[X86][AVX] Push sign extensions of comparison bool results through bitops (PR42025)

[RKSimon]

Summary

As discussed on PR42025, with more complex boolean math we can end up with many truncations/extensions of the comparison results through each bitop.

This patch handles the cases introduced in combineBitcastvxi1 by pushing the sign extension through the AND/OR/XOR ops so its just the original SETCC ops that gets extended.

[RKSimon]

ping?

[craig.topper]

LGTM

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

static bool checkBitcastSrcVectorSize(SDValue Src, unsigned Size) {
   case ISD::XOR:
   case ISD::OR:
     return checkBitcastSrcVectorSize(Src.getOperand(0), Size) &&
            checkBitcastSrcVectorSize(Src.getOperand(1), Size);
   }
   return false;
 }
 
+// Helper to push sign extension of vXi1 SETCC result through bitops.
+static SDValue signExtendBitcastSrcVector(SelectionDAG &DAG, EVT SExtVT,
+                                          SDValue Src, const SDLoc &DL) {
+  switch (Src.getOpcode()) {
+  case ISD::SETCC:
+    return DAG.getNode(ISD::SIGN_EXTEND, DL, SExtVT, Src);
+  case ISD::AND:
+  case ISD::XOR:
+  case ISD::OR:
+    return DAG.getNode(
+        Src.getOpcode(), DL, SExtVT,
+        signExtendBitcastSrcVector(DAG, SExtVT, Src.getOperand(0), DL),
+        signExtendBitcastSrcVector(DAG, SExtVT, Src.getOperand(1), DL));
+  }
+  llvm_unreachable("Unexpected node type for vXi1 sign extension");
+}
+
 // Try to match patterns such as
 // (i16 bitcast (v16i1 x))
 // ->
 // (i16 movmsk (16i8 sext (v16i1 x)))
 // before the illegal vector is scalarized on subtargets that don't have legal
 // vxi1 types.
 static SDValue combineBitcastvxi1(SelectionDAG &DAG, EVT VT, SDValue Src,
                                   const SDLoc &DL,
static SDValue combineBitcastvxi1(SelectionDAG &DAG, EVT VT, SDValue Src,
   // For these two cases, we can shuffle the upper element bytes to a
   // consecutive sequence at the start of the vector and treat the results as
   // v16i8 or v32i8, and for v16i8 this is the preferable solution. However,
   // for v16i16 this is not the case, because the shuffle is expensive, so we
   // avoid sign-extending to this type entirely.
   // For example, t0 := (v8i16 sext(v8i1 x)) needs to be shuffled as:
   // (v16i8 shuffle <0,2,4,6,8,10,12,14,u,u,...,u> (v16i8 bitcast t0), undef)
   MVT SExtVT;
+  bool PropagateSExt = false;
   switch (SrcVT.getSimpleVT().SimpleTy) {
   default:
     return SDValue();
   case MVT::v2i1:
     SExtVT = MVT::v2i64;
     break;
   case MVT::v4i1:
     SExtVT = MVT::v4i32;
     // For cases such as (i4 bitcast (v4i1 setcc v4i64 v1, v2))
     // sign-extend to a 256-bit operation to avoid truncation.
-    if (Subtarget.hasAVX() && checkBitcastSrcVectorSize(Src, 256))
+    if (Subtarget.hasAVX() && checkBitcastSrcVectorSize(Src, 256)) {
       SExtVT = MVT::v4i64;
+      PropagateSExt = true;
+    }
     break;
   case MVT::v8i1:
     SExtVT = MVT::v8i16;
     // For cases such as (i8 bitcast (v8i1 setcc v8i32 v1, v2)),
     // sign-extend to a 256-bit operation to match the compare.
     // If the setcc operand is 128-bit, prefer sign-extending to 128-bit over
     // 256-bit because the shuffle is cheaper than sign extending the result of
     // the compare.
-    // TODO : use checkBitcastSrcVectorSize
-    if (Src.getOpcode() == ISD::SETCC && Subtarget.hasAVX() &&
-        (Src.getOperand(0).getValueType().is256BitVector() ||
-         Src.getOperand(0).getValueType().is512BitVector())) {
+    if (Subtarget.hasAVX() && (checkBitcastSrcVectorSize(Src, 256) ||
+                               checkBitcastSrcVectorSize(Src, 512))) {
       SExtVT = MVT::v8i32;
+      PropagateSExt = true;
     }
     break;
   case MVT::v16i1:
     SExtVT = MVT::v16i8;
     // For the case (i16 bitcast (v16i1 setcc v16i16 v1, v2)),
     // it is not profitable to sign-extend to 256-bit because this will
     // require an extra cross-lane shuffle which is more expensive than
     // truncating the result of the compare to 128-bits.
     break;
   case MVT::v32i1:
     SExtVT = MVT::v32i8;
     break;
   case MVT::v64i1:
     // If we have AVX512F, but not AVX512BW and the input is truncated from
     // v64i8 checked earlier. Then split the input and make two pmovmskbs.
     if (Subtarget.hasAVX512() && !Subtarget.hasBWI()) {
       SExtVT = MVT::v64i8;
       break;
     }
     return SDValue();
   };
 
-  SDValue V = DAG.getNode(ISD::SIGN_EXTEND, DL, SExtVT, Src);
+  SDValue V = PropagateSExt ? signExtendBitcastSrcVector(DAG, SExtVT, Src, DL)
+                            : DAG.getNode(ISD::SIGN_EXTEND, DL, SExtVT, Src);
 
   if (SExtVT == MVT::v16i8 || SExtVT == MVT::v32i8 || SExtVT == MVT::v64i8) {
     V = getPMOVMSKB(DL, V, DAG, Subtarget);
   } else {
     if (SExtVT == MVT::v8i16)
       V = DAG.getNode(X86ISD::PACKSS, DL, MVT::v16i8, V,
                       DAG.getUNDEF(MVT::v8i16));
     V = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, V);

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

 ; SSE2-SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm1[0,2]
 ; SSE2-SSSE3-NEXT:    andps %xmm0, %xmm2
 ; SSE2-SSSE3-NEXT:    movmskps %xmm2, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm1
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vandpd %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskpd %ymm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v4i64:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm2, %ymm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vmovmskpd %ymm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtq %ymm1, %ymm0, %k1

 ; SSE2-SSSE3-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2],xmm7[0,2]
 ; SSE2-SSSE3-NEXT:    andps %xmm2, %xmm6
 ; SSE2-SSSE3-NEXT:    movmskps %xmm6, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX12-LABEL: v4f64:
 ; AVX12:       # %bb.0:
+; AVX12-NEXT:    vcmpltpd %ymm2, %ymm3, %ymm2
 ; AVX12-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
-; AVX12-NEXT:    vcmpltpd %ymm2, %ymm3, %ymm1
-; AVX12-NEXT:    vandpd %ymm1, %ymm0, %ymm0
+; AVX12-NEXT:    vandpd %ymm2, %ymm0, %ymm0
 ; AVX12-NEXT:    vmovmskpd %ymm0, %eax
 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX12-NEXT:    vzeroupper
 ; AVX12-NEXT:    retq
 ;
 ; AVX512F-LABEL: v4f64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpltpd %ymm0, %ymm1, %k1

 ; SSE2-SSSE3-NEXT:    pand %xmm0, %xmm4
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm4
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm4, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v8i32_and:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
+; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v8i32_and:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8i32_and:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtd %ymm1, %ymm0, %k1
 ; AVX512F-NEXT:    vpcmpgtd %ymm3, %ymm2, %k0 {%k1}

 ; SSE2-SSSE3-NEXT:    por %xmm0, %xmm4
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm4
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm4, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v8i32_or:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
+; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v8i32_or:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8i32_or:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; AVX512F-NEXT:    vpcmpgtd %ymm3, %ymm2, %k1

 ; SSE2-SSSE3-NEXT:    pand %xmm6, %xmm8
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm8
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm8, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v8i32_or_and:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm6, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm6
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm7
+; AVX1-NEXT:    vpcmpgtd %xmm6, %xmm7, %xmm6
 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm6, %xmm3
+; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm4, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v8i32_or_and:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpgtd %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpcmpgtd %ymm2, %ymm3, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm5, %ymm4, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8i32_or_and:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; AVX512F-NEXT:    vpcmpgtd %ymm2, %ymm3, %k1

 ; SSE2-SSSE3-NEXT:    pand %xmm2, %xmm6
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm6
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm6, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX12-LABEL: v8f32_and:
 ; AVX12:       # %bb.0:
+; AVX12-NEXT:    vcmpltps %ymm2, %ymm3, %ymm2
 ; AVX12-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX12-NEXT:    vcmpltps %ymm2, %ymm3, %ymm1
-; AVX12-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX12-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX12-NEXT:    vpmovmskb %xmm0, %eax
+; AVX12-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX12-NEXT:    vmovmskps %ymm0, %eax
 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX12-NEXT:    vzeroupper
 ; AVX12-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8f32_and:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpltps %ymm0, %ymm1, %k1
 ; AVX512F-NEXT:    vcmpltps %ymm2, %ymm3, %k0 {%k1}

 ; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm6
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm6
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm6, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX12-LABEL: v8f32_xor:
 ; AVX12:       # %bb.0:
+; AVX12-NEXT:    vcmpltps %ymm2, %ymm3, %ymm2
 ; AVX12-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX12-NEXT:    vcmpltps %ymm2, %ymm3, %ymm1
-; AVX12-NEXT:    vxorps %ymm1, %ymm0, %ymm0
-; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX12-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX12-NEXT:    vpmovmskb %xmm0, %eax
+; AVX12-NEXT:    vxorps %ymm2, %ymm0, %ymm0
+; AVX12-NEXT:    vmovmskps %ymm0, %eax
 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX12-NEXT:    vzeroupper
 ; AVX12-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8f32_xor:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpltps %ymm0, %ymm1, %k0
 ; AVX512F-NEXT:    vcmpltps %ymm2, %ymm3, %k1

 ; SSE2-SSSE3-NEXT:    pand %xmm4, %xmm8
 ; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm8
 ; SSE2-SSSE3-NEXT:    pmovmskb %xmm8, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX12-LABEL: v8f32_xor_and:
 ; AVX12:       # %bb.0:
+; AVX12-NEXT:    vcmpeq_uqps %ymm3, %ymm2, %ymm2
 ; AVX12-NEXT:    vcmpnleps %ymm1, %ymm0, %ymm0
-; AVX12-NEXT:    vcmpeq_uqps %ymm3, %ymm2, %ymm1
-; AVX12-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX12-NEXT:    vxorps %ymm2, %ymm0, %ymm0
 ; AVX12-NEXT:    vcmpltps %ymm4, %ymm5, %ymm1
-; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX12-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX12-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX12-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX12-NEXT:    vpmovmskb %xmm0, %eax
+; AVX12-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX12-NEXT:    vmovmskps %ymm0, %eax
 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX12-NEXT:    vzeroupper
 ; AVX12-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8f32_xor_and:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpnleps %ymm1, %ymm0, %k0
 ; AVX512F-NEXT:    vcmpeq_uqps %ymm3, %ymm2, %k1

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

 ; SSE-NEXT:    pand %xmm0, %xmm3
 ; SSE-NEXT:    packsswb %xmm0, %xmm3
 ; SSE-NEXT:    pmovmskb %xmm3, %eax
 ; SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: v8i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm8
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm9
+; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm8
+; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm9
 ; AVX1-NEXT:    vpcmpgtq %xmm8, %xmm9, %xmm8
+; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpackssdw %xmm8, %xmm5, %xmm8
+; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm7
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpcmpgtq %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpackssdw %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm8, %ymm4, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm8, %xmm1, %xmm8
+; AVX1-NEXT:    vpackssdw %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm8, %xmm0, %xmm0
-; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm5, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm6, %xmm4, %xmm3
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vandps %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v8i64:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpgtq %ymm7, %ymm5, %ymm5
+; AVX2-NEXT:    vpcmpgtq %ymm6, %ymm4, %ymm4
+; AVX2-NEXT:    vpackssdw %ymm5, %ymm4, %ymm4
 ; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm1, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpcmpgtq %ymm7, %ymm5, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpgtq %ymm6, %ymm4, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
 ; AVX512F-NEXT:    vpcmpgtq %zmm3, %zmm2, %k0 {%k1}

 ; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
 ; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
 ; SSE-NEXT:    pand %xmm0, %xmm3
 ; SSE-NEXT:    packsswb %xmm0, %xmm3
 ; SSE-NEXT:    pmovmskb %xmm3, %eax
 ; SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE-NEXT:    retq
 ;
-; AVX12-LABEL: v8f64:
-; AVX12:       # %bb.0:
-; AVX12-NEXT:    vcmpltpd %ymm1, %ymm3, %ymm1
-; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX12-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; AVX12-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
-; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX12-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX12-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX12-NEXT:    vcmpltpd %ymm5, %ymm7, %ymm1
-; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX12-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX12-NEXT:    vcmpltpd %ymm4, %ymm6, %ymm2
-; AVX12-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX12-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
-; AVX12-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX12-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX12-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX12-NEXT:    vpmovmskb %xmm0, %eax
-; AVX12-NEXT:    # kill: def $al killed $al killed $eax
-; AVX12-NEXT:    vzeroupper
-; AVX12-NEXT:    retq
+; AVX1-LABEL: v8f64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vcmpltpd %ymm5, %ymm7, %ymm5
+; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm7
+; AVX1-NEXT:    vpackssdw %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vcmpltpd %ymm4, %ymm6, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm6
+; AVX1-NEXT:    vpackssdw %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm4, %ymm4
+; AVX1-NEXT:    vcmpltpd %ymm1, %ymm3, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vandps %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
+; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: v8f64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vcmpltpd %ymm5, %ymm7, %ymm5
+; AVX2-NEXT:    vcmpltpd %ymm4, %ymm6, %ymm4
+; AVX2-NEXT:    vpackssdw %ymm5, %ymm4, %ymm4
+; AVX2-NEXT:    vcmpltpd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
+; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
+; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: v8f64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
 ; AVX512F-NEXT:    vcmpltpd %zmm2, %zmm3, %k0 {%k1}
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512F-NEXT:    vzeroupper