[X86][SSE] Prefer trunc(movd(x)) to pextrb(x,0)

[RKSimon]

Summary

If we're extracting the 0'th index of a v16i8 vector we're better off using MOVD than PEXTRB, unless we're storing the value or we require the implicit zero extension of PEXTRB.

The biggest perf diff is on SLM targets where MOVD (uops=1, lat=3 tp=1) is notably faster than PEXTRB (uops=2, lat=5, tp=4).

This matches what we already do for PEXTRW.

[lebedev.ri]

Test changes seem to be omitted from the diff?

[RKSimon]

In D76138#1921775, @lebedev.ri wrote:

Test changes seem to be omitted from the diff?

I loathe git at times :-)

[RKSimon]

rebase, with tests this time

[craig.topper]

LGTM

llvm/lib/Target/X86/X86ISelLowering.cpp

static SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) {
   SDValue Idx = Op.getOperand(1);
   assert(isa<ConstantSDNode>(Idx) && "Constant index expected");
   SDLoc dl(Op);
 
   if (!Vec.getSimpleValueType().is128BitVector())
     return SDValue();
 
   if (VT.getSizeInBits() == 8) {
+    // If IdxVal is 0, it's cheaper to do a move instead of a pextrb, unless
+    // we're going to zero extend the register or fold the store.
+    if (llvm::isNullConstant(Idx) && !MayFoldIntoZeroExtend(Op) &&
+        !MayFoldIntoStore(Op))
+      return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
+                         DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32,
+                                     DAG.getBitcast(MVT::v4i32, Vec), Idx));
+
     SDValue Extract = DAG.getNode(X86ISD::PEXTRB, dl, MVT::i32, Vec, Idx);
     return DAG.getNode(ISD::TRUNCATE, dl, VT, Extract);
   }
 
   if (VT == MVT::f32) {
     // EXTRACTPS outputs to a GPR32 register which will require a movd to copy
     // the result back to FR32 register. It's only worth matching if the
     // result has a single use which is a store or a bitcast to i32.  And in

llvm/test/CodeGen/X86/avg.ll

 ; AVX512BW-NEXT:    vmovq %xmm3, %rbx
 ; AVX512BW-NEXT:    vpextrq $1, %xmm3, %rbp
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm3, %xmm3
 ; AVX512BW-NEXT:    vmovq %xmm3, %rdi
 ; AVX512BW-NEXT:    vpextrq $1, %xmm3, %rsi
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX512BW-NEXT:    vmovq %xmm2, %rdx
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r10
+; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r15
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
 ; AVX512BW-NEXT:    vmovq %xmm2, %r8
 ; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r9
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX512BW-NEXT:    vmovq %xmm2, %r14
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r12
-; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
 ; AVX512BW-NEXT:    vmovq %xmm2, %r11
+; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r10
+; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
+; AVX512BW-NEXT:    vmovq %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    vpextrq $1, %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX512BW-NEXT:    vmovq %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    vpextrq $1, %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX512BW-NEXT:    vmovq %xmm3, %rax
-; AVX512BW-NEXT:    addq %rbx, %rax
-; AVX512BW-NEXT:    movq %rax, %rbx
-; AVX512BW-NEXT:    vpextrq $1, %xmm3, %r13
-; AVX512BW-NEXT:    addq %rbp, %r13
-; AVX512BW-NEXT:    vextracti128 $1, %ymm3, %xmm3
 ; AVX512BW-NEXT:    vmovq %xmm3, %rcx
-; AVX512BW-NEXT:    addq %rdi, %rcx
+; AVX512BW-NEXT:    addq %rbx, %rcx
+; AVX512BW-NEXT:    vpextrq $1, %xmm3, %rax
+; AVX512BW-NEXT:    addq %rbp, %rax
+; AVX512BW-NEXT:    movq %rax, %rbp
+; AVX512BW-NEXT:    vextracti128 $1, %ymm3, %xmm3
+; AVX512BW-NEXT:    vmovq %xmm3, %r14
+; AVX512BW-NEXT:    addq %rdi, %r14
 ; AVX512BW-NEXT:    vpextrq $1, %xmm3, %rax
 ; AVX512BW-NEXT:    addq %rsi, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX512BW-NEXT:    vmovq %xmm2, %r15
-; AVX512BW-NEXT:    addq %rdx, %r15
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rbp
-; AVX512BW-NEXT:    addq %r10, %rbp
+; AVX512BW-NEXT:    vmovq %xmm2, %rax
+; AVX512BW-NEXT:    addq %rdx, %rax
+; AVX512BW-NEXT:    movq %rax, %rdx
+; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r12
+; AVX512BW-NEXT:    addq %r15, %r12
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
 ; AVX512BW-NEXT:    vmovq %xmm2, %rax
 ; AVX512BW-NEXT:    addq %r8, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rax
 ; AVX512BW-NEXT:    addq %r9, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX512BW-NEXT:    vmovq %xmm2, %rax
-; AVX512BW-NEXT:    addq %r14, %rax
+; AVX512BW-NEXT:    addq %r11, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rax
-; AVX512BW-NEXT:    addq %r12, %rax
+; AVX512BW-NEXT:    addq %r10, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm2
-; AVX512BW-NEXT:    vmovq %xmm2, %r12
-; AVX512BW-NEXT:    addq %r11, %r12
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %r11
-; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Folded Reload
+; AVX512BW-NEXT:    vmovq %xmm2, %r13
+; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
+; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rbx
+; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX512BW-NEXT:    vmovq %xmm0, %r9
+; AVX512BW-NEXT:    vmovq %xmm0, %r10
+; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
+; AVX512BW-NEXT:    vpextrq $1, %xmm0, %r9
 ; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %r8
-; AVX512BW-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
 ; AVX512BW-NEXT:    vmovq %xmm1, %rax
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, %rsi
-; AVX512BW-NEXT:    addq %rax, %rsi
+; AVX512BW-NEXT:    vmovq %xmm0, %r8
+; AVX512BW-NEXT:    addq %rax, %r8
 ; AVX512BW-NEXT:    vpextrq $1, %xmm1, %rdi
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX512BW-NEXT:    addq %rdi, %rdx
-; AVX512BW-NEXT:    addq $-1, %rbx
-; AVX512BW-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512BW-NEXT:    movl $0, %r10d
-; AVX512BW-NEXT:    adcq $-1, %r10
-; AVX512BW-NEXT:    addq $-1, %r13
-; AVX512BW-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512BW-NEXT:    movl $0, %eax
-; AVX512BW-NEXT:    adcq $-1, %rax
-; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rsi
+; AVX512BW-NEXT:    addq %rdi, %rsi
 ; AVX512BW-NEXT:    addq $-1, %rcx
-; AVX512BW-NEXT:    movq %rcx, (%rsp) # 8-byte Spill
-; AVX512BW-NEXT:    movl $0, %r14d
-; AVX512BW-NEXT:    adcq $-1, %r14
-; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX512BW-NEXT:    movl $0, %ebx
-; AVX512BW-NEXT:    adcq $-1, %rbx
-; AVX512BW-NEXT:    addq $-1, %r15
-; AVX512BW-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512BW-NEXT:    movl $0, %r13d
-; AVX512BW-NEXT:    adcq $-1, %r13
+; AVX512BW-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    movl $0, %r11d
+; AVX512BW-NEXT:    adcq $-1, %r11
 ; AVX512BW-NEXT:    addq $-1, %rbp
 ; AVX512BW-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    movl $0, %edi
+; AVX512BW-NEXT:    adcq $-1, %rdi
+; AVX512BW-NEXT:    addq $-1, %r14
+; AVX512BW-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    movl $0, %r15d
 ; AVX512BW-NEXT:    adcq $-1, %r15
 ; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; AVX512BW-NEXT:    movl $0, %r14d
+; AVX512BW-NEXT:    adcq $-1, %r14
+; AVX512BW-NEXT:    addq $-1, %rdx
+; AVX512BW-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    movl $0, %eax
+; AVX512BW-NEXT:    adcq $-1, %rax
+; AVX512BW-NEXT:    movq %rax, (%rsp) # 8-byte Spill
+; AVX512BW-NEXT:    addq $-1, %r12
+; AVX512BW-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    movl $0, %r12d
+; AVX512BW-NEXT:    adcq $-1, %r12
+; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512BW-NEXT:    addq $-1, %r12
+; AVX512BW-NEXT:    addq $-1, %r13
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512BW-NEXT:    addq $-1, %r11
+; AVX512BW-NEXT:    addq $-1, %rbx
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512BW-NEXT:    addq $-1, %r10
+; AVX512BW-NEXT:    movl $0, %edx
+; AVX512BW-NEXT:    adcq $-1, %rdx
 ; AVX512BW-NEXT:    addq $-1, %r9
 ; AVX512BW-NEXT:    movl $0, %ecx
 ; AVX512BW-NEXT:    adcq $-1, %rcx
 ; AVX512BW-NEXT:    addq $-1, %r8
 ; AVX512BW-NEXT:    movl $0, %eax
 ; AVX512BW-NEXT:    adcq $-1, %rax
 ; AVX512BW-NEXT:    addq $-1, %rsi
 ; AVX512BW-NEXT:    movl $0, %ebp
 ; AVX512BW-NEXT:    adcq $-1, %rbp
-; AVX512BW-NEXT:    addq $-1, %rdx
-; AVX512BW-NEXT:    movl $0, %edi
-; AVX512BW-NEXT:    adcq $-1, %rdi
-; AVX512BW-NEXT:    shldq $63, %rdx, %rdi
-; AVX512BW-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    shldq $63, %rsi, %rbp
+; AVX512BW-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    shldq $63, %r8, %rax
 ; AVX512BW-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; AVX512BW-NEXT:    shldq $63, %r9, %rcx
-; AVX512BW-NEXT:    movq %rcx, %r8
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %r11, %r9
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %r12, %r11
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
+; AVX512BW-NEXT:    movq %rcx, %rbp
+; AVX512BW-NEXT:    shldq $63, %r10, %rdx
+; AVX512BW-NEXT:    movq %rdx, %r9
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rbx, %r10
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %r13, %r8
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %r12
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %rdi
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rax, %r13
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %rdx
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rax, %rbx
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
 ; AVX512BW-NEXT:    shldq $63, %rax, %rsi
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %r15
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rax, %rdx
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %r13
+; AVX512BW-NEXT:    shldq $63, %rax, %r12
+; AVX512BW-NEXT:    movq (%rsp), %rcx # 8-byte Reload
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rax, %rcx
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %rbx
-; AVX512BW-NEXT:    movq (%rsp), %rax # 8-byte Reload
 ; AVX512BW-NEXT:    shldq $63, %rax, %r14
-; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %rcx
+; AVX512BW-NEXT:    shldq $63, %rax, %r15
 ; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512BW-NEXT:    shldq $63, %rax, %r10
-; AVX512BW-NEXT:    vmovq %r10, %xmm0
-; AVX512BW-NEXT:    vmovq %rcx, %xmm1
+; AVX512BW-NEXT:    shldq $63, %rax, %rdi
+; AVX512BW-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX512BW-NEXT:    shldq $63, %rax, %r11
+; AVX512BW-NEXT:    vmovq %r11, %xmm0
+; AVX512BW-NEXT:    vmovq %rdi, %xmm1
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX512BW-NEXT:    vmovd %eax, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX512BW-NEXT:    vpinsrb $1, %eax, %xmm2, %xmm1
-; AVX512BW-NEXT:    vmovq %r14, %xmm2
-; AVX512BW-NEXT:    vmovq %rbx, %xmm3
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
+; AVX512BW-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm1
+; AVX512BW-NEXT:    vmovq %r15, %xmm2
+; AVX512BW-NEXT:    vmovq %r14, %xmm3
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovq %r13, %xmm1
-; AVX512BW-NEXT:    vmovq %r15, %xmm2
+; AVX512BW-NEXT:    vmovq %rcx, %xmm1
+; AVX512BW-NEXT:    vmovq %r12, %xmm2
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %rsi, %xmm2
-; AVX512BW-NEXT:    vmovq %rdx, %xmm3
+; AVX512BW-NEXT:    vmovq %rdx, %xmm2
+; AVX512BW-NEXT:    vmovq %rsi, %xmm3
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %rdi, %xmm1
-; AVX512BW-NEXT:    vmovq %r12, %xmm2
+; AVX512BW-NEXT:    vmovq %rbx, %xmm1
+; AVX512BW-NEXT:    vmovq %r13, %xmm2
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $8, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $9, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %r11, %xmm2
-; AVX512BW-NEXT:    vmovq %r9, %xmm3
+; AVX512BW-NEXT:    vmovq %r8, %xmm2
+; AVX512BW-NEXT:    vmovq %r10, %xmm3
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $10, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $11, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %r8, %xmm1
-; AVX512BW-NEXT:    vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 8-byte Folded Reload
-; AVX512BW-NEXT:    # xmm2 = mem[0],zero
+; AVX512BW-NEXT:    vmovq %r9, %xmm1
+; AVX512BW-NEXT:    vmovq %rbp, %xmm2
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %rbp, %xmm2
+; AVX512BW-NEXT:    vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 8-byte Folded Reload
+; AVX512BW-NEXT:    # xmm2 = mem[0],zero
 ; AVX512BW-NEXT:    vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 8-byte Folded Reload
 ; AVX512BW-NEXT:    # xmm3 = mem[0],zero
 ; AVX512BW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vmovdqu %xmm0, (%rax)
 ; AVX512BW-NEXT:    addq $24, %rsp
 ; AVX512BW-NEXT:    popq %rbx
 ; AVX512BW-NEXT:    popq %r12
 ; AVX512BW-NEXT:    popq %r13
 ; AVX512BW-NEXT:    popq %r14
 ; AVX512BW-NEXT:    popq %r15

llvm/test/CodeGen/X86/avx512-vec3-crash.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mcpu=skx -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
 
 ; This test crashed during type legalization of SETCC result type.
 define <3 x i8 > @foo(<3 x i8>%x, <3 x i8>%a, <3 x i8>%b) {
 ; CHECK-LABEL: foo:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovd %edi, %xmm0
 ; CHECK-NEXT:    vpinsrb $1, %esi, %xmm0, %xmm0
 ; CHECK-NEXT:    vpinsrb $2, %edx, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovd %ecx, %xmm1
 ; CHECK-NEXT:    vpinsrb $1, %r8d, %xmm1, %xmm1
 ; CHECK-NEXT:    vpinsrb $2, %r9d, %xmm1, %xmm1
 ; CHECK-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; CHECK-NEXT:    vpextrb $0, %xmm0, %eax
+; CHECK-NEXT:    vmovd %xmm0, %eax
 ; CHECK-NEXT:    vpextrb $1, %xmm0, %edx
 ; CHECK-NEXT:    vpextrb $2, %xmm0, %ecx
 ; CHECK-NEXT:    # kill: def $al killed $al killed $eax
 ; CHECK-NEXT:    # kill: def $dl killed $dl killed $edx
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    retq
   %cmp.i = icmp slt <3 x i8> %x, %a
   %res = sext <3 x i1> %cmp.i to <3 x i8>
   ret <3 x i8> %res
 }
 

llvm/test/CodeGen/X86/bitcast-vector-bool.ll

 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX12-NEXT:    retq
 ;
 ; AVX512-LABEL: bitcast_v16i8_to_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovb2m %xmm0, %k0
 ; AVX512-NEXT:    kmovw %k0, -{{[0-9]+}}(%rsp)
 ; AVX512-NEXT:    vmovdqa -{{[0-9]+}}(%rsp), %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %ecx
+; AVX512-NEXT:    vmovd %xmm0, %ecx
 ; AVX512-NEXT:    vpextrb $1, %xmm0, %eax
 ; AVX512-NEXT:    addb %cl, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = icmp slt <16 x i8> %a0, zeroinitializer
   %2 = bitcast <16 x i1> %1 to <2 x i8>
   %3 = extractelement <2 x i8> %2, i32 0
   %4 = extractelement <2 x i8> %2, i32 1

 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bitcast_v16i16_to_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovw2m %ymm0, %k0
 ; AVX512-NEXT:    kmovw %k0, -{{[0-9]+}}(%rsp)
 ; AVX512-NEXT:    vmovdqa -{{[0-9]+}}(%rsp), %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %ecx
+; AVX512-NEXT:    vmovd %xmm0, %ecx
 ; AVX512-NEXT:    vpextrb $1, %xmm0, %eax
 ; AVX512-NEXT:    addb %cl, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = icmp slt <16 x i16> %a0, zeroinitializer
   %2 = bitcast <16 x i1> %1 to <2 x i8>
   %3 = extractelement <2 x i8> %2, i32 0

 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bitcast_v16i32_to_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpcmpgtd %zmm0, %zmm1, %k0
 ; AVX512-NEXT:    kmovw %k0, -{{[0-9]+}}(%rsp)
 ; AVX512-NEXT:    vmovdqa -{{[0-9]+}}(%rsp), %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %ecx
+; AVX512-NEXT:    vmovd %xmm0, %ecx
 ; AVX512-NEXT:    vpextrb $1, %xmm0, %eax
 ; AVX512-NEXT:    addb %cl, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = icmp slt <16 x i32> %a0, zeroinitializer
   %2 = bitcast <16 x i1> %1 to <2 x i8>
   %3 = extractelement <2 x i8> %2, i32 0

llvm/test/CodeGen/X86/buildvec-insertvec.ll

 ; SSE2-NEXT:    movd %xmm0, (%rdi)
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: foo:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    cvttps2dq %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $8, %xmm0, %eax
 ; SSE41-NEXT:    pextrb $4, %xmm0, %ecx
-; SSE41-NEXT:    pextrb $0, %xmm0, %edx
-; SSE41-NEXT:    movd %edx, %xmm0
 ; SSE41-NEXT:    pinsrb $1, %ecx, %xmm0
 ; SSE41-NEXT:    pinsrb $2, %eax, %xmm0
 ; SSE41-NEXT:    movl $255, %eax
 ; SSE41-NEXT:    pinsrb $3, %eax, %xmm0
 ; SSE41-NEXT:    movd %xmm0, (%rdi)
 ; SSE41-NEXT:    retq
   %t0 = fptoui <3 x float> %in to <3 x i8>
   %t1 = shufflevector <3 x i8> %t0, <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>

llvm/test/CodeGen/X86/extract-concat.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=sse4.2  | FileCheck %s --check-prefixes=SSE42
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=avx     | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=avx2    | FileCheck %s --check-prefixes=AVX,AVX2
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=avx512f | FileCheck %s --check-prefixes=AVX,AVX512F
 
 define void @foo(<4 x float> %in, <4 x i8>* %out) {
 ; SSE42-LABEL: foo:
 ; SSE42:       # %bb.0:
 ; SSE42-NEXT:    cvttps2dq %xmm0, %xmm0
 ; SSE42-NEXT:    pextrb $8, %xmm0, %eax
 ; SSE42-NEXT:    pextrb $4, %xmm0, %ecx
-; SSE42-NEXT:    pextrb $0, %xmm0, %edx
-; SSE42-NEXT:    movd %edx, %xmm0
 ; SSE42-NEXT:    pinsrb $1, %ecx, %xmm0
 ; SSE42-NEXT:    pinsrb $2, %eax, %xmm0
 ; SSE42-NEXT:    movl $255, %eax
 ; SSE42-NEXT:    pinsrb $3, %eax, %xmm0
 ; SSE42-NEXT:    movd %xmm0, (%rdi)
 ; SSE42-NEXT:    retq
 ;
 ; AVX-LABEL: foo:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $8, %xmm0, %eax
 ; AVX-NEXT:    vpextrb $4, %xmm0, %ecx
-; AVX-NEXT:    vpextrb $0, %xmm0, %edx
-; AVX-NEXT:    vmovd %edx, %xmm0
 ; AVX-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
 ; AVX-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
 ; AVX-NEXT:    movl $255, %eax
 ; AVX-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, (%rdi)
 ; AVX-NEXT:    retq
   %t0 = fptosi <4 x float> %in to <4 x i32>
   %t1 = trunc <4 x i32> %t0 to <4 x i16>

llvm/test/CodeGen/X86/horizontal-reduce-smax.ll

 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI3_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI3_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $127, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2

 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $127, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <16 x i8> %3, %4

 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pxor LCPI7_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    xorb $127, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorb $127, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2

 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    xorb $127, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorb $127, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    xorb $127, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ; X86-SSE42-NEXT:    pmaxsb %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxsb %xmm2, %xmm1
 ; X86-SSE42-NEXT:    pmaxsb %xmm0, %xmm1
 ; X86-SSE42-NEXT:    pxor LCPI11_0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxsb %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    xorb $127, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorb $127, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm4

 ; X64-SSE42-NEXT:    pmaxsb %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxsb %xmm2, %xmm1
 ; X64-SSE42-NEXT:    pmaxsb %xmm0, %xmm1
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxsb %xmm2, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    xorb $127, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorb $127, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    xorb $127, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI14_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI14_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $127, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $127, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI15_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI15_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $127, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $127, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

llvm/test/CodeGen/X86/horizontal-reduce-smin.ll

 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI3_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI3_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $-128, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2

 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $-128, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <16 x i8> %3, %4

 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminsb %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pxor LCPI7_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    xorb $-128, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorb $-128, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2

 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminsb %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    xorb $-128, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorb $-128, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    xorb $-128, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ; X86-SSE42-NEXT:    pminsb %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminsb %xmm2, %xmm1
 ; X86-SSE42-NEXT:    pminsb %xmm0, %xmm1
 ; X86-SSE42-NEXT:    pxor LCPI11_0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpminsb %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    xorb $-128, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorb $-128, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm2, %xmm4

 ; X64-SSE42-NEXT:    pminsb %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminsb %xmm2, %xmm1
 ; X64-SSE42-NEXT:    pminsb %xmm0, %xmm1
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpminsb %xmm2, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    xorb $-128, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorb $-128, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    xorb $-128, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI14_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI14_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $-128, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $-128, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pxor LCPI15_0, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpxor LCPI15_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    xorb $-128, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    xorb $-128, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

llvm/test/CodeGen/X86/horizontal-reduce-umax.ll

 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    notb %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1

 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v16i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v16i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v16i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <16 x i8> %3, %4

 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    notb %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    notb %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0

 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ; X86-SSE42-NEXT:    pmaxub %xmm2, %xmm1
 ; X86-SSE42-NEXT:    pmaxub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
 ; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxub %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    notb %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    notb %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxub %xmm3, %xmm1

 ; X64-SSE42-NEXT:    pmaxub %xmm2, %xmm1
 ; X64-SSE42-NEXT:    pmaxub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
 ; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxub %xmm2, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    notb %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

 ; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    notb %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
 ; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>

llvm/test/CodeGen/X86/horizontal-reduce-umin.ll

 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]

 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ult <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ult <16 x i8> %3, %4
   %6  = select <16 x i1> %5, <16 x i8> %3, <16 x i8> %4

 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ult <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ult <32 x i8> %3, %4

 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminub %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm2, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE42-NEXT:    psrlw $8, %xmm0
 ; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpminub %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpminub %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pminub %xmm3, %xmm1
 ; X64-SSE2-NEXT:    pminub %xmm2, %xmm1

 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminub %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm2, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE42-NEXT:    psrlw $8, %xmm0
 ; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpminub %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpminub %xmm2, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ult <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ult <64 x i8> %3, %4

 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pminub %xmm0, %xmm1

 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ult <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ult <32 x i8> %3, %4

 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE42-NEXT:    psrlw $8, %xmm1
 ; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    movd %xmm0, %eax
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    vzeroupper
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pminub %xmm0, %xmm1

 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE42-NEXT:    psrlw $8, %xmm1
 ; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
 ; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
-; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    movd %xmm0, %eax
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    vzeroupper
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ult <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ult <64 x i8> %3, %4

llvm/test/CodeGen/X86/scalar_widen_div.ll

 define <4 x i8> @test_char_rem(<4 x i8> %num, <4 x i8> %rem) {
 ; CHECK-LABEL: test_char_rem:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    pextrb $1, %xmm1, %ecx
 ; CHECK-NEXT:    pextrb $1, %xmm0, %eax
 ; CHECK-NEXT:    cbtw
 ; CHECK-NEXT:    idivb %cl
 ; CHECK-NEXT:    movsbl %ah, %ecx
-; CHECK-NEXT:    pextrb $0, %xmm1, %edx
-; CHECK-NEXT:    pextrb $0, %xmm0, %eax
+; CHECK-NEXT:    movd %xmm1, %edx
+; CHECK-NEXT:    movd %xmm0, %eax
 ; CHECK-NEXT:    cbtw
 ; CHECK-NEXT:    idivb %dl
 ; CHECK-NEXT:    movsbl %ah, %eax
 ; CHECK-NEXT:    movd %eax, %xmm2
 ; CHECK-NEXT:    pinsrb $1, %ecx, %xmm2
 ; CHECK-NEXT:    pextrb $2, %xmm1, %ecx
 ; CHECK-NEXT:    pextrb $2, %xmm0, %eax
 ; CHECK-NEXT:    cbtw

llvm/test/CodeGen/X86/var-permute-128.ll

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: var_shuffle_v16i8_from_v32i8_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pushq %rbp
 ; SSE41-NEXT:    movq %rsp, %rbp
 ; SSE41-NEXT:    andq $-32, %rsp
 ; SSE41-NEXT:    subq $544, %rsp # imm = 0x220
-; SSE41-NEXT:    pextrb $0, %xmm2, %eax
+; SSE41-NEXT:    movd %xmm2, %eax
 ; SSE41-NEXT:    movaps %xmm1, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    movaps %xmm0, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    andl $31, %eax
 ; SSE41-NEXT:    movaps %xmm1, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    movaps %xmm0, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    movaps %xmm1, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    movaps %xmm0, {{[0-9]+}}(%rsp)
 ; SSE41-NEXT:    movaps %xmm1, {{[0-9]+}}(%rsp)

llvm/test/CodeGen/X86/var-permute-512.ll

 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    pushq %rbp
 ; AVX512F-NEXT:    movq %rsp, %rbp
 ; AVX512F-NEXT:    andq $-64, %rsp
 ; AVX512F-NEXT:    subq $4160, %rsp # imm = 0x1040
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
 ; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm3
 ; AVX512F-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; AVX512F-NEXT:    vpextrb $0, %xmm4, %eax
+; AVX512F-NEXT:    vmovd %xmm4, %eax
 ; AVX512F-NEXT:    vextractf64x4 $1, %zmm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
 ; AVX512F-NEXT:    vextractf64x4 $1, %zmm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vextractf64x4 $1, %zmm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vextractf64x4 $1, %zmm0, {{[0-9]+}}(%rsp)
 ; AVX512F-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%rsp)

 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $13, 3904(%rsp,%rax), %xmm0, %xmm0
 ; AVX512F-NEXT:    vpextrb $14, %xmm4, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $14, 3968(%rsp,%rax), %xmm0, %xmm0
 ; AVX512F-NEXT:    vpextrb $15, %xmm4, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $15, 4032(%rsp,%rax), %xmm0, %xmm0
-; AVX512F-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512F-NEXT:    vmovd %xmm2, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    movzbl 2048(%rsp,%rax), %eax
 ; AVX512F-NEXT:    vmovd %eax, %xmm4
 ; AVX512F-NEXT:    vpextrb $1, %xmm2, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $1, 2112(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $2, %xmm2, %eax
 ; AVX512F-NEXT:    andl $63, %eax

 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $13, 2880(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $14, %xmm2, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $14, 2944(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $15, %xmm2, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $15, 3008(%rsp,%rax), %xmm4, %xmm2
-; AVX512F-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX512F-NEXT:    vmovd %xmm3, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    movzbl 1024(%rsp,%rax), %eax
 ; AVX512F-NEXT:    vmovd %eax, %xmm4
 ; AVX512F-NEXT:    vpextrb $1, %xmm3, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $1, 1088(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $2, %xmm3, %eax
 ; AVX512F-NEXT:    andl $63, %eax

 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $13, 1856(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $14, %xmm3, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $14, 1920(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $15, %xmm3, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $15, 1984(%rsp,%rax), %xmm4, %xmm3
-; AVX512F-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512F-NEXT:    vmovd %xmm1, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512F-NEXT:    vmovd %eax, %xmm4
 ; AVX512F-NEXT:    vpextrb $1, %xmm1, %eax
 ; AVX512F-NEXT:    andl $63, %eax
 ; AVX512F-NEXT:    vpinsrb $1, 64(%rsp,%rax), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpextrb $2, %xmm1, %eax
 ; AVX512F-NEXT:    andl $63, %eax

 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    pushq %rbp
 ; AVX512BW-NEXT:    movq %rsp, %rbp
 ; AVX512BW-NEXT:    andq $-64, %rsp
 ; AVX512BW-NEXT:    subq $128, %rsp
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm3
 ; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512BW-NEXT:    vpextrb $0, %xmm4, %eax
+; AVX512BW-NEXT:    vmovd %xmm4, %eax
 ; AVX512BW-NEXT:    vmovaps %zmm0, (%rsp)
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vmovd %eax, %xmm0
 ; AVX512BW-NEXT:    vpextrb $1, %xmm4, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $1, (%rsp,%rax), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpextrb $2, %xmm4, %eax

 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $13, (%rsp,%rax), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpextrb $14, %xmm4, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $14, (%rsp,%rax), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpextrb $15, %xmm4, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $15, (%rsp,%rax), %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX512BW-NEXT:    vmovd %xmm3, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vmovd %eax, %xmm4
 ; AVX512BW-NEXT:    vpextrb $1, %xmm3, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $1, (%rsp,%rax), %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpextrb $2, %xmm3, %eax
 ; AVX512BW-NEXT:    andl $63, %eax

 ; AVX512BW-NEXT:    vpextrb $14, %xmm3, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpextrb $15, %xmm3, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm4, %xmm3
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512BW-NEXT:    vmovd %xmm2, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vmovd %eax, %xmm4
 ; AVX512BW-NEXT:    vpextrb $1, %xmm2, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $1, (%rsp,%rax), %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpextrb $2, %xmm2, %eax
 ; AVX512BW-NEXT:    andl $63, %eax

 ; AVX512BW-NEXT:    vpextrb $14, %xmm2, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpextrb $15, %xmm2, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm4, %xmm2
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512BW-NEXT:    vmovd %xmm1, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    movzbl (%rsp,%rax), %eax
 ; AVX512BW-NEXT:    vmovd %eax, %xmm4
 ; AVX512BW-NEXT:    vpextrb $1, %xmm1, %eax
 ; AVX512BW-NEXT:    andl $63, %eax
 ; AVX512BW-NEXT:    vpinsrb $1, (%rsp,%rax), %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpextrb $2, %xmm1, %eax
 ; AVX512BW-NEXT:    andl $63, %eax

llvm/test/CodeGen/X86/vector-bitreverse.ll

 ; AVX-NEXT:    addl %edi, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; XOP-LABEL: test_bitreverse_i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vmovd %edi, %xmm0
 ; XOP-NEXT:    vpperm {{.*}}(%rip), %xmm0, %xmm0, %xmm0
-; XOP-NEXT:    vpextrb $0, %xmm0, %eax
+; XOP-NEXT:    vmovd %xmm0, %eax
 ; XOP-NEXT:    # kill: def $al killed $al killed $eax
 ; XOP-NEXT:    retq
   %b = call i8 @llvm.bitreverse.i8(i8 %a)
   ret i8 %b
 }
 
 define i16 @test_bitreverse_i16(i16 %a) nounwind {
 ; SSE-LABEL: test_bitreverse_i16:

llvm/test/CodeGen/X86/vector-idiv-sdiv-128.ll

 ;
 ; SSE41-LABEL: test_rem_variable_16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pextrb $1, %xmm1, %ecx
 ; SSE41-NEXT:    pextrb $1, %xmm0, %eax
 ; SSE41-NEXT:    cbtw
 ; SSE41-NEXT:    idivb %cl
 ; SSE41-NEXT:    movsbl %ah, %ecx
-; SSE41-NEXT:    pextrb $0, %xmm1, %edx
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm1, %edx
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    cbtw
 ; SSE41-NEXT:    idivb %dl
 ; SSE41-NEXT:    movsbl %ah, %eax
 ; SSE41-NEXT:    movd %eax, %xmm2
 ; SSE41-NEXT:    pinsrb $1, %ecx, %xmm2
 ; SSE41-NEXT:    pextrb $2, %xmm1, %ecx
 ; SSE41-NEXT:    pextrb $2, %xmm0, %eax
 ; SSE41-NEXT:    cbtw

 ;
 ; AVX-LABEL: test_rem_variable_16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpextrb $1, %xmm1, %ecx
 ; AVX-NEXT:    vpextrb $1, %xmm0, %eax
 ; AVX-NEXT:    cbtw
 ; AVX-NEXT:    idivb %cl
 ; AVX-NEXT:    movsbl %ah, %ecx
-; AVX-NEXT:    vpextrb $0, %xmm1, %edx
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm1, %edx
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    cbtw
 ; AVX-NEXT:    idivb %dl
 ; AVX-NEXT:    movsbl %ah, %eax
 ; AVX-NEXT:    vmovd %eax, %xmm2
 ; AVX-NEXT:    vpinsrb $1, %ecx, %xmm2, %xmm2
 ; AVX-NEXT:    vpextrb $2, %xmm1, %ecx
 ; AVX-NEXT:    vpextrb $2, %xmm0, %eax
 ; AVX-NEXT:    cbtw

llvm/test/CodeGen/X86/vector-reduce-add.ll

; AVX512-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v2i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    paddb %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v2i8_load(<2 x i8>* %p) {
-; SSE2-LABEL: test_v2i8_load:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movzwl (%rdi), %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v2i8_load:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movzwl (%rdi), %eax
-; SSE41-NEXT:    movd %eax, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v2i8_load:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movzwl (%rdi), %eax
+; SSE-NEXT:    movd %eax, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    paddb %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8_load:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    movzwl (%rdi), %eax
 ; AVX-NEXT:    vmovd %eax, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8_load:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    movzwl (%rdi), %eax
 ; AVX512-NEXT:    vmovd %eax, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %a0 = load <2 x i8>, <2 x i8>* %p
   %1 = call i8 @llvm.experimental.vector.reduce.add.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSE2-NEXT:    psadbw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; SSE41-NEXT:    psadbw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; AVX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8_load(<4 x i8>* %p) {
-; SSE2-LABEL: test_v4i8_load:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8_load:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8_load:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    psadbw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8_load:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8_load:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %a0 = load <4 x i8>, <4 x i8>* %p
   %1 = call i8 @llvm.experimental.vector.reduce.add.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v8i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    psadbw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8_load(<8 x i8>* %p) {
-; SSE2-LABEL: test_v8i8_load:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v8i8_load:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v8i8_load:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    psadbw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8_load:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8_load:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %a0 = load <8 x i8>, <8 x i8>* %p
   %1 = call i8 @llvm.experimental.vector.reduce.add.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
-; SSE2-LABEL: test_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    psadbw %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    paddb %xmm0, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    psadbw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    paddb %xmm0, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    psadbw %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
-; SSE2-LABEL: test_v32i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    paddb %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    psadbw %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v32i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    paddb %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    paddb %xmm0, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    psadbw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v32i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    paddb %xmm0, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    psadbw %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {
-; SSE2-LABEL: test_v64i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    paddb %xmm3, %xmm1
-; SSE2-NEXT:    paddb %xmm2, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    paddb %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v64i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    paddb %xmm2, %xmm1
-; SSE41-NEXT:    paddb %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    paddb %xmm1, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v64i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    paddb %xmm3, %xmm1
+; SSE-NEXT:    paddb %xmm2, %xmm1
+; SSE-NEXT:    paddb %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    psadbw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {
-; SSE2-LABEL: test_v128i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    paddb %xmm7, %xmm3
-; SSE2-NEXT:    paddb %xmm5, %xmm3
-; SSE2-NEXT:    paddb %xmm1, %xmm3
-; SSE2-NEXT:    paddb %xmm6, %xmm2
-; SSE2-NEXT:    paddb %xmm4, %xmm2
-; SSE2-NEXT:    paddb %xmm3, %xmm2
-; SSE2-NEXT:    paddb %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v128i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    paddb %xmm7, %xmm3
-; SSE41-NEXT:    paddb %xmm5, %xmm3
-; SSE41-NEXT:    paddb %xmm1, %xmm3
-; SSE41-NEXT:    paddb %xmm6, %xmm2
-; SSE41-NEXT:    paddb %xmm4, %xmm2
-; SSE41-NEXT:    paddb %xmm3, %xmm2
-; SSE41-NEXT:    paddb %xmm0, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE41-NEXT:    paddb %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v128i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    paddb %xmm7, %xmm3
+; SSE-NEXT:    paddb %xmm5, %xmm3
+; SSE-NEXT:    paddb %xmm1, %xmm3
+; SSE-NEXT:    paddb %xmm6, %xmm2
+; SSE-NEXT:    paddb %xmm4, %xmm2
+; SSE-NEXT:    paddb %xmm3, %xmm2
+; SSE-NEXT:    paddb %xmm0, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
+; SSE-NEXT:    paddb %xmm2, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    psadbw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpaddb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpaddb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpaddb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.add.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.add.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/vector-reduce-and.ll

; AVX512-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; 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: test_v2i8:
-; SSE41:       # %bb.0:
-; 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: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
-; SSE2-LABEL: test_v4i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; 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: test_v8i8:
-; SSE41:       # %bb.0:
-; 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: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
-; SSE2-LABEL: test_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
-; SSE2-LABEL: test_v32i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v32i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v32i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpand %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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {
-; SSE2-LABEL: test_v64i8:
-; 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: test_v64i8:
-; 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: test_v64i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pand %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm2, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpandq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {
-; SSE2-LABEL: test_v128i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand %xmm6, %xmm2
-; SSE2-NEXT:    pand %xmm7, %xmm3
-; SSE2-NEXT:    pand %xmm5, %xmm3
-; SSE2-NEXT:    pand %xmm1, %xmm3
-; SSE2-NEXT:    pand %xmm4, %xmm2
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE2-NEXT:    pand %xmm2, %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: test_v128i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pand %xmm6, %xmm2
-; SSE41-NEXT:    pand %xmm7, %xmm3
-; SSE41-NEXT:    pand %xmm5, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
-; SSE41-NEXT:    pand %xmm4, %xmm2
-; SSE41-NEXT:    pand %xmm3, %xmm2
-; SSE41-NEXT:    pand %xmm0, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pand %xmm2, %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: test_v128i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pand %xmm6, %xmm2
+; SSE-NEXT:    pand %xmm7, %xmm3
+; SSE-NEXT:    pand %xmm5, %xmm3
+; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pand %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vandps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vandps %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpand %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpandq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpandq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.and.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.and.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/vector-reduce-mul.ll

; AVX512DQVL-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pmullw %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v2i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pmullw %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:

 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v4i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v4i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; AVX512VL-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v4i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:

 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v8i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v8i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX512VL-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpalignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3]
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:

 ; SSE41-NEXT:    packuswb %xmm3, %xmm0
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; SSE41-NEXT:    pmullw %xmm2, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    packuswb %xmm3, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[2],zero,xmm0[4],zero,xmm0[6],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: test_v16i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BWVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BWVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BWVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v16i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQ-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQ-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQ-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: test_v16i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQVL-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQVL-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512DQVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQVL-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {

 ; SSE41-NEXT:    packuswb %xmm3, %xmm2
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; SSE41-NEXT:    pmullw %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    packuswb %xmm3, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]

 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero

 ; AVX512BW-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpmullw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpmullw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: test_v32i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero

 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BWVL-NEXT:    vpmullw %zmm1, %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BWVL-NEXT:    vpmullw %zmm1, %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BWVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v32i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]

 ; AVX512DQ-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: test_v32i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]

 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE41-NEXT:    packuswb %xmm3, %xmm1
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
 ; SSE41-NEXT:    pand %xmm2, %xmm1
 ; SSE41-NEXT:    packuswb %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm2[8],xmm0[8],xmm2[9],xmm0[9],xmm2[10],xmm0[10],xmm2[11],xmm0[11],xmm2[12],xmm0[12],xmm2[13],xmm0[13],xmm2[14],xmm0[14],xmm2[15],xmm0[15]
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4

 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23]
 ; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm3 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]

 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]

 ; AVX512BW-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: test_v64i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]

 ; AVX512BWVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v64i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23]

 ; AVX512DQ-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; AVX512DQ-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: test_v64i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31]

 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41-NEXT:    packuswb %xmm2, %xmm0
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
 ; SSE41-NEXT:    pand %xmm9, %xmm0
 ; SSE41-NEXT:    packuswb %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
 ; AVX1-NEXT:    vpmullw %xmm4, %xmm5, %xmm4

 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm3[8],ymm0[8],ymm3[9],ymm0[9],ymm3[10],ymm0[10],ymm3[11],ymm0[11],ymm3[12],ymm0[12],ymm3[13],ymm0[13],ymm3[14],ymm0[14],ymm3[15],ymm0[15],ymm3[24],ymm0[24],ymm3[25],ymm0[25],ymm3[26],ymm0[26],ymm3[27],ymm0[27],ymm3[28],ymm0[28],ymm3[29],ymm0[29],ymm3[30],ymm0[30],ymm3[31],ymm0[31]
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31]

 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v128i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]

 ; AVX512BW-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: test_v128i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]

 ; AVX512BWVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_v128i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm3[8],ymm0[8],ymm3[9],ymm0[9],ymm3[10],ymm0[10],ymm3[11],ymm0[11],ymm3[12],ymm0[12],ymm3[13],ymm0[13],ymm3[14],ymm0[14],ymm3[15],ymm0[15],ymm3[24],ymm0[24],ymm3[25],ymm0[25],ymm3[26],ymm0[26],ymm3[27],ymm0[27],ymm3[28],ymm0[28],ymm3[29],ymm0[29],ymm3[30],ymm0[30],ymm3[31],ymm0[31]
 ; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31]

 ; AVX512DQ-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: test_v128i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[4],ymm0[4],ymm3[5],ymm0[5],ymm3[6],ymm0[6],ymm3[7],ymm0[7],ymm3[16],ymm0[16],ymm3[17],ymm0[17],ymm3[18],ymm0[18],ymm3[19],ymm0[19],ymm3[20],ymm0[20],ymm3[21],ymm0[21],ymm3[22],ymm0[22],ymm3[23],ymm0[23]
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23]

 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.mul.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/vector-reduce-or.ll

; AVX512-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; 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: test_v2i8:
-; SSE41:       # %bb.0:
-; 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: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
-; SSE2-LABEL: test_v4i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; 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: test_v8i8:
-; SSE41:       # %bb.0:
-; 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: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
-; SSE2-LABEL: test_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
-; SSE2-LABEL: test_v32i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v32i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    por %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v32i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpor %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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {
-; SSE2-LABEL: test_v64i8:
-; 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: test_v64i8:
-; 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: test_v64i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    por %xmm3, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {
-; SSE2-LABEL: test_v128i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    por %xmm6, %xmm2
-; SSE2-NEXT:    por %xmm7, %xmm3
-; SSE2-NEXT:    por %xmm5, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
-; SSE2-NEXT:    por %xmm4, %xmm2
-; SSE2-NEXT:    por %xmm3, %xmm2
-; SSE2-NEXT:    por %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE2-NEXT:    por %xmm2, %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: test_v128i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    por %xmm6, %xmm2
-; SSE41-NEXT:    por %xmm7, %xmm3
-; SSE41-NEXT:    por %xmm5, %xmm3
-; SSE41-NEXT:    por %xmm1, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm2
-; SSE41-NEXT:    por %xmm3, %xmm2
-; SSE41-NEXT:    por %xmm0, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE41-NEXT:    por %xmm2, %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: test_v128i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    por %xmm6, %xmm2
+; SSE-NEXT:    por %xmm7, %xmm3
+; SSE-NEXT:    por %xmm5, %xmm3
+; SSE-NEXT:    por %xmm1, %xmm3
+; SSE-NEXT:    por %xmm4, %xmm2
+; SSE-NEXT:    por %xmm3, %xmm2
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
+; SSE-NEXT:    por %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vorps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpor %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.or.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.or.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/vector-reduce-smax.ll

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:

 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:

 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:

 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    xorb $127, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {

 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm2, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pmaxsb %xmm4, %xmm2
 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm2
 ; SSE41-NEXT:    pmaxsb %xmm0, %xmm2
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm2, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmaxsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 

llvm/test/CodeGen/X86/vector-reduce-smin.ll

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminsb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:

 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    pminsb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:

 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE41-NEXT:    pminsb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminsb %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movd %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:

 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    xorb $-128, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {

 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pminsb %xmm3, %xmm1
 ; SSE41-NEXT:    pminsb %xmm2, %xmm1
 ; SSE41-NEXT:    pminsb %xmm0, %xmm1
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pminsb %xmm4, %xmm2
 ; SSE41-NEXT:    pminsb %xmm3, %xmm2
 ; SSE41-NEXT:    pminsb %xmm0, %xmm2
 ; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm2, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 

llvm/test/CodeGen/X86/vector-reduce-umax.ll

; AVX512VL-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pmaxub %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v2i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pmaxub %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pmaxub %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
-; SSE2-LABEL: test_v4i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pmaxub %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pmaxub %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pmaxub %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pmaxub %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pmaxub %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pmaxub %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pmaxub %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pmaxub %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pmaxub %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v8i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmaxub %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pmaxub %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pmaxub %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pmaxub %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pmaxub %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pmaxub %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:

 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
 ; SSE41-NEXT:    pxor %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    notb %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
 ; SSE41-NEXT:    pxor %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pmaxub %xmm2, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm0, %xmm1
 ; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
 ; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 

 ; SSE41-NEXT:    pmaxub %xmm3, %xmm2
 ; SSE41-NEXT:    pmaxub %xmm0, %xmm2
 ; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
 ; SSE41-NEXT:    pxor %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxub %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v128i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v128i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 

llvm/test/CodeGen/X86/vector-reduce-umin.ll

; AVX512-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pminub %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v2i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pminub %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pminub %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
-; SSE2-LABEL: test_v4i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pminub %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pminub %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pminub %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pminub %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pminub %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pminub %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pminub %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pminub %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pminub %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v8i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pminub %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pminub %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pminub %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pminub %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pminub %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pminub %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:

 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    phminposuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminub %xmm3, %xmm1
 ; SSE41-NEXT:    pminub %xmm2, %xmm1
 ; SSE41-NEXT:    pminub %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm1, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminub %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminub %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41-NEXT:    pminub %xmm6, %xmm2
 ; SSE41-NEXT:    pminub %xmm4, %xmm2
 ; SSE41-NEXT:    pminub %xmm3, %xmm2
 ; SSE41-NEXT:    pminub %xmm0, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pminub %xmm2, %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vpminub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpminub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminub %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpminub %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminub %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminub %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminub %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminub %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.umin.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/vector-reduce-xor.ll

; AVX512-NEXT:    retq
   ret i16 %1
 }
 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
-; SSE2-LABEL: test_v2i8:
-; SSE2:       # %bb.0:
-; 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: test_v2i8:
-; SSE41:       # %bb.0:
-; 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: test_v2i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
-; SSE2-LABEL: test_v4i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v4i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v4i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
-; SSE2-LABEL: test_v8i8:
-; SSE2:       # %bb.0:
-; 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: test_v8i8:
-; SSE41:       # %bb.0:
-; 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: test_v8i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
-; SSE2-LABEL: test_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
-; SSE2-LABEL: test_v32i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    # kill: def $al killed $al killed $eax
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: test_v32i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
-; SSE41-NEXT:    # kill: def $al killed $al killed $eax
-; SSE41-NEXT:    retq
+; SSE-LABEL: test_v32i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {
-; SSE2-LABEL: test_v64i8:
-; 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: test_v64i8:
-; 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: test_v64i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm3, %xmm1
+; SSE-NEXT:    pxor %xmm2, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {
-; SSE2-LABEL: test_v128i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm6, %xmm2
-; SSE2-NEXT:    pxor %xmm7, %xmm3
-; SSE2-NEXT:    pxor %xmm5, %xmm3
-; SSE2-NEXT:    pxor %xmm1, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm2
-; SSE2-NEXT:    pxor %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm2, %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: test_v128i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm6, %xmm2
-; SSE41-NEXT:    pxor %xmm7, %xmm3
-; SSE41-NEXT:    pxor %xmm5, %xmm3
-; SSE41-NEXT:    pxor %xmm1, %xmm3
-; SSE41-NEXT:    pxor %xmm4, %xmm2
-; SSE41-NEXT:    pxor %xmm3, %xmm2
-; SSE41-NEXT:    pxor %xmm0, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pxor %xmm2, %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: test_v128i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm6, %xmm2
+; SSE-NEXT:    pxor %xmm7, %xmm3
+; SSE-NEXT:    pxor %xmm5, %xmm3
+; SSE-NEXT:    pxor %xmm1, %xmm3
+; SSE-NEXT:    pxor %xmm4, %xmm2
+; SSE-NEXT:    pxor %xmm3, %xmm2
+; SSE-NEXT:    pxor %xmm0, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
+; SSE-NEXT:    pxor %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pxor %xmm0, %xmm1
+; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vxorps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vxorps %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpxor %ymm1, %ymm2, %ymm1
 ; 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:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.xor.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.xor.v2i64(<2 x i64>)

llvm/test/CodeGen/X86/widen_bitops-0.ll

 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    pand %xmm0, %xmm1
-; X32-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X32-SSE-NEXT:    movd %xmm1, %eax
 ; X32-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X32-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X32-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X32-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X32-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X32-SSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: and_v3i8_as_i24:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movd %ecx, %xmm0
 ; X64-SSE-NEXT:    pinsrb $1, %r8d, %xmm0
 ; X64-SSE-NEXT:    pinsrb $2, %r9d, %xmm0
 ; X64-SSE-NEXT:    movd %edi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $1, %esi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $2, %edx, %xmm1
 ; X64-SSE-NEXT:    pand %xmm0, %xmm1
-; X64-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X64-SSE-NEXT:    movd %xmm1, %eax
 ; X64-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X64-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X64-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X64-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-SSE-NEXT:    retq
   %1 = bitcast <3 x i8> %a to i24
   %2 = bitcast <3 x i8> %b to i24
   %3 = and i24 %1, %2
   %4 = bitcast i24 %3 to <3 x i8>
   ret <3 x i8>  %4
 }
 
 define <3 x i8> @xor_v3i8_as_i24(<3 x i8> %a, <3 x i8> %b) nounwind {
 ; X32-SSE-LABEL: xor_v3i8_as_i24:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    pxor %xmm0, %xmm1
-; X32-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X32-SSE-NEXT:    movd %xmm1, %eax
 ; X32-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X32-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X32-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X32-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X32-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X32-SSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: xor_v3i8_as_i24:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movd %ecx, %xmm0
 ; X64-SSE-NEXT:    pinsrb $1, %r8d, %xmm0
 ; X64-SSE-NEXT:    pinsrb $2, %r9d, %xmm0
 ; X64-SSE-NEXT:    movd %edi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $1, %esi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $2, %edx, %xmm1
 ; X64-SSE-NEXT:    pxor %xmm0, %xmm1
-; X64-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X64-SSE-NEXT:    movd %xmm1, %eax
 ; X64-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X64-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X64-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X64-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-SSE-NEXT:    retq
   %1 = bitcast <3 x i8> %a to i24
   %2 = bitcast <3 x i8> %b to i24
   %3 = xor i24 %1, %2
   %4 = bitcast i24 %3 to <3 x i8>
   ret <3 x i8>  %4
 }
 
 define <3 x i8> @or_v3i8_as_i24(<3 x i8> %a, <3 x i8> %b) nounwind {
 ; X32-SSE-LABEL: or_v3i8_as_i24:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm0
 ; X32-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; X32-SSE-NEXT:    pinsrb $1, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    pinsrb $2, {{[0-9]+}}(%esp), %xmm1
 ; X32-SSE-NEXT:    por %xmm0, %xmm1
-; X32-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X32-SSE-NEXT:    movd %xmm1, %eax
 ; X32-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X32-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X32-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X32-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X32-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X32-SSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: or_v3i8_as_i24:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movd %ecx, %xmm0
 ; X64-SSE-NEXT:    pinsrb $1, %r8d, %xmm0
 ; X64-SSE-NEXT:    pinsrb $2, %r9d, %xmm0
 ; X64-SSE-NEXT:    movd %edi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $1, %esi, %xmm1
 ; X64-SSE-NEXT:    pinsrb $2, %edx, %xmm1
 ; X64-SSE-NEXT:    por %xmm0, %xmm1
-; X64-SSE-NEXT:    pextrb $0, %xmm1, %eax
+; X64-SSE-NEXT:    movd %xmm1, %eax
 ; X64-SSE-NEXT:    pextrb $1, %xmm1, %edx
 ; X64-SSE-NEXT:    pextrb $2, %xmm1, %ecx
 ; X64-SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-SSE-NEXT:    # kill: def $dl killed $dl killed $edx
 ; X64-SSE-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-SSE-NEXT:    retq
   %1 = bitcast <3 x i8> %a to i24
   %2 = bitcast <3 x i8> %b to i24