[X86][SSE] Add support for <64 x i1> bool reduction

This generalizes the existing <32 x i1> pre-AVX2 split code to support reductions from <64 x i1> as well, we can probably generalize to any larger pow2 case in the future if the (unlikely) need ever arises.

We still need to tweak combineBitcastvxi1 to improve AVX512F codegen as its assumes vXi1 types should be handled on the mask registers even when they aren't legal.

Differential Revision: https://reviews.llvm.org/D67070

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@371328 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 552c91703a7d96e52144bbc9f9f57c908c3fc545..f5c9971a3c582259f0c93498347e74d2c8fd744f 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -35859,11 +35859,12 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
   SDLoc DL(Extract);
   EVT MatchVT = Match.getValueType();
   unsigned NumElts = MatchVT.getVectorNumElements();
+  unsigned MaxElts = Subtarget.hasInt256() ? 32 : 16;
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
 
   if (ExtractVT == MVT::i1) {
     // Special case for (pre-legalization) vXi1 reductions.
-    if (NumElts > 32)
+    if (NumElts > 64 || !isPowerOf2_32(NumElts))
       return SDValue();
     if (TLI.isTypeLegal(MatchVT)) {
       // If this is a legal AVX512 predicate type then we can just bitcast.
@@ -35871,18 +35872,18 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
       Movmsk = DAG.getBitcast(MovmskVT, Match);
     } else {
       // Use combineBitcastvxi1 to create the MOVMSK.
-      if (NumElts == 32 && !Subtarget.hasInt256()) {
+      while (NumElts > MaxElts) {
         SDValue Lo, Hi;
         std::tie(Lo, Hi) = DAG.SplitVector(Match, DL);
         Match = DAG.getNode(BinOp, DL, Lo.getValueType(), Lo, Hi);
-        NumElts = 16;
+        NumElts /= 2;
       }
       EVT MovmskVT = EVT::getIntegerVT(*DAG.getContext(), NumElts);
       Movmsk = combineBitcastvxi1(DAG, MovmskVT, Match, DL, Subtarget);
     }
     if (!Movmsk)
       return SDValue();
-    Movmsk = DAG.getZExtOrTrunc(Movmsk, DL, MVT::i32);
+    Movmsk = DAG.getZExtOrTrunc(Movmsk, DL, NumElts > 32 ? MVT::i64 : MVT::i32);
   } else {
     // Bail with AVX512VL (which uses predicate registers).
     if (Subtarget.hasVLX())
@@ -35923,13 +35924,15 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
     Movmsk = getPMOVMSKB(DL, BitcastLogicOp, DAG, Subtarget);
     NumElts = MaskSrcVT.getVectorNumElements();
   }
-  assert(NumElts <= 32 && "Not expecting more than 32 elements");
+  assert((NumElts <= 32 || NumElts == 64) &&
+         "Not expecting more than 64 elements");
 
+  MVT CmpVT = NumElts == 64 ? MVT::i64 : MVT::i32;
   if (BinOp == ISD::XOR) {
     // parity -> (AND (CTPOP(MOVMSK X)), 1)
-    SDValue Mask = DAG.getConstant(1, DL, MVT::i32);
-    SDValue Result = DAG.getNode(ISD::CTPOP, DL, MVT::i32, Movmsk);
-    Result = DAG.getNode(ISD::AND, DL, MVT::i32, Result, Mask);
+    SDValue Mask = DAG.getConstant(1, DL, CmpVT);
+    SDValue Result = DAG.getNode(ISD::CTPOP, DL, CmpVT, Movmsk);
+    Result = DAG.getNode(ISD::AND, DL, CmpVT, Result, Mask);
     return DAG.getZExtOrTrunc(Result, DL, ExtractVT);
   }
 
@@ -35937,18 +35940,18 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
   ISD::CondCode CondCode;
   if (BinOp == ISD::OR) {
     // any_of -> MOVMSK != 0
-    CmpC = DAG.getConstant(0, DL, MVT::i32);
+    CmpC = DAG.getConstant(0, DL, CmpVT);
     CondCode = ISD::CondCode::SETNE;
   } else {
     // all_of -> MOVMSK == ((1 << NumElts) - 1)
-    CmpC = DAG.getConstant((1ULL << NumElts) - 1, DL, MVT::i32);
+    CmpC = DAG.getConstant((1ULL << NumElts) - 1, DL, CmpVT);
     CondCode = ISD::CondCode::SETEQ;
   }
 
   // The setcc produces an i8 of 0/1, so extend that to the result width and
   // negate to get the final 0/-1 mask value.
   EVT SetccVT =
-      TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), MVT::i32);
+      TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), CmpVT);
   SDValue Setcc = DAG.getSetCC(DL, SetccVT, Movmsk, CmpC, CondCode);
   SDValue Zext = DAG.getZExtOrTrunc(Setcc, DL, ExtractVT);
   SDValue Zero = DAG.getConstant(0, DL, ExtractVT);

diff --git a/test/CodeGen/X86/vector-reduce-and-bool.ll b/test/CodeGen/X86/vector-reduce-and-bool.ll
index 2d5d13d4a6eb0f0624592dec916c3f05600608e5..b9cdb2f022540e3128a51bb6d431f333d6563272 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -804,77 +804,36 @@ define i1 @trunc_v32i16_v32i1(<32 x i16>) {
 }
 
 define i1 @trunc_v64i8_v64i1(<64 x i8>) {
-; SSE2-LABEL: trunc_v64i8_v64i1:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: trunc_v64i8_v64i1:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pand %xmm3, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: trunc_v64i8_v64i1:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pand %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm2, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    psllw $7, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    cmpw $-1, %ax
+; SSE-NEXT:    sete %al
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_v64i8_v64i1:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vandps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vandps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vandps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    cmpw $-1, %ax
+; AVX1-NEXT:    sete %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_v64i8_v64i1:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vpsllw $7, %ymm0, %ymm0
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    cmpl $-1, %eax
+; AVX2-NEXT:    sete %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -906,20 +865,8 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestq %k0, %k0
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -927,20 +874,8 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestq %k0, %k0
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <64 x i8> %0 to <64 x i1>
@@ -1749,9 +1684,8 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; SSE-NEXT:    por %xmm0, %xmm1
 ; SSE-NEXT:    pcmpeqb %xmm3, %xmm1
 ; SSE-NEXT:    pmovmskb %xmm1, %eax
-; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    cmpw $-1, %ax
 ; SSE-NEXT:    sete %al
-; SSE-NEXT:    negb %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: icmp_v64i8_v64i1:
@@ -1762,10 +1696,10 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX1-NEXT:    cmpw $-1, %ax
 ; AVX1-NEXT:    sete %al
-; AVX1-NEXT:    negb %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
@@ -1777,7 +1711,6 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    cmpl $-1, %eax
 ; AVX2-NEXT:    sete %al
-; AVX2-NEXT:    negb %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -1810,40 +1743,16 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW-LABEL: icmp_v64i8_v64i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    kandq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestq %k0, %k0
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v64i8_v64i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    kandq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestq %k0, %k0
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <64 x i8> %0, zeroinitializer

diff --git a/test/CodeGen/X86/vector-reduce-or-bool.ll b/test/CodeGen/X86/vector-reduce-or-bool.ll
index 91157ee54a5d45688195f44759393dc8159f097f..67303054c2ea1b4ab973450f3b2c329f319a18ea 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-or-bool.ll
+++ b/test/CodeGen/X86/vector-reduce-or-bool.ll
@@ -795,77 +795,36 @@ define i1 @trunc_v32i16_v32i1(<32 x i16>) {
 }
 
 define i1 @trunc_v64i8_v64i1(<64 x i8>) {
-; SSE2-LABEL: trunc_v64i8_v64i1:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    por %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm2, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: trunc_v64i8_v64i1:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    por %xmm3, %xmm1
-; SSE41-NEXT:    por %xmm2, %xmm1
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: trunc_v64i8_v64i1:
+; SSE:       # %bb.0:
+; SSE-NEXT:    por %xmm3, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    psllw $7, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    testw %ax, %ax
+; SSE-NEXT:    setne %al
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_v64i8_v64i1:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    testw %ax, %ax
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_v64i8_v64i1:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vpsllw $7, %ymm0, %ymm0
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    testl %eax, %eax
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -897,20 +856,8 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestq %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -918,20 +865,8 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestq %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <64 x i8> %0 to <64 x i1>
@@ -1735,9 +1670,10 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; SSE-NEXT:    por %xmm3, %xmm1
 ; SSE-NEXT:    por %xmm2, %xmm1
 ; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    psllw $7, %xmm1
 ; SSE-NEXT:    pmovmskb %xmm1, %eax
-; SSE-NEXT:    negl %eax
-; SSE-NEXT:    sbbb %al, %al
+; SSE-NEXT:    testw %ax, %ax
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: icmp_v64i8_v64i1:
@@ -1752,9 +1688,10 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm4, %xmm0
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    negl %eax
-; AVX1-NEXT:    sbbb %al, %al
+; AVX1-NEXT:    testw %ax, %ax
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
@@ -1765,8 +1702,8 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX2-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
-; AVX2-NEXT:    negl %eax
-; AVX2-NEXT:    sbbb %al, %al
+; AVX2-NEXT:    testl %eax, %eax
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -1799,40 +1736,16 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW-LABEL: icmp_v64i8_v64i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    korq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestq %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v64i8_v64i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    korq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestq %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <64 x i8> %0, zeroinitializer

diff --git a/test/CodeGen/X86/vector-reduce-xor-bool.ll b/test/CodeGen/X86/vector-reduce-xor-bool.ll
index 36395a1b80b75bb3d30d889e97c561d11aae34c0..1dc839734e0e542f5e16e8b7251c3e28f7e86f45 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-xor-bool.ll
+++ b/test/CodeGen/X86/vector-reduce-xor-bool.ll
@@ -904,77 +904,45 @@ define i1 @trunc_v32i16_v32i1(<32 x i16>) {
 }
 
 define i1 @trunc_v64i8_v64i1(<64 x i8>) {
-; SSE2-LABEL: trunc_v64i8_v64i1:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm3, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: trunc_v64i8_v64i1:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm3, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: trunc_v64i8_v64i1:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm3, %xmm1
+; SSE-NEXT:    pxor %xmm2, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    psllw $7, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    movl %eax, %ecx
+; SSE-NEXT:    shrl $8, %ecx
+; SSE-NEXT:    xorb %al, %cl
+; SSE-NEXT:    setnp %al
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_v64i8_v64i1:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vxorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vxorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vxorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vxorps %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    movl %eax, %ecx
+; AVX1-NEXT:    shrl $8, %ecx
+; AVX1-NEXT:    xorb %al, %cl
+; AVX1-NEXT:    setnp %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_v64i8_v64i1:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vpsllw $7, %ymm0, %ymm0
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    movl %eax, %ecx
+; AVX2-NEXT:    shrl $16, %ecx
+; AVX2-NEXT:    xorl %eax, %ecx
+; AVX2-NEXT:    movl %ecx, %eax
+; AVX2-NEXT:    shrl $8, %eax
+; AVX2-NEXT:    xorb %cl, %al
+; AVX2-NEXT:    setnp %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -1006,20 +974,17 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kmovq %k0, %rax
+; AVX512BW-NEXT:    movq %rax, %rcx
+; AVX512BW-NEXT:    shrq $32, %rcx
+; AVX512BW-NEXT:    xorl %eax, %ecx
+; AVX512BW-NEXT:    movl %ecx, %eax
+; AVX512BW-NEXT:    shrl $16, %eax
+; AVX512BW-NEXT:    xorl %ecx, %eax
+; AVX512BW-NEXT:    movl %eax, %ecx
+; AVX512BW-NEXT:    shrl $8, %ecx
+; AVX512BW-NEXT:    xorb %al, %cl
+; AVX512BW-NEXT:    setnp %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -1027,20 +992,17 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kmovq %k0, %rax
+; AVX512VL-NEXT:    movq %rax, %rcx
+; AVX512VL-NEXT:    shrq $32, %rcx
+; AVX512VL-NEXT:    xorl %eax, %ecx
+; AVX512VL-NEXT:    movl %ecx, %eax
+; AVX512VL-NEXT:    shrl $16, %eax
+; AVX512VL-NEXT:    xorl %ecx, %eax
+; AVX512VL-NEXT:    movl %eax, %ecx
+; AVX512VL-NEXT:    shrl $8, %ecx
+; AVX512VL-NEXT:    xorb %al, %cl
+; AVX512VL-NEXT:    setnp %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <64 x i8> %0 to <64 x i1>
@@ -1943,53 +1905,23 @@ define i1 @icmp_v32i16_v32i1(<32 x i16>) {
 }
 
 define i1 @icmp_v64i8_v64i1(<64 x i8>) {
-; SSE2-LABEL: icmp_v64i8_v64i1:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE2-NEXT:    pcmpeqb %xmm4, %xmm0
-; SSE2-NEXT:    pcmpeqb %xmm4, %xmm3
-; SSE2-NEXT:    pcmpeqb %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm3, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: icmp_v64i8_v64i1:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE41-NEXT:    pcmpeqb %xmm4, %xmm0
-; SSE41-NEXT:    pcmpeqb %xmm4, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
-; SSE41-NEXT:    pxor %xmm3, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: icmp_v64i8_v64i1:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm0
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm3
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE-NEXT:    pxor %xmm3, %xmm1
+; SSE-NEXT:    pxor %xmm2, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    psllw $7, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    movl %eax, %ecx
+; SSE-NEXT:    shrl $8, %ecx
+; SSE-NEXT:    xorb %al, %cl
+; SSE-NEXT:    setnp %al
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: icmp_v64i8_v64i1:
 ; AVX1:       # %bb.0:
@@ -2003,16 +1935,12 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpxor %xmm0, %xmm4, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    movl %eax, %ecx
+; AVX1-NEXT:    shrl $8, %ecx
+; AVX1-NEXT:    xorb %al, %cl
+; AVX1-NEXT:    setnp %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
@@ -2022,18 +1950,14 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX2-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    movl %eax, %ecx
+; AVX2-NEXT:    shrl $16, %ecx
+; AVX2-NEXT:    xorl %eax, %ecx
+; AVX2-NEXT:    movl %ecx, %eax
+; AVX2-NEXT:    shrl $8, %eax
+; AVX2-NEXT:    xorb %cl, %al
+; AVX2-NEXT:    setnp %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -2066,40 +1990,34 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
 ; AVX512BW-LABEL: icmp_v64i8_v64i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512BW-NEXT:    kxorq %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kmovq %k0, %rax
+; AVX512BW-NEXT:    movq %rax, %rcx
+; AVX512BW-NEXT:    shrq $32, %rcx
+; AVX512BW-NEXT:    xorl %eax, %ecx
+; AVX512BW-NEXT:    movl %ecx, %eax
+; AVX512BW-NEXT:    shrl $16, %eax
+; AVX512BW-NEXT:    xorl %ecx, %eax
+; AVX512BW-NEXT:    movl %eax, %ecx
+; AVX512BW-NEXT:    shrl $8, %ecx
+; AVX512BW-NEXT:    xorb %al, %cl
+; AVX512BW-NEXT:    setnp %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v64i8_v64i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512VL-NEXT:    kshiftrq $32, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $16, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $8, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $4, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $2, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrq $1, %k0, %k1
-; AVX512VL-NEXT:    kxorq %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kmovq %k0, %rax
+; AVX512VL-NEXT:    movq %rax, %rcx
+; AVX512VL-NEXT:    shrq $32, %rcx
+; AVX512VL-NEXT:    xorl %eax, %ecx
+; AVX512VL-NEXT:    movl %ecx, %eax
+; AVX512VL-NEXT:    shrl $16, %eax
+; AVX512VL-NEXT:    xorl %ecx, %eax
+; AVX512VL-NEXT:    movl %eax, %ecx
+; AVX512VL-NEXT:    shrl $8, %ecx
+; AVX512VL-NEXT:    xorb %al, %cl
+; AVX512VL-NEXT:    setnp %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <64 x i8> %0, zeroinitializer