7
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm3
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm6, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm1, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm6
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm2
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm5, %xmm5
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm5, %xmm5
-; NoVLX-NEXT:    vmovq %xmm1, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm9, %ymm4, %ymm1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm0
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vinserti128 $1, %xmm7, %ymm3, %ymm3
-; NoVLX-NEXT:    vpternlogd $255, %zmm6, %zmm6, %zmm6 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm6, %xmm6
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovq %xmm8, %rcx
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm5 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm5, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    movl %ecx, %eax
+; NoVLX-NEXT:    shrl $16, %eax
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rcx, %rax
+; NoVLX-NEXT:    shrq $32, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpextrq $1, %xmm8, %rax
+; NoVLX-NEXT:    shrq $48, %rcx
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movl %eax, %ecx
+; NoVLX-NEXT:    shrl $16, %ecx
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    movq %rax, %rcx
+; NoVLX-NEXT:    shrq $32, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm4, %xmm4
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm7, %ymm3
+; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm1, %ymm1
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    shrq $48, %rax
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm4, %ymm2
+; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm4 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
+; NoVLX-NEXT:    vpxor %ymm4, %ymm3, %ymm3
+; NoVLX-NEXT:    vpxor %ymm4, %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm2, %ymm2
+; NoVLX-NEXT:    vpmovsxwd %ymm2, %zmm2
+; NoVLX-NEXT:    vpxor %ymm4, %ymm0, %ymm0
+; NoVLX-NEXT:    vpxor %ymm4, %ymm1, %ymm1
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %ymm5, %ymm1, %ymm1
-; NoVLX-NEXT:    vpxor %ymm5, %ymm3, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm2, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %1 = bitcast <8 x i64> %__b to <32 x i16>
   %2 = icmp ult <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>
   %4 = and <32 x i1> %2, %3
   %5 = shufflevector <32 x i1> %4, <32 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultw_v32i1_v64i1_mask_mem(i32 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $96, %rsp
-; NoVLX-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; NoVLX-NEXT:    vmovq %xmm1, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
 ; NoVLX-NEXT:    vmovd %eax, %xmm2
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm3
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
+; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm4
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpextrq $1, %xmm1, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm1
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vmovq %xmm0, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm2
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
-; NoVLX-NEXT:    vmovq %xmm3, %rcx
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vmovq %xmm4, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vmovd %ecx, %xmm4
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
+; NoVLX-NEXT:    vmovd %ecx, %xmm3
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
-; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm3, %rax
+; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm3, %xmm3
+; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
-; NoVLX-NEXT:    vmovq %xmm0, %rcx
+; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm3, %xmm3
 ; NoVLX-NEXT:    movl %ecx, %eax
 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vmovd %ecx, %xmm4
 ; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; NoVLX-NEXT:    vpextrq $1, %xmm0, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm2, %rax
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; NoVLX-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; NoVLX-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm4, %xmm4
-; NoVLX-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5 {%k2} {z}
-; NoVLX-NEXT:    vpmovdb %zmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    shrl $16, %edi
+; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm1
 ; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm2, %ymm3
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm3, %ymm0
-; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpxor 32(%rsi), %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm2, %ymm1
+; NoVLX-NEXT:    vpxor 32(%rsi), %ymm2, %ymm3
+; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpmovdb %zmm1, %xmm1
-; NoVLX-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; NoVLX-NEXT:    vpmovsxbd %xmm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %eax
-; NoVLX-NEXT:    shlq $32, %rax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpxor (%rsi), %ymm2, %ymm2
+; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    kmovw %edi, %k2
+; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %ecx
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    shll $16, %eax
+; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <32 x i16>
   %2 = icmp ult <32 x i16> %0, %1
   %3 = bitcast i32 %__u to <32 x i1>

 ; VLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultd_v4i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v4i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v4i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultd_v4i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v4i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %1 = bitcast <2 x i64> %__b to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %4 = and <4 x i1> %2, %extract.i
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v4i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x i32>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i32> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <4 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultd_v8i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v8i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v8i1_v32i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultd_v8i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v8i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %1 = bitcast <4 x i64> %__b to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v8i1_v64i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x i32>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i32> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastd (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <8 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultd_v16i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v16i1_v32i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v16i1_v32i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultd_v16i1_v32i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <16 x i32> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultd_v16i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v16i1_v64i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %1 = bitcast <8 x i64> %__b to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v16i1_v64i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x i32>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vpcmpltud (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <16 x i32> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultd_v16i1_v64i1_mask_mem_b(i16 zeroext %__u, <8 x i64> %__a, i32* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultd_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vpcmpltud (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x i32>
   %load = load i32, i32* %__b
   %vec = insertelement <16 x i32> undef, i32 %load, i32 0
   %1 = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <16 x i32> %0, %1

 ; VLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultq_v2i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v2i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v2i1_v32i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v2i1_v32i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp ult <2 x i64> %0, %1

 ; VLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultq_v2i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v2i1_v64i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %1 = bitcast <2 x i64> %__b to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
   %4 = and <2 x i1> %2, %extract.i
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v2i1_v64i1_mask_mem(i8 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x i64>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp ult <2 x i64> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v2i1_v64i1_mask_mem_b(i8 zeroext %__u, <2 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %xmm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <2 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <2 x i64> %vec, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %2 = icmp ult <2 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultq_v4i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v4i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v4i1_v32i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultq_v4i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v4i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %1 = bitcast <4 x i64> %__b to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x i64>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i64> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v4i1_v64i1_mask_mem_b(i8 zeroext %__u, <4 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vpbroadcastq (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <4 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <4 x i64> %vec, <4 x i64> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <4 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultq_v8i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpltuq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v8i1_v32i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v8i1_v32i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vpcmpltuq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultq_v8i1_v32i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i64> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultq_v8i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpltuq (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v8i1_v64i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %1 = bitcast <8 x i64> %__b to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v8i1_v64i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x i64>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vpcmpltuq (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i64> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vpcmpultq_v8i1_v64i1_mask_mem_b(i8 zeroext %__u, <8 x i64> %__a, i64* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vpcmpltuq (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x i64>
   %load = load i64, i64* %__b
   %vec = insertelement <8 x i64> undef, i64 %load, i32 0
   %1 = shufflevector <8 x i64> %vec, <8 x i64> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = icmp ult <8 x i64> %0, %1

 ; VLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %1 = bitcast <2 x i64> %__b to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v4i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v4i1_v32i1_mask_mem_b(<2 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load float, float* %__b
   %vec = insertelement <4 x float> undef, float %load, i32 0
   %1 = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask(i4 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %1 = bitcast <2 x i64> %__b to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask_mem(i4 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovaps (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask_mem_b(i4 zeroext %__u, <2 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load float, float* %__b
   %vec = insertelement <4 x float> undef, float %load, i32 0
   %1 = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x float> %0, %1

 ; VLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %1 = bitcast <2 x i64> %__b to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v4i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v4i1_v64i1_mask_mem_b(<2 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to4}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load float, float* %__b
   %vec = insertelement <4 x float> undef, float %load, i32 0
   %1 = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask(i4 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %1 = bitcast <2 x i64> %__b to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask_mem(i4 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovaps (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <4 x float>
   %2 = fcmp oeq <4 x float> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask_mem_b(i4 zeroext %__u, <2 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to4}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <4 x float>
   %load = load float, float* %__b
   %vec = insertelement <4 x float> undef, float %load, i32 0
   %1 = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x float> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %1 = bitcast <4 x i64> %__b to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v8i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v8i1_v32i1_mask_mem_b(<4 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load float, float* %__b
   %vec = insertelement <8 x float> undef, float %load, i32 0
   %1 = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v8i1_v32i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %1 = bitcast <4 x i64> %__b to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v8i1_v32i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load float, float* %__b
   %vec = insertelement <8 x float> undef, float %load, i32 0
   %1 = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x float> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %1 = bitcast <4 x i64> %__b to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v8i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v8i1_v64i1_mask_mem_b(<4 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to8}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load float, float* %__b
   %vec = insertelement <8 x float> undef, float %load, i32 0
   %1 = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v8i1_v64i1_mask(i8 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %1 = bitcast <4 x i64> %__b to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v8i1_v64i1_mask_mem(i8 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovaps (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <8 x float>
   %2 = fcmp oeq <8 x float> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to8}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %ymm1
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $8, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $8, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <8 x float>
   %load = load float, float* %__b
   %vec = insertelement <8 x float> undef, float %load, i32 0
   %1 = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x float> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %1 = bitcast <8 x i64> %__b to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v16i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqps_v16i1_v32i1_mask_mem_b(<8 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v16i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load float, float* %__b
   %vec = insertelement <16 x float> undef, float %load, i32 0
   %1 = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v16i1_v32i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %1 = bitcast <8 x i64> %__b to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqps_v16i1_v32i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load float, float* %__b
   %vec = insertelement <16 x float> undef, float %load, i32 0
   %1 = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <16 x float> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %1 = bitcast <8 x i64> %__b to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v16i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps (%rdi), %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqps_v16i1_v64i1_mask_mem_b(<8 x i64> %__a, float* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqps_v16i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqps_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpeqps (%rdi){1to16}, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load float, float* %__b
   %vec = insertelement <16 x float> undef, float %load, i32 0
   %1 = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v16i1_v64i1_mask(i16 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps %zmm1, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %1 = bitcast <8 x i64> %__b to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>
   %4 = and <16 x i1> %2, %3
   %5 = shufflevector <16 x i1> %4, <16 x i1> zeroinitializer, <64 x i32> <i32 0,i32 1,i32 2,i32 3,i32 4,i32 5,i32 6,i32 7,i32 8,i32 9,i32 10,i32 11,i32 12,i32 13,i32 14,i32 15,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31,i32 16,i32 17,i32 18,i32 19,i32 20,i32 21,i32 22,i32 23,i32 24,i32 25,i32 26,i32 27,i32 28,i32 29,i32 30,i32 31>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqps_v16i1_v64i1_mask_mem(i16 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqps_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps (%rsi), %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <16 x float>
   %2 = fcmp oeq <16 x float> %0, %1
   %3 = bitcast i16 %__u to <16 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqps_v16i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
-; NoVLX-NEXT:    kmovw %edi, %k1
-; NoVLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k1 {%k1}
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %edi, %k1
+; NoVLX-NEXT:    vcmpeqps (%rsi){1to16}, %zmm0, %k0 {%k1}
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <16 x float>
   %load = load float, float* %__b
   %vec = insertelement <16 x float> undef, float %load, i32 0
   %1 = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <16 x float> %0, %1

 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %1 = bitcast <2 x i64> %__b to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v2i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovapd (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v2i1_v32i1_mask_mem_b(<2 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0]
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load double, double* %__b
   %vec = insertelement <2 x double> undef, double %load, i32 0
   %1 = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 0, i32 0>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v2i1_v32i1_mask(i2 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %1 = bitcast <2 x i64> %__b to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = bitcast i2 %__u to <2 x i1>
   %4 = and <2 x i1> %2, %3
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem(i2 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovapd (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = bitcast i2 %__u to <2 x i1>
   %4 = and <2 x i1> %2, %3
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem_b(i2 zeroext %__u, <2 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0]
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load double, double* %__b
   %vec = insertelement <2 x double> undef, double %load, i32 0
   %1 = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 0, i32 0>
   %2 = fcmp oeq <2 x double> %0, %1

 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %1 = bitcast <2 x i64> %__b to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v2i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovapd (%rdi), %xmm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v2i1_v64i1_mask_mem_b(<2 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to2}, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0]
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load double, double* %__b
   %vec = insertelement <2 x double> undef, double %load, i32 0
   %1 = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 0, i32 0>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = shufflevector <2 x i1> %2, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v2i1_v64i1_mask(i2 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm1 killed %xmm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %1 = bitcast <2 x i64> %__b to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = bitcast i2 %__u to <2 x i1>
   %4 = and <2 x i1> %2, %3
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem(i2 zeroext %__u, <2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovapd (%rsi), %xmm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <2 x double>
   %2 = fcmp oeq <2 x double> %0, %1
   %3 = bitcast i2 %__u to <2 x i1>
   %4 = and <2 x i1> %2, %3
   %5 = shufflevector <2 x i1> %4, <2 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem_b(i2 zeroext %__u, <2 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to2}, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %xmm0 killed %xmm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0]
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $14, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $14, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <2 x double>
   %load = load double, double* %__b
   %vec = insertelement <2 x double> undef, double %load, i32 0
   %1 = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 0, i32 0>
   %2 = fcmp oeq <2 x double> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %1 = bitcast <4 x i64> %__b to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v4i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovapd (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v4i1_v32i1_mask_mem_b(<4 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v4i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastsd (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load double, double* %__b
   %vec = insertelement <4 x double> undef, double %load, i32 0
   %1 = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v4i1_v32i1_mask(i4 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v4i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %1 = bitcast <4 x i64> %__b to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem(i4 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovapd (%rsi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load double, double* %__b
   %vec = insertelement <4 x double> undef, double %load, i32 0
   %1 = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x double> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %1 = bitcast <4 x i64> %__b to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v4i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vmovapd (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v4i1_v64i1_mask_mem_b(<4 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v4i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to4}, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    vbroadcastsd (%rdi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load double, double* %__b
   %vec = insertelement <4 x double> undef, double %load, i32 0
   %1 = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = shufflevector <4 x i1> %2, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v4i1_v64i1_mask(i4 zeroext %__u, <4 x i64> %__a, <4 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v4i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %1 = bitcast <4 x i64> %__b to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>
   %4 = and <4 x i1> %2, %3
   %5 = shufflevector <4 x i1> %4, <4 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem(i4 zeroext %__u, <4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vmovapd (%rsi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <4 x double>
   %2 = fcmp oeq <4 x double> %0, %1
   %3 = bitcast i4 %__u to <4 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to4}, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edx
-; NoVLX-NEXT:    kmovw %k0, %esi
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kshiftlw $12, %k0, %k0
+; NoVLX-NEXT:    kshiftrw $12, %k0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <4 x double>
   %load = load double, double* %__b
   %vec = insertelement <4 x double> undef, double %load, i32 0
   %1 = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <4 x double> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %1 = bitcast <8 x i64> %__b to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v8i1_v32i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vcmpeqpd (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vcmpoeqpd_v8i1_v32i1_mask_mem_b(<8 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v8i1_v32i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    vcmpeqpd (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load double, double* %__b
   %vec = insertelement <8 x double> undef, double %load, i32 0
   %1 = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v8i1_v32i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %1 = bitcast <8 x i64> %__b to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <32 x i1> %5 to i32
   ret i32 %6
 }
 
 define zeroext i32 @test_masked_vcmpoeqpd_v8i1_v32i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v32i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load double, double* %__b
   %vec = insertelement <8 x double> undef, double %load, i32 0
   %1 = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x double> %0, %1

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %1 = bitcast <8 x i64> %__b to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v8i1_v64i1_mask_mem(<8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vcmpeqpd (%rdi), %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vcmpoeqpd_v8i1_v64i1_mask_mem_b(<8 x i64> %__a, double* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vcmpoeqpd_v8i1_v64i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to8}, %zmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    vcmpeqpd (%rdi){1to8}, %zmm0, %k0
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load double, double* %__b
   %vec = insertelement <8 x double> undef, double %load, i32 0
   %1 = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v8i1_v64i1_mask(i8 zeroext %__u, <8 x i64> %__a, <8 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd %zmm1, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %1 = bitcast <8 x i64> %__b to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>
   %4 = and <8 x i1> %2, %3
   %5 = shufflevector <8 x i1> %4, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %6 = bitcast <64 x i1> %5 to i64
   ret i64 %6
 }
 
 define zeroext i64 @test_masked_vcmpoeqpd_v8i1_v64i1_mask_mem(i8 zeroext %__u, <8 x i64> %__a, <8 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vcmpoeqpd_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load <8 x i64>, <8 x i64>* %__b
   %1 = bitcast <8 x i64> %load to <8 x double>
   %2 = fcmp oeq <8 x double> %0, %1
   %3 = bitcast i8 %__u to <8 x i1>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to8}, %zmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v8i1_v64i1_mask_mem_b:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $64, %rsp
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vcmpeqpd (%rsi){1to8}, %zmm0, %k0 {%k1}
-; NoVLX-NEXT:    kshiftrw $7, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r8d
-; NoVLX-NEXT:    kshiftrw $6, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r9d
-; NoVLX-NEXT:    kshiftrw $5, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %r10d
-; NoVLX-NEXT:    kshiftrw $4, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %esi
-; NoVLX-NEXT:    kshiftrw $3, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %edi
-; NoVLX-NEXT:    kshiftrw $2, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %eax
-; NoVLX-NEXT:    kshiftrw $1, %k0, %k1
-; NoVLX-NEXT:    kmovw %k1, %ecx
-; NoVLX-NEXT:    kmovw %k0, %edx
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $0, %edx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $3, %edi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
-; NoVLX-NEXT:    shlq $32, %rcx
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    orq %rcx, %rax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    kmovw %k0, %eax
+; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <8 x double>
   %load = load double, double* %__b
   %vec = insertelement <8 x double> undef, double %load, i32 0
   %1 = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   %2 = fcmp oeq <8 x double> %0, %1

 ; VLX:       # %bb.0:
 ; VLX-NEXT:    vcmpleps {sae}, %zmm1, %zmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_cmpm_rnd_zero:
 ; NoVLX:       # %bb.0:
-; NoVLX-NEXT:    pushq %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_offset 16
-; NoVLX-NEXT:    .cfi_offset %rbp, -16
-; NoVLX-NEXT:    movq %rsp, %rbp
-; NoVLX-NEXT:    .cfi_def_cfa_register %rbp
-; NoVLX-NEXT:    andq $-32, %rsp
-; NoVLX-NEXT:    subq $32, %rsp
-; NoVLX-NEXT:    vcmpleps {sae}, %zmm1, %zmm0, %k1
-; NoVLX-NEXT:    kxorw %k0, %k0, %k0
-; NoVLX-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; NoVLX-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NoVLX-NEXT:    vpmovdb %zmm0, %xmm0
-; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    kmovw %k0, (%rsp)
-; NoVLX-NEXT:    movl (%rsp), %eax
-; NoVLX-NEXT:    movq %rbp, %rsp
-; NoVLX-NEXT:    popq %rbp
+; NoVLX-NEXT:    vcmpleps {sae}, %zmm1, %zmm0, %k0
+; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
   %res = call i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float> %a, <16 x float> %b, i32 2, i16 -1, i32 8)
   %cast = bitcast i16 %res to <16 x i1>
   %shuffle = shufflevector <16 x i1> %cast, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %cast2 = bitcast <32 x i1> %shuffle to i32
   ret i32 %cast2
 }

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

 ; AVX2-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
 ; AVX512F-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm1
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    movl (%rsp), %eax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm0
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k2}
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k1}
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %ecx, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %ymm1, %ymm0, %k1
 ; AVX512BW-NEXT:    vpcmpgtb %ymm3, %ymm2, %k0 {%k1}
 ; AVX512BW-NEXT:    kmovd %k0, %eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
   %x0 = icmp sgt <32 x i8> %a, %b
   %x1 = icmp sgt <32 x i8> %c, %d
   %y = and <32 x i1> %x0, %x1
   %res = bitcast <32 x i1> %y to i32
   ret i32 %res
 }

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

 ; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
+; AVX512F-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmovsxwd %ymm1, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
 ; AVX512F-NEXT:    vpcmpgtw %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmovsxwd %ymm0, %zmm0
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; AVX512F-NEXT:    vpcmpgtw %ymm7, %ymm5, %ymm0
+; AVX512F-NEXT:    vpmovsxwd %ymm0, %zmm0
+; AVX512F-NEXT:    vpcmpgtw %ymm6, %ymm4, %ymm1
 ; AVX512F-NEXT:    vpmovsxwd %ymm1, %zmm1
-; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512F-NEXT:    vpcmpgtw %ymm6, %ymm4, %ymm2
-; AVX512F-NEXT:    vpmovsxwd %ymm2, %zmm2
-; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512F-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpgtw %ymm7, %ymm5, %ymm2
-; AVX512F-NEXT:    vpmovsxwd %ymm2, %zmm2
-; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    movl (%rsp), %eax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k2}
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %ecx, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtw %zmm1, %zmm0, %k1
 ; AVX512BW-NEXT:    vpcmpgtw %zmm3, %zmm2, %k0 {%k1}
 ; AVX512BW-NEXT:    kmovd %k0, %eax

 ; AVX2-NEXT:    orq %rcx, %rax
 ; AVX2-NEXT:    movq %rbp, %rsp
 ; AVX2-NEXT:    popq %rbp
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $64, %rsp
 ; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm3
+; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm3
+; AVX512F-NEXT:    vptestmd %zmm3, %zmm3, %k1
+; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k2
 ; AVX512F-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpgtb %ymm7, %ymm5, %ymm2
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpcmpgtb %ymm6, %ymm4, %ymm2
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX512F-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512F-NEXT:    vpslld $31, %zmm2, %zmm2
-; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k3
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k4
+; AVX512F-NEXT:    vpcmpgtb %ymm7, %ymm5, %ymm0
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    movl (%rsp), %ecx
-; AVX512F-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; AVX512F-NEXT:    vpcmpgtb %ymm6, %ymm4, %ymm2
+; AVX512F-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm3
+; AVX512F-NEXT:    vpmovsxbd %xmm2, %zmm2
+; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0 {%k4}
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    vptestmd %zmm3, %zmm3, %k0 {%k3}
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    shll $16, %ecx
+; AVX512F-NEXT:    orl %eax, %ecx
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k2}
+; AVX512F-NEXT:    kmovw %k0, %edx
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0 {%k1}
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %edx, %eax
 ; AVX512F-NEXT:    shlq $32, %rax
 ; AVX512F-NEXT:    orq %rcx, %rax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %zmm1, %zmm0, %k1
 ; AVX512BW-NEXT:    vpcmpgtb %zmm3, %zmm2, %k0 {%k1}
 ; AVX512BW-NEXT:    kmovq %k0, %rax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
   %x0 = icmp sgt <64 x i8> %a, %b
   %x1 = icmp sgt <64 x i8> %c, %d
   %y = and <64 x i1> %x0, %x1
   %res = bitcast <64 x i1> %y to i64
   ret i64 %res
 }

test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll

 ; AVX2-FAST-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vpsrlw $7, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: ext_i32_32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
+; AVX512F-NEXT:    kmovw %edi, %k2
 ; AVX512F-NEXT:    movl {{.*}}(%rip), %eax
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
 ; AVX512F-NEXT:    vpbroadcastd %eax, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: ext_i32_32i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    kmovd %edi, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 {{.*}}(%rip), %ymm0 {%k1} {z}
 ; AVX512VLBW-NEXT:    retq
   %1 = bitcast i32 %a0 to <32 x i1>

 ; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm1, %ymm2, %ymm2
 ; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpsrlw $15, %ymm1, %ymm1
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: ext_i32_32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm1
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX512F-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrl $16, %edi
+; AVX512F-NEXT:    kmovw %edi, %k2
+; AVX512F-NEXT:    movl {{.*}}(%rip), %eax
+; AVX512F-NEXT:    vpbroadcastd %eax, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k2} {z}
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: ext_i32_32i16:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    kmovd %edi, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 {{.*}}(%rip), %zmm0 {%k1} {z}
 ; AVX512VLBW-NEXT:    retq
   %1 = bitcast i32 %a0 to <32 x i1>

 ; AVX2-FAST-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vpsrlw $7, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vpand %ymm3, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: ext_i64_64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $64, %rsp
-; AVX512F-NEXT:    movl %edi, (%rsp)
-; AVX512F-NEXT:    shrq $32, %rdi
-; AVX512F-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movq %rdi, %rax
+; AVX512F-NEXT:    movq %rdi, %rcx
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    movl %edi, %edx
+; AVX512F-NEXT:    shrl $16, %edx
+; AVX512F-NEXT:    shrq $32, %rax
+; AVX512F-NEXT:    shrq $48, %rcx
+; AVX512F-NEXT:    kmovw %ecx, %k2
+; AVX512F-NEXT:    kmovw %eax, %k3
+; AVX512F-NEXT:    kmovw %edx, %k4
 ; AVX512F-NEXT:    movl {{.*}}(%rip), %eax
-; AVX512F-NEXT:    kmovw (%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
 ; AVX512F-NEXT:    vpbroadcastd %eax, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k2} {z}
+; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k4} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k2
-; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k1} {z}
+; AVX512F-NEXT:    vpbroadcastd %eax, %zmm1 {%k3} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vpbroadcastd %eax, %zmm2 {%k2} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
 ; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: ext_i64_64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    kmovq %rdi, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 {{.*}}(%rip), %zmm0 {%k1} {z}
 ; AVX512VLBW-NEXT:    retq
   %1 = bitcast i64 %a0 to <64 x i1>
   %2 = zext <64 x i1> %1 to <64 x i8>
   ret <64 x i8> %2
 }

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

 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
 ; AVX512F-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    movl (%rsp), %eax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %ecx, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %ymm1, %ymm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
 ; AVX512BW-NEXT:    vzeroupper

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

 ; AVX2-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $32, %rsp
-; AVX512F-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpmovsxwd %ymm1, %zmm1
-; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vpcmpgtw %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmovsxwd %ymm0, %zmm0
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    movl (%rsp), %eax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm0
+; AVX512F-NEXT:    vpmovsxwd %ymm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %ecx, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtw %zmm1, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
 ; AVX512BW-NEXT:    vzeroupper

 ; AVX2-NEXT:    orq %rcx, %rax
 ; AVX2-NEXT:    movq %rbp, %rsp
 ; AVX2-NEXT:    popq %rbp
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $64, %rsp
 ; AVX512F-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX512F-NEXT:    vpmovsxbd %xmm2, %zmm2
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm2
 ; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    shll $16, %ecx
+; AVX512F-NEXT:    orl %eax, %ecx
 ; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
 ; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    kmovw %k0, %edx
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    movl (%rsp), %ecx
-; AVX512F-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    shll $16, %eax
+; AVX512F-NEXT:    orl %edx, %eax
 ; AVX512F-NEXT:    shlq $32, %rax
 ; AVX512F-NEXT:    orq %rcx, %rax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %zmm1, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovq %k0, %rax
 ; AVX512BW-NEXT:    vzeroupper

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

 ; AVX2-NEXT:    andb $1, %al
 ; AVX2-NEXT:    movb %al, (%rdi)
 ; AVX2-NEXT:    movq %rdi, %rax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm4
 ; AVX512F-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm3
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm4
+; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm2
+; AVX512F-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX512F-NEXT:    # kill: def %xmm0 killed %xmm0 killed %ymm0
-; AVX512F-NEXT:    vmovdqa %xmm4, %xmm2
+; AVX512F-NEXT:    vmovdqa %xmm4, %xmm1
+; AVX512F-NEXT:    # kill: def %xmm2 killed %xmm2 killed %ymm2
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v64i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm4
 ; AVX512DQ-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm4, %xmm3
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm4
+; AVX512DQ-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm2
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX512DQ-NEXT:    # kill: def %xmm0 killed %xmm0 killed %ymm0
-; AVX512DQ-NEXT:    vmovdqa %xmm4, %xmm2
+; AVX512DQ-NEXT:    vmovdqa %xmm4, %xmm1
+; AVX512DQ-NEXT:    # kill: def %xmm2 killed %xmm2 killed %ymm2
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %zmm1, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    retq

 ; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    vpacksswb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v32f32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vcmpltps %zmm0, %zmm2, %k1
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vcmpltps %zmm1, %zmm3, %k1
+; AVX512F-NEXT:    vcmpltps %zmm0, %zmm2, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v32f32:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vcmpltps %zmm0, %zmm2, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vcmpltps %zmm1, %zmm3, %k0
+; AVX512DQ-NEXT:    vcmpltps %zmm0, %zmm2, %k1
+; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v32f32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vcmpltps %zmm0, %zmm2, %k0
 ; AVX512BW-NEXT:    vcmpltps %zmm1, %zmm3, %k1
 ; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0

 ; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    vpacksswb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v32i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpcmpgtd %zmm2, %zmm0, %k1
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpcmpgtd %zmm3, %zmm1, %k1
+; AVX512F-NEXT:    vpcmpgtd %zmm2, %zmm0, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v32i32:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpcmpgtd %zmm2, %zmm0, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpcmpgtd %zmm3, %zmm1, %k0
+; AVX512DQ-NEXT:    vpcmpgtd %zmm2, %zmm0, %k1
+; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v32i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtd %zmm2, %zmm0, %k0
 ; AVX512BW-NEXT:    vpcmpgtd %zmm3, %zmm1, %k1
 ; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0

 ; AVX2-NEXT:    movb %al, (%rdi)
 ; AVX2-NEXT:    movq %rdi, %rax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v128i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtb %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpgtb %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vpcmpgtb %ymm6, %ymm2, %ymm2
-; AVX512F-NEXT:    vpcmpgtb %ymm7, %ymm3, %ymm3
-; AVX512F-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX512F-NEXT:    vpmovsxbd %xmm4, %zmm4
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm4
 ; AVX512F-NEXT:    vptestmd %zmm4, %zmm4, %k0
-; AVX512F-NEXT:    kmovw %k0, 14(%rdi)
-; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512F-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512F-NEXT:    kmovw %k0, 12(%rdi)
-; AVX512F-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512F-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512F-NEXT:    kmovw %k0, 10(%rdi)
-; AVX512F-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512F-NEXT:    kmovw %k0, 8(%rdi)
-; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512F-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512F-NEXT:    kmovw %k0, 6(%rdi)
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, 4(%rdi)
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, 2(%rdi)
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512F-NEXT:    vpcmpgtb %ymm5, %ymm1, %ymm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k2
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k3
+; AVX512F-NEXT:    vpcmpgtb %ymm6, %ymm2, %ymm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k4
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k5
+; AVX512F-NEXT:    vpcmpgtb %ymm7, %ymm3, %ymm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k6
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k7
+; AVX512F-NEXT:    kmovw %k7, 14(%rdi)
+; AVX512F-NEXT:    kmovw %k6, 12(%rdi)
+; AVX512F-NEXT:    kmovw %k5, 10(%rdi)
+; AVX512F-NEXT:    kmovw %k4, 8(%rdi)
+; AVX512F-NEXT:    kmovw %k3, 6(%rdi)
+; AVX512F-NEXT:    kmovw %k2, 4(%rdi)
+; AVX512F-NEXT:    kmovw %k1, 2(%rdi)
 ; AVX512F-NEXT:    kmovw %k0, (%rdi)
 ; AVX512F-NEXT:    movq %rdi, %rax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v128i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpcmpgtb %ymm4, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpcmpgtb %ymm5, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpcmpgtb %ymm6, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpcmpgtb %ymm7, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX512DQ-NEXT:    vpmovsxbd %xmm4, %zmm4
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm4
 ; AVX512DQ-NEXT:    vptestmd %zmm4, %zmm4, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 14(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512DQ-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 12(%rdi)
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; AVX512DQ-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512DQ-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 10(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512DQ-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 8(%rdi)
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512DQ-NEXT:    vpmovsxbd %xmm2, %zmm2
-; AVX512DQ-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 6(%rdi)
-; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 4(%rdi)
-; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512DQ-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512DQ-NEXT:    kmovw %k0, 2(%rdi)
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512DQ-NEXT:    vpcmpgtb %ymm5, %ymm1, %ymm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k2
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k3
+; AVX512DQ-NEXT:    vpcmpgtb %ymm6, %ymm2, %ymm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k4
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k5
+; AVX512DQ-NEXT:    vpcmpgtb %ymm7, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512DQ-NEXT:    vptestmd %zmm1, %zmm1, %k6
+; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512DQ-NEXT:    vptestmd %zmm0, %zmm0, %k7
+; AVX512DQ-NEXT:    kmovw %k7, 14(%rdi)
+; AVX512DQ-NEXT:    kmovw %k6, 12(%rdi)
+; AVX512DQ-NEXT:    kmovw %k5, 10(%rdi)
+; AVX512DQ-NEXT:    kmovw %k4, 8(%rdi)
+; AVX512DQ-NEXT:    kmovw %k3, 6(%rdi)
+; AVX512DQ-NEXT:    kmovw %k2, 4(%rdi)
+; AVX512DQ-NEXT:    kmovw %k1, 2(%rdi)
 ; AVX512DQ-NEXT:    kmovw %k0, (%rdi)
 ; AVX512DQ-NEXT:    movq %rdi, %rax
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v128i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtb %zmm3, %zmm1, %k0

 ; AVX2-NEXT:    vpacksswb %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    movq %rbp, %rsp
 ; AVX2-NEXT:    popq %rbp
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v32f64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm4, %k0
-; AVX512F-NEXT:    vcmpltpd %zmm1, %zmm5, %k1
-; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vcmpltpd %zmm2, %zmm6, %k0
 ; AVX512F-NEXT:    vcmpltpd %zmm3, %zmm7, %k1
 ; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
+; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm4, %k0
+; AVX512F-NEXT:    vcmpltpd %zmm1, %zmm5, %k2
+; AVX512F-NEXT:    kunpckbw %k0, %k2, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v32f64:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vcmpltpd %zmm0, %zmm4, %k0
-; AVX512DQ-NEXT:    vcmpltpd %zmm1, %zmm5, %k1
-; AVX512DQ-NEXT:    kunpckbw %k0, %k1, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vcmpltpd %zmm2, %zmm6, %k0
 ; AVX512DQ-NEXT:    vcmpltpd %zmm3, %zmm7, %k1
 ; AVX512DQ-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512DQ-NEXT:    vcmpltpd %zmm0, %zmm4, %k1
+; AVX512DQ-NEXT:    vcmpltpd %zmm1, %zmm5, %k2
+; AVX512DQ-NEXT:    kunpckbw %k1, %k2, %k1
+; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v32f64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vcmpltpd %zmm0, %zmm4, %k0
 ; AVX512BW-NEXT:    vcmpltpd %zmm1, %zmm5, %k1
 ; AVX512BW-NEXT:    kunpckbw %k0, %k1, %k0
 ; AVX512BW-NEXT:    vcmpltpd %zmm2, %zmm6, %k1

 ; AVX2-NEXT:    vpacksswb %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    movq %rbp, %rsp
 ; AVX2-NEXT:    popq %rbp
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: test_cmp_v32i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpcmpgtq %zmm4, %zmm0, %k0
-; AVX512F-NEXT:    vpcmpgtq %zmm5, %zmm1, %k1
-; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
-; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpcmpgtq %zmm6, %zmm2, %k0
 ; AVX512F-NEXT:    vpcmpgtq %zmm7, %zmm3, %k1
 ; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
+; AVX512F-NEXT:    vpcmpgtq %zmm4, %zmm0, %k0
+; AVX512F-NEXT:    vpcmpgtq %zmm5, %zmm1, %k2
+; AVX512F-NEXT:    kunpckbw %k0, %k2, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v32i64:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpcmpgtq %zmm4, %zmm0, %k0
-; AVX512DQ-NEXT:    vpcmpgtq %zmm5, %zmm1, %k1
-; AVX512DQ-NEXT:    kunpckbw %k0, %k1, %k0
-; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpcmpgtq %zmm6, %zmm2, %k0
 ; AVX512DQ-NEXT:    vpcmpgtq %zmm7, %zmm3, %k1
 ; AVX512DQ-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512DQ-NEXT:    vpcmpgtq %zmm4, %zmm0, %k1
+; AVX512DQ-NEXT:    vpcmpgtq %zmm5, %zmm1, %k2
+; AVX512DQ-NEXT:    kunpckbw %k1, %k2, %k1
+; AVX512DQ-NEXT:    vpmovm2d %k1, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm1
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_cmp_v32i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpcmpgtq %zmm4, %zmm0, %k0
 ; AVX512BW-NEXT:    vpcmpgtq %zmm5, %zmm1, %k1
 ; AVX512BW-NEXT:    kunpckbw %k0, %k1, %k0
 ; AVX512BW-NEXT:    vpcmpgtq %zmm6, %zmm2, %k1

test/CodeGen/X86/vector-shuffle-v1.ll

; VL_BW_DQ-NEXT:    retq
   %b2 = icmp eq <16 x i32> %b, %b1
   %c = shufflevector <16 x i1> %a2, <16 x i1> %b2, <16 x i32> <i32 3, i32 6, i32 22, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 21, i32 7, i32 0>
   ret <16 x i1> %c
 }
 
 define <32 x i1> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(<32 x i1> %a) {
 ; AVX512F-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[3,6,u,12,3,7,7,0,3,6,1,13,3,u,7,0,u,u,22,u,u,u,u,u,u,u,u,u,u,21,u,u]
-; AVX512F-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX512F-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,6,u,u,u,u,u,u,u,u,u,u,5,u,u,19,22,u,28,19,23,23,16,19,22,17,29,19,u,23,16]
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,255,255,255,255,255,255,255,255,255,255,0,255,255,0,0,255,0,0,0,0,0,0,0,0,0,0,255,0,0]
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
+; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; AVX512F-NEXT:    vmovdqa32 {{.*#+}} zmm2 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
+; AVX512F-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
+; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[3,6,u,12,3,7,7,0,3,6,1,13,3,u,7,0,u,u,22,u,u,u,u,u,u,u,u,u,u,21,u,u]
-; AVX512VL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,6,u,u,u,u,u,u,u,u,u,u,5,u,u,19,22,u,28,19,23,23,16,19,22,17,29,19,u,23,16]
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,255,255,255,255,255,255,255,255,255,255,0,255,255,0,0,255,0,0,0,0,0,0,0,0,0,0,255,0,0]
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovsxbd %xmm0, %zmm1
+; AVX512VL-NEXT:    vpslld $31, %zmm1, %zmm1
+; AVX512VL-NEXT:    vptestmd %zmm1, %zmm1, %k1
+; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512VL-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512VL-NEXT:    vptestmd %zmm0, %zmm0, %k2
+; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
+; AVX512VL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; AVX512VL-NEXT:    vmovdqa32 {{.*#+}} zmm2 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
+; AVX512VL-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
+; AVX512VL-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; VL_BW_DQ-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
 ; VL_BW_DQ:       # %bb.0:
 ; VL_BW_DQ-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; VL_BW_DQ-NEXT:    vpmovb2m %ymm0, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2w %k0, %zmm0
 ; VL_BW_DQ-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0,3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
; VL_BW_DQ-NEXT:    retq
   %c = shufflevector < 16 x i1> %b, <16 x i1> undef, <16 x i32> zeroinitializer
   %d = bitcast <16 x i1> %c to i16
   ret i16 %d
 }
 
 define i64 @shuf64i1_zero(i64 %a) {
 ; AVX512F-LABEL: shuf64i1_zero:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    pushq %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_offset 16
-; AVX512F-NEXT:    .cfi_offset %rbp, -16
-; AVX512F-NEXT:    movq %rsp, %rbp
-; AVX512F-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512F-NEXT:    andq $-32, %rsp
-; AVX512F-NEXT:    subq $96, %rsp
-; AVX512F-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
+; AVX512F-NEXT:    kmovw %edi, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpbroadcastb %xmm0, %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512F-NEXT:    vpbroadcastd %xmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, (%rsp)
-; AVX512F-NEXT:    movl (%rsp), %ecx
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    kmovw %k0, %ecx
+; AVX512F-NEXT:    shll $16, %ecx
+; AVX512F-NEXT:    orl %eax, %ecx
 ; AVX512F-NEXT:    movq %rcx, %rax
 ; AVX512F-NEXT:    shlq $32, %rax
 ; AVX512F-NEXT:    orq %rcx, %rax
-; AVX512F-NEXT:    movq %rbp, %rsp
-; AVX512F-NEXT:    popq %rbp
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuf64i1_zero:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    pushq %rbp
-; AVX512VL-NEXT:    .cfi_def_cfa_offset 16
-; AVX512VL-NEXT:    .cfi_offset %rbp, -16
-; AVX512VL-NEXT:    movq %rsp, %rbp
-; AVX512VL-NEXT:    .cfi_def_cfa_register %rbp
-; AVX512VL-NEXT:    andq $-32, %rsp
-; AVX512VL-NEXT:    subq $96, %rsp
-; AVX512VL-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
-; AVX512VL-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1
+; AVX512VL-NEXT:    kmovw %edi, %k1
 ; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastb %xmm0, %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512VL-NEXT:    vpslld $31, %zmm1, %zmm1
-; AVX512VL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512VL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
-; AVX512VL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512VL-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpbroadcastd %xmm0, %zmm0
 ; AVX512VL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512VL-NEXT:    kmovw %k0, (%rsp)
-; AVX512VL-NEXT:    movl (%rsp), %ecx
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    kmovw %k0, %ecx
+; AVX512VL-NEXT:    shll $16, %ecx
+; AVX512VL-NEXT:    orl %eax, %ecx
 ; AVX512VL-NEXT:    movq %rcx, %rax
 ; AVX512VL-NEXT:    shlq $32, %rax
 ; AVX512VL-NEXT:    orq %rcx, %rax
-; AVX512VL-NEXT:    movq %rbp, %rsp
-; AVX512VL-NEXT:    popq %rbp
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; VL_BW_DQ-LABEL: shuf64i1_zero:
 ; VL_BW_DQ:       # %bb.0:
 ; VL_BW_DQ-NEXT:    kmovq %rdi, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2b %k0, %zmm0
 ; VL_BW_DQ-NEXT:    vpbroadcastb %xmm0, %zmm0