[X86] Add a DAG combine to transform (i8 (bitcast (v8i1 (extract_subvector (v16i1 X), 0)))) -> (i8 (trunc (i16 (bitcast (v16i1 X))))) on KNL target

[craig.topper]

Summary

Without AVX512DQ we don't have KMOVB so we can't really copy 8-bits of a k-register to a GPR. We have to copy 16 bits instead. We do this even if the DAG copy is from v8i1->v16i1. If we detect the (i8 (bitcast (v8i1 (extract_subvector (v16i1 X), 0)))) we should rewrite the types to match the copy we do support. By doing this, we can help known bits to propagate without losing the upper 8 bits of the input to the extract_subvector. This allows some zero extends to be removed since we have an isel pattern to use kmovw for (zero_extend (i16 (bitcast (v16i1 X))).

[RKSimon]

LGTM

llvm/test/CodeGen/X86/win32-eh-available-externally.ll

1 ↗

(On Diff #216207)

drop this

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

if ((SrcVT == MVT::v4i1 || SrcVT == MVT::v2i1) && VT.isScalarInteger() &&
       SmallVector<SDValue, 4> Ops(NumConcats, DAG.getUNDEF(SrcVT));
       Ops[0] = N0;
       N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i1, Ops);
       N0 = DAG.getBitcast(MVT::i8, N0);
       return DAG.getNode(ISD::TRUNCATE, dl, VT, N0);
     }
   }
 
+  // Look for (i8 (bitcast (v8i1 (extract_subvector (v16i1 X), 0)))) and
+  // replace with (i8 (trunc (i16 (bitcast (v16i1 X))))). This can occur
+  // due to insert_subvector legalization on KNL. By promoting the copy to i16
+  // we can help with known bits propagation from the vXi1 domain to the
+  // scalar domain.
+  if (VT == MVT::i8 && SrcVT == MVT::v8i1 && Subtarget.hasAVX512() &&
+      !Subtarget.hasDQI() && N0.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
+      N0.getOperand(0).getValueType() == MVT::v16i1 &&
+      isNullConstant(N0.getOperand(1)))
+    return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT,
+                       DAG.getBitcast(MVT::i16, N0.getOperand(0)));
+
   // Since MMX types are special and don't usually play with other vector types,
   // it's better to handle them early to be sure we emit efficient code by
   // avoiding store-load conversions.
   if (VT == MVT::x86mmx) {
     // Detect MMX constant vectors.
     APInt UndefElts;
     SmallVector<APInt, 1> EltBits;
     if (getTargetConstantBitsFromNode(N0, 64, UndefElts, EltBits)) {

llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll

 ; X64-NEXT:    vcmplesd %xmm1, %xmm0, %k0
 ; X64-NEXT:    kmovw %k0, %ecx
 ; X64-NEXT:    vcmpunordsd {sae}, %xmm1, %xmm0, %k0
 ; X64-NEXT:    kmovw %k0, %edx
 ; X64-NEXT:    vcmpneqsd %xmm1, %xmm0, %k0 {%k1}
 ; X64-NEXT:    kmovw %k0, %esi
 ; X64-NEXT:    vcmpnltsd {sae}, %xmm1, %xmm0, %k0 {%k1}
 ; X64-NEXT:    kmovw %k0, %eax
-; X64-NEXT:    orb %sil, %al
-; X64-NEXT:    orb %dl, %al
-; X64-NEXT:    orb %cl, %al
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    orl %edx, %eax
+; X64-NEXT:    orl %ecx, %eax
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cmp_sd_all:
 ; X86:       # %bb.0:
-; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 8
-; X86-NEXT:    .cfi_offset %ebx, -8
+; X86-NEXT:    .cfi_offset %esi, -8
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    kmovw %eax, %k1
 ; X86-NEXT:    vcmplesd %xmm1, %xmm0, %k0
 ; X86-NEXT:    kmovw %k0, %ecx
 ; X86-NEXT:    vcmpunordsd {sae}, %xmm1, %xmm0, %k0
 ; X86-NEXT:    kmovw %k0, %edx
 ; X86-NEXT:    vcmpneqsd %xmm1, %xmm0, %k0 {%k1}
-; X86-NEXT:    kmovw %k0, %ebx
+; X86-NEXT:    kmovw %k0, %esi
 ; X86-NEXT:    vcmpnltsd {sae}, %xmm1, %xmm0, %k0 {%k1}
 ; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    orb %bl, %al
-; X86-NEXT:    orb %dl, %al
-; X86-NEXT:    orb %cl, %al
+; X86-NEXT:    orl %esi, %eax
+; X86-NEXT:    orl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
 ; X86-NEXT:    # kill: def $al killed $al killed $eax
-; X86-NEXT:    popl %ebx
+; X86-NEXT:    popl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NEXT:    retl
 
   %res1 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 2, i8 -1, i32 4)
   %res2 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 3, i8 -1, i32 8)
   %res3 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 4, i8 %x3, i32 4)
   %res4 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 5, i8 %x3, i32 8)
 

 ; X64-NEXT:    vcmpless %xmm1, %xmm0, %k0
 ; X64-NEXT:    kmovw %k0, %ecx
 ; X64-NEXT:    vcmpunordss {sae}, %xmm1, %xmm0, %k0
 ; X64-NEXT:    kmovw %k0, %edx
 ; X64-NEXT:    vcmpneqss %xmm1, %xmm0, %k0 {%k1}
 ; X64-NEXT:    kmovw %k0, %esi
 ; X64-NEXT:    vcmpnltss {sae}, %xmm1, %xmm0, %k0 {%k1}
 ; X64-NEXT:    kmovw %k0, %eax
-; X64-NEXT:    andb %sil, %al
-; X64-NEXT:    andb %dl, %al
-; X64-NEXT:    andb %cl, %al
+; X64-NEXT:    andl %esi, %eax
+; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    andl %ecx, %eax
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cmp_ss_all:
 ; X86:       # %bb.0:
-; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 8
-; X86-NEXT:    .cfi_offset %ebx, -8
+; X86-NEXT:    .cfi_offset %esi, -8
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    kmovw %eax, %k1
 ; X86-NEXT:    vcmpless %xmm1, %xmm0, %k0
 ; X86-NEXT:    kmovw %k0, %ecx
 ; X86-NEXT:    vcmpunordss {sae}, %xmm1, %xmm0, %k0
 ; X86-NEXT:    kmovw %k0, %edx
 ; X86-NEXT:    vcmpneqss %xmm1, %xmm0, %k0 {%k1}
-; X86-NEXT:    kmovw %k0, %ebx
+; X86-NEXT:    kmovw %k0, %esi
 ; X86-NEXT:    vcmpnltss {sae}, %xmm1, %xmm0, %k0 {%k1}
 ; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    andb %bl, %al
-; X86-NEXT:    andb %dl, %al
-; X86-NEXT:    andb %cl, %al
+; X86-NEXT:    andl %esi, %eax
+; X86-NEXT:    andl %edx, %eax
+; X86-NEXT:    andl %ecx, %eax
 ; X86-NEXT:    # kill: def $al killed $al killed $eax
-; X86-NEXT:    popl %ebx
+; X86-NEXT:    popl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NEXT:    retl
   %res1 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 2, i8 -1, i32 4)
   %res2 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 3, i8 -1, i32 8)
   %res3 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 4, i8 %x3, i32 4)
   %res4 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 5, i8 %x3, i32 8)
 
   %res11 = and i8 %res1, %res2

llvm/trunk/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll

; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_cmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_cmp_q_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1]
-; X86-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1 {%k2} # encoding: [0x62,0xf2,0xf5,0x2a,0x37,0xc8]
-; X86-NEXT:    vpcmpleq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xd9,0x02]
-; X86-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe9,0x05]
-; X86-NEXT:    vpcmpgtq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x37,0xf1]
-; X86-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X86-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1]
+; X86-NEXT:    vpcmpgtq %ymm0, %ymm1, %k2 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x37,0xd0]
+; X86-NEXT:    vpcmpleq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xd9,0x02]
+; X86-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe9,0x05]
+; X86-NEXT:    vpcmpgtq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x37,0xf1]
+; X86-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X86-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_cmp_q_256:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1]
-; X64-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1 {%k2} # encoding: [0x62,0xf2,0xf5,0x2a,0x37,0xc8]
-; X64-NEXT:    vpcmpleq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xd9,0x02]
-; X64-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe9,0x05]
-; X64-NEXT:    vpcmpgtq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x37,0xf1]
-; X64-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X64-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1]
+; X64-NEXT:    vpcmpgtq %ymm0, %ymm1, %k2 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x37,0xd0]
+; X64-NEXT:    vpcmpleq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xd9,0x02]
+; X64-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe9,0x05]
+; X64-NEXT:    vpcmpgtq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x37,0xf1]
+; X64-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X64-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.cmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.cmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.cmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 2, i8 %mask)
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_ucmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_ucmp_q_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1]
-; X86-NEXT:    vpcmpltuq %ymm1, %ymm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xc9,0x01]
-; X86-NEXT:    vpcmpleuq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xd9,0x02]
-; X86-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xe9,0x05]
-; X86-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xf1,0x06]
-; X86-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X86-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1]
+; X86-NEXT:    vpcmpltuq %ymm1, %ymm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd1,0x01]
+; X86-NEXT:    vpcmpleuq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd9,0x02]
+; X86-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe9,0x05]
+; X86-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xf1,0x06]
+; X86-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X86-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_ucmp_q_256:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1]
-; X64-NEXT:    vpcmpltuq %ymm1, %ymm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xc9,0x01]
-; X64-NEXT:    vpcmpleuq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xd9,0x02]
-; X64-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xe9,0x05]
-; X64-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xf1,0x06]
-; X64-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X64-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1]
+; X64-NEXT:    vpcmpltuq %ymm1, %ymm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd1,0x01]
+; X64-NEXT:    vpcmpleuq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd9,0x02]
+; X64-NEXT:    vpcmpneqq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe9,0x05]
+; X64-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xf1,0x06]
+; X64-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X64-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.ucmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.ucmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.ucmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 2, i8 %mask)
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_cmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_cmp_d_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1]
-; X86-NEXT:    vpcmpgtd %xmm0, %xmm1, %k1 {%k2} # encoding: [0x62,0xf1,0x75,0x0a,0x66,0xc8]
-; X86-NEXT:    vpcmpled %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xd9,0x02]
-; X86-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltd %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe9,0x05]
-; X86-NEXT:    vpcmpgtd %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x66,0xf1]
-; X86-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X86-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1]
+; X86-NEXT:    vpcmpgtd %xmm0, %xmm1, %k2 {%k1} # encoding: [0x62,0xf1,0x75,0x09,0x66,0xd0]
+; X86-NEXT:    vpcmpled %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xd9,0x02]
+; X86-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltd %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe9,0x05]
+; X86-NEXT:    vpcmpgtd %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x66,0xf1]
+; X86-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X86-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_cmp_d_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1]
-; X64-NEXT:    vpcmpgtd %xmm0, %xmm1, %k1 {%k2} # encoding: [0x62,0xf1,0x75,0x0a,0x66,0xc8]
-; X64-NEXT:    vpcmpled %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xd9,0x02]
-; X64-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltd %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe9,0x05]
-; X64-NEXT:    vpcmpgtd %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x66,0xf1]
-; X64-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X64-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1]
+; X64-NEXT:    vpcmpgtd %xmm0, %xmm1, %k2 {%k1} # encoding: [0x62,0xf1,0x75,0x09,0x66,0xd0]
+; X64-NEXT:    vpcmpled %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xd9,0x02]
+; X64-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltd %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe9,0x05]
+; X64-NEXT:    vpcmpgtd %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x66,0xf1]
+; X64-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X64-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.cmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.cmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.cmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 2, i8 %mask)
   %vec2 = insertelement <8 x i8> %vec1, i8 %res2, i32 2
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_ucmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_ucmp_d_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1]
-; X86-NEXT:    vpcmpltud %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xc9,0x01]
-; X86-NEXT:    vpcmpleud %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xd9,0x02]
-; X86-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltud %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xe9,0x05]
-; X86-NEXT:    vpcmpnleud %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xf1,0x06]
-; X86-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X86-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1]
+; X86-NEXT:    vpcmpltud %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd1,0x01]
+; X86-NEXT:    vpcmpleud %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd9,0x02]
+; X86-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltud %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe9,0x05]
+; X86-NEXT:    vpcmpnleud %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xf1,0x06]
+; X86-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X86-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_ucmp_d_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1]
-; X64-NEXT:    vpcmpltud %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xc9,0x01]
-; X64-NEXT:    vpcmpleud %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xd9,0x02]
-; X64-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltud %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xe9,0x05]
-; X64-NEXT:    vpcmpnleud %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xf1,0x06]
-; X64-NEXT:    kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c]
-; X64-NEXT:    kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1]
+; X64-NEXT:    vpcmpltud %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd1,0x01]
+; X64-NEXT:    vpcmpleud %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd9,0x02]
+; X64-NEXT:    vpcmpneqd %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltud %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe9,0x05]
+; X64-NEXT:    vpcmpnleud %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xf1,0x06]
+; X64-NEXT:    kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c]
+; X64-NEXT:    kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.ucmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.ucmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.ucmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 2, i8 %mask)
   %vec2 = insertelement <8 x i8> %vec1, i8 %res2, i32 2
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_cmp_q_128(<2 x i64> %a0, <2 x i64> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_cmp_q_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x29,0xc1]
-; X86-NEXT:    vpcmpgtq %xmm0, %xmm1, %k1 {%k2} # encoding: [0x62,0xf2,0xf5,0x0a,0x37,0xc8]
-; X86-NEXT:    vpcmpleq %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xd9,0x02]
-; X86-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltq %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe9,0x05]
-; X86-NEXT:    vpcmpgtq %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x37,0xf1]
-; X86-NEXT:    kshiftlw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0e]
-; X86-NEXT:    kshiftrw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0e]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x29,0xc1]
+; X86-NEXT:    vpcmpgtq %xmm0, %xmm1, %k2 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x37,0xd0]
+; X86-NEXT:    vpcmpleq %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xd9,0x02]
+; X86-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltq %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe9,0x05]
+; X86-NEXT:    vpcmpgtq %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x37,0xf1]
+; X86-NEXT:    kshiftlw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0e]
+; X86-NEXT:    kshiftrw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0e]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_cmp_q_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x29,0xc1]
-; X64-NEXT:    vpcmpgtq %xmm0, %xmm1, %k1 {%k2} # encoding: [0x62,0xf2,0xf5,0x0a,0x37,0xc8]
-; X64-NEXT:    vpcmpleq %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xd9,0x02]
-; X64-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltq %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe9,0x05]
-; X64-NEXT:    vpcmpgtq %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x37,0xf1]
-; X64-NEXT:    kshiftlw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0e]
-; X64-NEXT:    kshiftrw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0e]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x29,0xc1]
+; X64-NEXT:    vpcmpgtq %xmm0, %xmm1, %k2 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x37,0xd0]
+; X64-NEXT:    vpcmpleq %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xd9,0x02]
+; X64-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltq %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe9,0x05]
+; X64-NEXT:    vpcmpgtq %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x37,0xf1]
+; X64-NEXT:    kshiftlw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0e]
+; X64-NEXT:    kshiftrw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0e]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.cmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.cmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.cmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 2, i8 %mask)
   %vec2 = insertelement <8 x i8> %vec1, i8 %res2, i32 2
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %vec7 = insertelement <8 x i8> %vec6, i8 %res7, i32 7
   ret <8 x i8> %vec7
 }
 
 define <8 x i8> @test_mask_ucmp_q_128(<2 x i64> %a0, <2 x i64> %a1, i8 %mask) {
 ; X86-LABEL: test_mask_ucmp_q_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0]
-; X86-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x29,0xc1]
-; X86-NEXT:    vpcmpltuq %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xc9,0x01]
-; X86-NEXT:    vpcmpleuq %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xd9,0x02]
-; X86-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe1,0x04]
-; X86-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xe9,0x05]
-; X86-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xf1,0x06]
-; X86-NEXT:    kshiftlw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0e]
-; X86-NEXT:    kshiftrw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0e]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x29,0xc1]
+; X86-NEXT:    vpcmpltuq %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xd1,0x01]
+; X86-NEXT:    vpcmpleuq %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xd9,0x02]
+; X86-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe1,0x04]
+; X86-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xe9,0x05]
+; X86-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xf1,0x06]
+; X86-NEXT:    kshiftlw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0e]
+; X86-NEXT:    kshiftrw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0e]
 ; X86-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X86-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X86-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X86-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X86-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X86-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X86-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X86-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X86-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X86-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X86-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X86-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X86-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X86-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_ucmp_q_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7]
-; X64-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x0a,0x29,0xc1]
-; X64-NEXT:    vpcmpltuq %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xc9,0x01]
-; X64-NEXT:    vpcmpleuq %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xd9,0x02]
-; X64-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1f,0xe1,0x04]
-; X64-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xe9,0x05]
-; X64-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x0a,0x1e,0xf1,0x06]
-; X64-NEXT:    kshiftlw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0e]
-; X64-NEXT:    kshiftrw $14, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0e]
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x29,0xc1]
+; X64-NEXT:    vpcmpltuq %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xd1,0x01]
+; X64-NEXT:    vpcmpleuq %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xd9,0x02]
+; X64-NEXT:    vpcmpneqq %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1f,0xe1,0x04]
+; X64-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xe9,0x05]
+; X64-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x1e,0xf1,0x06]
+; X64-NEXT:    kshiftlw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0e]
+; X64-NEXT:    kshiftrw $14, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0e]
 ; X64-NEXT:    kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0]
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0]
 ; X64-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
+; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
 ; X64-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x01]
 ; X64-NEXT:    kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3]
 ; X64-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; X64-NEXT:    kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4]
 ; X64-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; X64-NEXT:    kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5]
 ; X64-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x05]
 ; X64-NEXT:    kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6]
 ; X64-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
-; X64-NEXT:    kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2]
+; X64-NEXT:    kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
 ; X64-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res0 = call i8 @llvm.x86.avx512.mask.ucmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 0, i8 %mask)
   %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0
   %res1 = call i8 @llvm.x86.avx512.mask.ucmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 1, i8 %mask)
   %vec1 = insertelement <8 x i8> %vec0, i8 %res1, i32 1
   %res2 = call i8 @llvm.x86.avx512.mask.ucmp.q.128(<2 x i64> %a0, <2 x i64> %a1, i32 2, i8 %mask)
   %vec2 = insertelement <8 x i8> %vec1, i8 %res2, i32 2

llvm/trunk/test/CodeGen/X86/avx512vlcd-intrinsics-fast-isel.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -fast-isel -mtriple=i686-unknown-unknown -mattr=+avx512cd,+avx512vl | FileCheck %s --check-prefixes=CHECK,X86
 ; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+avx512vl | FileCheck %s --check-prefixes=CHECK,X64
 
 define <2 x i64> @test_mm_broadcastmb_epi64(<2 x i64> %a, <2 x i64> %b) {
 ; X86-LABEL: test_mm_broadcastmb_epi64:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0
 ; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    movzbl %al, %eax
 ; X86-NEXT:    vmovd %eax, %xmm0
 ; X86-NEXT:    vpbroadcastq %xmm0, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_broadcastmb_epi64:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0
 ; X64-NEXT:    vpbroadcastmb2q %k0, %xmm0
entry:
   ret <2 x i64> %vecinit1.i.i
 }
 
 define <4 x i64> @test_mm256_broadcastmb_epi64(<4 x i64> %a, <4 x i64> %b) {
 ; X86-LABEL: test_mm256_broadcastmb_epi64:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0
 ; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    movzbl %al, %eax
 ; X86-NEXT:    vmovd %eax, %xmm0
 ; X86-NEXT:    vpbroadcastq %xmm0, %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_broadcastmb_epi64:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0
 ; X64-NEXT:    vpbroadcastmb2q %k0, %ymm0