[X86][SSE] Add pre-AVX2 support for (i32 bitcast(v32i1)) -> 2xMOVMSK

[RKSimon]

Summary

Currently we only support (i32 bitcast(v32i1)) using the AVX2 VPMOVMSKB ymm instruction.

This patch adds support for splitting pre-AVX2 targets into 2 x (V)PMOVMSKB xmm instructions and merging the integer results.

In future we could probably generalize this to handle more cases.

[filcab]

lib/Target/X86/X86ISelLowering.cpp

29284 ↗

(On Diff #106193)

Suggestion: We can probably split the else if, as I think it makes it easy to miss that the block above already returned. What do you think about:

if (SextVT == MVT::v32i8 && ...) {
  //...
  return ...
}

if (SExtVT == MVT::v8i16) {
  ...

[RKSimon]

@filcab Split up if-else chain

[RKSimon]

@zvi Any comments?

[RKSimon]

ping

[zvi]

Simon, I apologize for the delayed response. Will review ASAP.

[zvi]

LGTM. Thanks!

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

static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
   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:
-    // TODO: Handle pre-AVX2 cases by splitting to two v16i1's.
-    if (!Subtarget.hasInt256())
-      return SDValue();
     SExtVT = MVT::v32i8;
     break;
   };
 
   SDLoc DL(BitCast);
   SDValue V = DAG.getSExtOrTrunc(N0, DL, SExtVT);
+
+  if (SExtVT == MVT::v32i8 && !Subtarget.hasInt256()) {
+    // Handle pre-AVX2 cases by splitting to two v16i1's.
+    const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+    MVT ShiftTy = TLI.getScalarShiftAmountTy(DAG.getDataLayout(), MVT::i32);
+    SDValue Lo = extract128BitVector(V, 0, DAG, DL);
+    SDValue Hi = extract128BitVector(V, 16, DAG, DL);
+    Lo = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, Lo);
+    Hi = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, Hi);
+    Hi = DAG.getNode(ISD::SHL, DL, MVT::i32, Hi,
+                     DAG.getConstant(16, DL, ShiftTy));
+    V = DAG.getNode(ISD::OR, DL, MVT::i32, Lo, Hi);
+    return DAG.getZExtOrTrunc(V, DL, VT);
+  }
+
   if (SExtVT == MVT::v8i16) {
     V = DAG.getBitcast(MVT::v16i8, V);
     V = DAG.getVectorShuffle(
         MVT::v16i8, DL, V, DAG.getUNDEF(MVT::v16i8),
         {0, 2, 4, 6, 8, 10, 12, 14, -1, -1, -1, -1, -1, -1, -1, -1});
   } else
     assert(SExtVT.getScalarType() != MVT::i16 &&
            "Vectors of i16 must be shuffled");

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

 ; SSE2-SSSE3-LABEL: v32i8:
 ; SSE2-SSSE3:       # BB#0:
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm2, %xmm0
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm3, %xmm1
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm6, %xmm4
 ; SSE2-SSSE3-NEXT:    pand %xmm0, %xmm4
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm7, %xmm5
 ; SSE2-SSSE3-NEXT:    pand %xmm1, %xmm5
-; SSE2-SSSE3-NEXT:    movdqa %xmm5, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movdqa %xmm4, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSE2-SSSE3-NEXT:    andb $1, %cl
-; SSE2-SSSE3-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSE2-SSSE3-NEXT:    andb $1, %cl
-; SSE2-SSSE3-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-SSSE3-NEXT:    shll $16, %ecx
-; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSE2-SSSE3-NEXT:    pmovmskb %xmm4, %ecx
+; SSE2-SSSE3-NEXT:    pmovmskb %xmm5, %eax
+; SSE2-SSSE3-NEXT:    shll $16, %eax
 ; SSE2-SSSE3-NEXT:    orl %ecx, %eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v32i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi0:
-; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi1:
-; AVX1-NEXT:    .cfi_offset %rbp, -16
-; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi2:
-; AVX1-NEXT:    .cfi_def_cfa_register %rbp
-; AVX1-NEXT:    andq $-32, %rsp
-; AVX1-NEXT:    subq $32, %rsp
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpcmpgtb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
+; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm1
 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    movl (%rsp), %eax
-; AVX1-NEXT:    movq %rbp, %rsp
-; AVX1-NEXT:    popq %rbp
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v32i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0

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

; AVX512BW-NEXT:    retq
   %y = and <8 x i1> %x0, %x1
   %res = bitcast <8 x i1> %y to i8
   ret i8 %res
 }
 
 define i32 @v32i16(<32 x i16> %a, <32 x i16> %b, <32 x i16> %c, <32 x i16> %d) {
 ; SSE-LABEL: v32i16:
 ; SSE:       # BB#0:
+; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm8
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm8
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    pcmpgtw %xmm5, %xmm1
 ; SSE-NEXT:    movdqa {{.*#+}} xmm5 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
 ; SSE-NEXT:    pshufb %xmm5, %xmm1
 ; SSE-NEXT:    pcmpgtw %xmm4, %xmm0
 ; SSE-NEXT:    pshufb %xmm5, %xmm0
 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    psllw $7, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; SSE-NEXT:    pand %xmm12, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pcmpgtb %xmm0, %xmm4
 ; SSE-NEXT:    pcmpgtw %xmm7, %xmm3
 ; SSE-NEXT:    pshufb %xmm5, %xmm3
 ; SSE-NEXT:    pcmpgtw %xmm6, %xmm2
 ; SSE-NEXT:    pshufb %xmm5, %xmm2
 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE-NEXT:    psllw $7, %xmm2
+; SSE-NEXT:    pand %xmm12, %xmm2
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    pcmpgtb %xmm2, %xmm0
 ; SSE-NEXT:    pcmpgtw {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    pshufb %xmm5, %xmm11
-; SSE-NEXT:    pcmpgtw {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    pshufb %xmm5, %xmm8
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm8 = xmm8[0],xmm11[0]
-; SSE-NEXT:    pand %xmm0, %xmm8
 ; SSE-NEXT:    pcmpgtw {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    pshufb %xmm5, %xmm10
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm10 = xmm10[0],xmm11[0]
+; SSE-NEXT:    psllw $7, %xmm10
+; SSE-NEXT:    pand %xmm12, %xmm10
+; SSE-NEXT:    pxor %xmm2, %xmm2
+; SSE-NEXT:    pcmpgtb %xmm10, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm2
 ; SSE-NEXT:    pcmpgtw {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    pshufb %xmm5, %xmm9
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm10[0]
-; SSE-NEXT:    pand %xmm2, %xmm9
-; SSE-NEXT:    pextrb $15, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $14, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $13, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $11, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $9, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $7, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $5, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $3, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $1, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm9, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $15, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $14, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $13, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $11, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $9, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $7, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $5, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $3, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $1, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm8, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    movl -{{[0-9]+}}(%rsp), %ecx
-; SSE-NEXT:    shll $16, %ecx
-; SSE-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSE-NEXT:    pcmpgtw {{[0-9]+}}(%rsp), %xmm8
+; SSE-NEXT:    pshufb %xmm5, %xmm8
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm8 = xmm8[0],xmm9[0]
+; SSE-NEXT:    psllw $7, %xmm8
+; SSE-NEXT:    pand %xmm12, %xmm8
+; SSE-NEXT:    pcmpgtb %xmm8, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    pmovmskb %xmm2, %ecx
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    shll $16, %eax
 ; SSE-NEXT:    orl %ecx, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: v32i16:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi0:
-; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi1:
-; AVX1-NEXT:    .cfi_offset %rbp, -16
-; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi2:
-; AVX1-NEXT:    .cfi_def_cfa_register %rbp
-; AVX1-NEXT:    andq $-32, %rsp
-; AVX1-NEXT:    subq $32, %rsp
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm8
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm9
 ; AVX1-NEXT:    vpcmpgtw %xmm8, %xmm9, %xmm8
 ; AVX1-NEXT:    vpcmpgtw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpacksswb %xmm8, %xmm1, %xmm8
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpcmpgtw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm8, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpcmpgtw %xmm7, %xmm5, %xmm2
 ; AVX1-NEXT:    vpacksswb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpand %xmm1, %xmm8, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm6, %xmm4, %xmm3
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    movl (%rsp), %eax
-; AVX1-NEXT:    movq %rbp, %rsp
-; AVX1-NEXT:    popq %rbp
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    vpsllw $7, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v32i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
 ; AVX2-NEXT:    vpacksswb %xmm3, %xmm1, %xmm1

 ; SSE-NEXT:    orl %edx, %eax
 ; SSE-NEXT:    shlq $32, %rax
 ; SSE-NEXT:    orq %rcx, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: v64i8:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi3:
+; AVX1-NEXT:  .Lcfi0:
 ; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi4:
+; AVX1-NEXT:  .Lcfi1:
 ; AVX1-NEXT:    .cfi_offset %rbp, -16
 ; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi5:
+; AVX1-NEXT:  .Lcfi2:
 ; AVX1-NEXT:    .cfi_def_cfa_register %rbp
 ; AVX1-NEXT:    andq $-32, %rsp
 ; AVX1-NEXT:    subq $64, %rsp
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm8
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm9
 ; AVX1-NEXT:    vpcmpgtb %xmm8, %xmm9, %xmm8
 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm8, %ymm1, %ymm8

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

; AVX512-NEXT:    retq
   %x = fcmp ogt <8 x float> %a, %b
   %res = bitcast <8 x i1> %x to i8
   ret i8 %res
 }
 
 define i32 @v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; SSE2-SSSE3-LABEL: v32i8:
 ; SSE2-SSSE3:       # BB#0:
-; SSE2-SSSE3-NEXT:    pcmpgtb %xmm3, %xmm1
-; SSE2-SSSE3-NEXT:    movdqa %xmm1, -{{[0-9]+}}(%rsp)
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm2, %xmm0
-; SSE2-SSSE3-NEXT:    movdqa %xmm0, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSE2-SSSE3-NEXT:    andb $1, %cl
-; SSE2-SSSE3-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSE2-SSSE3-NEXT:    andb $1, %cl
-; SSE2-SSSE3-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    andb $1, %al
-; SSE2-SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-SSSE3-NEXT:    movl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-SSSE3-NEXT:    shll $16, %ecx
-; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %ecx
+; SSE2-SSSE3-NEXT:    pcmpgtb %xmm3, %xmm1
+; SSE2-SSSE3-NEXT:    pmovmskb %xmm1, %eax
+; SSE2-SSSE3-NEXT:    shll $16, %eax
 ; SSE2-SSSE3-NEXT:    orl %ecx, %eax
 ; SSE2-SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v32i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi0:
-; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi1:
-; AVX1-NEXT:    .cfi_offset %rbp, -16
-; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi2:
-; AVX1-NEXT:    .cfi_def_cfa_register %rbp
-; AVX1-NEXT:    andq $-32, %rsp
-; AVX1-NEXT:    subq $32, %rsp
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm2, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
+; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
+; AVX1-NEXT:    vpmovmskb %xmm2, %ecx
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    movl (%rsp), %eax
-; AVX1-NEXT:    movq %rbp, %rsp
-; AVX1-NEXT:    popq %rbp
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v32i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper

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

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
 
 define i32 @v32i16(<32 x i16> %a, <32 x i16> %b) {
 ; SSE-LABEL: v32i16:
 ; SSE:       # BB#0:
-; SSE-NEXT:    pcmpgtw %xmm7, %xmm3
-; SSE-NEXT:    pextrb $14, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm3, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pcmpgtw %xmm6, %xmm2
-; SSE-NEXT:    pextrb $14, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm2, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
 ; SSE-NEXT:    pcmpgtw %xmm5, %xmm1
-; SSE-NEXT:    pextrb $14, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
 ; SSE-NEXT:    pcmpgtw %xmm4, %xmm0
-; SSE-NEXT:    pextrb $14, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $12, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $10, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $8, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $6, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $4, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $2, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    andb $1, %al
-; SSE-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE-NEXT:    movl -{{[0-9]+}}(%rsp), %ecx
-; SSE-NEXT:    shll $16, %ecx
-; SSE-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSE-NEXT:    packsswb %xmm1, %xmm0
+; SSE-NEXT:    pmovmskb %xmm0, %ecx
+; SSE-NEXT:    pcmpgtw %xmm7, %xmm3
+; SSE-NEXT:    pcmpgtw %xmm6, %xmm2
+; SSE-NEXT:    packsswb %xmm3, %xmm2
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    shll $16, %eax
 ; SSE-NEXT:    orl %ecx, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: v32i16:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi0:
-; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi1:
-; AVX1-NEXT:    .cfi_offset %rbp, -16
-; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi2:
-; AVX1-NEXT:    .cfi_def_cfa_register %rbp
-; AVX1-NEXT:    andq $-32, %rsp
-; AVX1-NEXT:    subq $32, %rsp
-; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpcmpgtw %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpextrb $14, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm4, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpcmpgtw %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, (%rsp)
-; AVX1-NEXT:    movl (%rsp), %eax
-; AVX1-NEXT:    movq %rbp, %rsp
-; AVX1-NEXT:    popq %rbp
+; AVX1-NEXT:    vpacksswb %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpcmpgtw %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpcmpgtw %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpacksswb %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: v32i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpgtw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0

 ; SSE-NEXT:    orl %edx, %eax
 ; SSE-NEXT:    shlq $32, %rax
 ; SSE-NEXT:    orq %rcx, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: v64i8:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    pushq %rbp
-; AVX1-NEXT:  .Lcfi3:
+; AVX1-NEXT:  .Lcfi0:
 ; AVX1-NEXT:    .cfi_def_cfa_offset 16
-; AVX1-NEXT:  .Lcfi4:
+; AVX1-NEXT:  .Lcfi1:
 ; AVX1-NEXT:    .cfi_offset %rbp, -16
 ; AVX1-NEXT:    movq %rsp, %rbp
-; AVX1-NEXT:  .Lcfi5:
+; AVX1-NEXT:  .Lcfi2:
 ; AVX1-NEXT:    .cfi_def_cfa_register %rbp
 ; AVX1-NEXT:    andq $-32, %rsp
 ; AVX1-NEXT:    subq $64, %rsp
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpcmpgtb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpextrb $15, %xmm4, %eax
 ; AVX1-NEXT:    andb $1, %al