[X86] Rerun "update_llc_test_checks" tool on CodeGen tests. NFC.

[gadi.haber]

Summary

This is NFC after rerunning the "update_llc_test_checks.py" tool on the CodeGen X86 tests in order to submit a patch.
Minor differences due to added "End of Function" lines.

[zvi]

LGTM

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

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl  | FileCheck %s --check-prefix=ALL --check-prefix=KNL
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx  | FileCheck %s --check-prefix=ALL --check-prefix=SKX
 
 define double @test1(double %a, double %b) nounwind {
 ; ALL-LABEL: test1:
 ; ALL:       ## BB#0:
 ; ALL-NEXT:    vucomisd %xmm1, %xmm0
 ; ALL-NEXT:    jne LBB0_1
 ; ALL-NEXT:    jnp LBB0_2
 ; ALL-NEXT:  LBB0_1: ## %l1
 ; ALL-NEXT:    vsubsd %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    retq
 ; ALL-NEXT:  LBB0_2: ## %l2
 ; ALL-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    retq
+; ALL-NEXT:    ## -- End function
   %tobool = fcmp une double %a, %b
   br i1 %tobool, label %l1, label %l2
 
 l1:
   %c = fsub double %a, %b
   ret double %c
 l2:
   %c1 = fadd double %a, %b
   ret double %c1
 }
 
 define float @test2(float %a, float %b) nounwind {
 ; ALL-LABEL: test2:
 ; ALL:       ## BB#0:
 ; ALL-NEXT:    vucomiss %xmm0, %xmm1
 ; ALL-NEXT:    jbe LBB1_2
 ; ALL-NEXT:  ## BB#1: ## %l1
 ; ALL-NEXT:    vsubss %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    retq
 ; ALL-NEXT:  LBB1_2: ## %l2
 ; ALL-NEXT:    vaddss %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    retq
+; ALL-NEXT:    ## -- End function
   %tobool = fcmp olt float %a, %b
   br i1 %tobool, label %l1, label %l2
 
 l1:
   %c = fsub float %a, %b
   ret float %c
 l2:
   %c1 = fadd float %a, %b

llvm/trunk/test/CodeGen/X86/avx512-insert-extract.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck --check-prefix=KNL %s
 ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck --check-prefix=SKX --check-prefix=SKX_ONLY %s
 ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=skx -mattr=avx512vbmi | FileCheck --check-prefix=SKX --check-prefix=SKX_VBMI %s
 
 define <16 x float> @test1(<16 x float> %x, float* %br, float %y) nounwind {
 ; KNL-LABEL: test1:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vinsertps {{.*#+}} xmm2 = xmm0[0],mem[0],xmm0[2,3]
 ; KNL-NEXT:    vinsertf32x4 $0, %xmm2, %zmm0, %zmm2
 ; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
 ; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0],xmm0[3]
 ; KNL-NEXT:    vinsertf32x4 $3, %xmm0, %zmm2, %zmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test1:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vinsertps {{.*#+}} xmm2 = xmm0[0],mem[0],xmm0[2,3]
 ; SKX-NEXT:    vinsertf32x4 $0, %xmm2, %zmm0, %zmm2
 ; SKX-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
 ; SKX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0],xmm0[3]
 ; SKX-NEXT:    vinsertf32x4 $3, %xmm0, %zmm2, %zmm0
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %rrr = load float, float* %br
   %rrr2 = insertelement <16 x float> %x, float %rrr, i32 1
   %rrr3 = insertelement <16 x float> %rrr2, float %y, i32 14
   ret <16 x float> %rrr3
 }
 
 define <8 x double> @test2(<8 x double> %x, double* %br, double %y) nounwind {
 ; KNL-LABEL: test2:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vmovhpd {{.*#+}} xmm2 = xmm0[0],mem[0]
 ; KNL-NEXT:    vinsertf32x4 $0, %xmm2, %zmm0, %zmm2
 ; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
 ; KNL-NEXT:    vmovsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; KNL-NEXT:    vinsertf32x4 $3, %xmm0, %zmm2, %zmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test2:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vmovhpd {{.*#+}} xmm2 = xmm0[0],mem[0]
 ; SKX-NEXT:    vinsertf64x2 $0, %xmm2, %zmm0, %zmm2
 ; SKX-NEXT:    vextractf64x2 $3, %zmm0, %xmm0
 ; SKX-NEXT:    vmovsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SKX-NEXT:    vinsertf64x2 $3, %xmm0, %zmm2, %zmm0
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %rrr = load double, double* %br
   %rrr2 = insertelement <8 x double> %x, double %rrr, i32 1
   %rrr3 = insertelement <8 x double> %rrr2, double %y, i32 6
   ret <8 x double> %rrr3
 }
 
 define <16 x float> @test3(<16 x float> %x) nounwind {
 ; KNL-LABEL: test3:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vextractf32x4 $1, %zmm0, %xmm1
 ; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[2,3]
 ; KNL-NEXT:    vinsertf32x4 $0, %xmm1, %zmm0, %zmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test3:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vextractf32x4 $1, %zmm0, %xmm1
 ; SKX-NEXT:    vinsertps {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[2,3]
 ; SKX-NEXT:    vinsertf32x4 $0, %xmm1, %zmm0, %zmm0
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %eee = extractelement <16 x float> %x, i32 4
   %rrr2 = insertelement <16 x float> %x, float %eee, i32 1
   ret <16 x float> %rrr2
 }
 
 define <8 x i64> @test4(<8 x i64> %x) nounwind {
 ; KNL-LABEL: test4:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
 ; KNL-NEXT:    vmovq %xmm1, %rax
 ; KNL-NEXT:    vpinsrq $1, %rax, %xmm0, %xmm1
 ; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm0, %zmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test4:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vextracti64x2 $2, %zmm0, %xmm1
 ; SKX-NEXT:    vmovq %xmm1, %rax
 ; SKX-NEXT:    vpinsrq $1, %rax, %xmm0, %xmm1
 ; SKX-NEXT:    vinserti64x2 $0, %xmm1, %zmm0, %zmm0
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %eee = extractelement <8 x i64> %x, i32 4
   %rrr2 = insertelement <8 x i64> %x, i64 %eee, i32 1
   ret <8 x i64> %rrr2
 }
 
 define i32 @test5(<4 x float> %x) nounwind {
 ; KNL-LABEL: test5:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vextractps $3, %xmm0, %eax
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test5:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vextractps $3, %xmm0, %eax
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %ef = extractelement <4 x float> %x, i32 3
   %ei = bitcast float %ef to i32
   ret i32 %ei
 }
 
 define void @test6(<4 x float> %x, float* %out) nounwind {
 ; KNL-LABEL: test6:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vextractps $3, %xmm0, (%rdi)
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test6:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vextractps $3, %xmm0, (%rdi)
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %ef = extractelement <4 x float> %x, i32 3
   store float %ef, float* %out, align 4
   ret void
 }
 
 define float @test7(<16 x float> %x, i32 %ind) nounwind {
 ; KNL-LABEL: test7:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    andq $-64, %rsp
 ; KNL-NEXT:    subq $128, %rsp
 ; KNL-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; KNL-NEXT:    vmovaps %zmm0, (%rsp)
 ; KNL-NEXT:    andl $15, %edi
 ; KNL-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test7:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    pushq %rbp
 ; SKX-NEXT:    movq %rsp, %rbp
 ; SKX-NEXT:    andq $-64, %rsp
 ; SKX-NEXT:    subq $128, %rsp
 ; SKX-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; SKX-NEXT:    vmovaps %zmm0, (%rsp)
 ; SKX-NEXT:    andl $15, %edi
 ; SKX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SKX-NEXT:    movq %rbp, %rsp
 ; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %e = extractelement <16 x float> %x, i32 %ind
   ret float %e
 }
 
 define double @test8(<8 x double> %x, i32 %ind) nounwind {
 ; KNL-LABEL: test8:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    andq $-64, %rsp
 ; KNL-NEXT:    subq $128, %rsp
 ; KNL-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; KNL-NEXT:    vmovaps %zmm0, (%rsp)
 ; KNL-NEXT:    andl $7, %edi
 ; KNL-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test8:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    pushq %rbp
 ; SKX-NEXT:    movq %rsp, %rbp
 ; SKX-NEXT:    andq $-64, %rsp
 ; SKX-NEXT:    subq $128, %rsp
 ; SKX-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; SKX-NEXT:    vmovaps %zmm0, (%rsp)
 ; SKX-NEXT:    andl $7, %edi
 ; SKX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
 ; SKX-NEXT:    movq %rbp, %rsp
 ; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %e = extractelement <8 x double> %x, i32 %ind
   ret double %e
 }
 
 define float @test9(<8 x float> %x, i32 %ind) nounwind {
 ; KNL-LABEL: test9:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    andq $-32, %rsp
 ; KNL-NEXT:    subq $64, %rsp
 ; KNL-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; KNL-NEXT:    vmovaps %ymm0, (%rsp)
 ; KNL-NEXT:    andl $7, %edi
 ; KNL-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test9:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    pushq %rbp
 ; SKX-NEXT:    movq %rsp, %rbp
 ; SKX-NEXT:    andq $-32, %rsp
 ; SKX-NEXT:    subq $64, %rsp
 ; SKX-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; SKX-NEXT:    vmovaps %ymm0, (%rsp)
 ; SKX-NEXT:    andl $7, %edi
 ; SKX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SKX-NEXT:    movq %rbp, %rsp
 ; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %e = extractelement <8 x float> %x, i32 %ind
   ret float %e
 }
 
 define i32 @test10(<16 x i32> %x, i32 %ind) nounwind {
 ; KNL-LABEL: test10:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    andq $-64, %rsp
 ; KNL-NEXT:    subq $128, %rsp
 ; KNL-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; KNL-NEXT:    vmovaps %zmm0, (%rsp)
 ; KNL-NEXT:    andl $15, %edi
 ; KNL-NEXT:    movl (%rsp,%rdi,4), %eax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test10:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    pushq %rbp
 ; SKX-NEXT:    movq %rsp, %rbp
 ; SKX-NEXT:    andq $-64, %rsp
 ; SKX-NEXT:    subq $128, %rsp
 ; SKX-NEXT:    ## kill: %EDI<def> %EDI<kill> %RDI<def>
 ; SKX-NEXT:    vmovaps %zmm0, (%rsp)
 ; SKX-NEXT:    andl $15, %edi
 ; SKX-NEXT:    movl (%rsp,%rdi,4), %eax
 ; SKX-NEXT:    movq %rbp, %rsp
 ; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %e = extractelement <16 x i32> %x, i32 %ind
   ret i32 %e
 }
 
 define <16 x i32> @test11(<16 x i32>%a, <16 x i32>%b) {
 ; KNL-LABEL: test11:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vpcmpltud %zmm1, %zmm0, %k0

llvm/trunk/test/CodeGen/X86/avx512-vec-cmp.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s --check-prefix=CHECK --check-prefix=KNL
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
 
 define <16 x float> @test1(<16 x float> %x, <16 x float> %y) nounwind {
 ; CHECK-LABEL: test1:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmpleps %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = fcmp ole <16 x float> %x, %y
   %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %y
   ret <16 x float> %max
 }
 
 define <8 x double> @test2(<8 x double> %x, <8 x double> %y) nounwind {
 ; CHECK-LABEL: test2:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmplepd %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vblendmpd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = fcmp ole <8 x double> %x, %y
   %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %y
   ret <8 x double> %max
 }
 
 define <16 x i32> @test3(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %yp) nounwind {
 ; CHECK-LABEL: test3:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpeqd (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <16 x i32>, <16 x i32>* %yp, align 4
   %mask = icmp eq <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test4_unsigned(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind {
 ; CHECK-LABEL: test4_unsigned:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpnltud %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm2, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = icmp uge <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y
   ret <16 x i32> %max
 }
 
 define <8 x i64> @test5(<8 x i64> %x, <8 x i64> %y) nounwind {
 ; CHECK-LABEL: test5:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpeqq %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpblendmq %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = icmp eq <8 x i64> %x, %y
   %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %y
   ret <8 x i64> %max
 }
 
 define <8 x i64> @test6_unsigned(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1) nounwind {
 ; CHECK-LABEL: test6_unsigned:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpnleuq %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpblendmq %zmm2, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = icmp ugt <8 x i64> %x, %y
   %max = select <8 x i1> %mask, <8 x i64> %x1, <8 x i64> %y
   ret <8 x i64> %max
 }
 
 define <4 x float> @test7(<4 x float> %a, <4 x float> %b) {
 ; KNL-LABEL: test7:
 ; KNL:       ## BB#0:

 ; KNL-LABEL: test9:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    ## kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
 ; KNL-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test9:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vpcmpeqd %ymm1, %ymm0, %k1
 ; SKX-NEXT:    vpblendmd %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %mask = icmp eq <8 x i32> %x, %y
   %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y
   ret <8 x i32> %max
 }
 
 define <8 x float> @test10(<8 x float> %x, <8 x float> %y) nounwind {
 ; KNL-LABEL: test10:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    ## kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
 ; KNL-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test10:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
 ; SKX-NEXT:    vblendmps %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %mask = fcmp oeq <8 x float> %x, %y
   %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y
   ret <8 x float> %max
 }
 
 define <8 x i32> @test11_unsigned(<8 x i32> %x, <8 x i32> %y) nounwind {
 ; CHECK-LABEL: test11_unsigned:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = icmp ugt <8 x i32> %x, %y
   %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y
   ret <8 x i32> %max
 }
 
 define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind {
 ; KNL-LABEL: test12:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
 ; KNL-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
 ; KNL-NEXT:    kunpckbw %k0, %k1, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test12:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
 ; SKX-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
 ; SKX-NEXT:    kunpckbw %k0, %k1, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
 ; SKX-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %res = icmp eq <16 x i64> %a, %b
   %res1 = bitcast <16 x i1> %res to i16
   ret i16 %res1
 }
 
 define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
 ; KNL-LABEL: test12_v32i32:
 ; KNL:       ## BB#0:

 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, (%rsp)
 ; KNL-NEXT:    movl (%rsp), %eax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test12_v32i32:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vpcmpeqd %zmm2, %zmm0, %k0
 ; SKX-NEXT:    vpcmpeqd %zmm3, %zmm1, %k1
 ; SKX-NEXT:    kunpckwd %k0, %k1, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %res = icmp eq <32 x i32> %a, %b
   %res1 = bitcast <32 x i1> %res to i32
   ret i32 %res1
 }
 
 define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind {
 ; KNL-LABEL: test12_v64i16:
 ; KNL:       ## BB#0:

 ; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    movl (%rsp), %ecx
 ; KNL-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; KNL-NEXT:    shlq $32, %rax
 ; KNL-NEXT:    orq %rcx, %rax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test12_v64i16:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vpcmpeqw %zmm2, %zmm0, %k0
 ; SKX-NEXT:    vpcmpeqw %zmm3, %zmm1, %k1
 ; SKX-NEXT:    kunpckdq %k0, %k1, %k0
 ; SKX-NEXT:    kmovq %k0, %rax
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %res = icmp eq <64 x i16> %a, %b
   %res1 = bitcast <64 x i1> %res to i64
   ret i64 %res1
 }
 
 define <16 x i32> @test13(<16 x float>%a, <16 x float>%b)
 ; CHECK-LABEL: test13:
 ; CHECK:       ## BB#0:

 }
 
 define <16 x i32> @test16(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind {
 ; CHECK-LABEL: test16:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpled %zmm0, %zmm1, %k1
 ; CHECK-NEXT:    vpblendmd %zmm2, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask = icmp sge <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test17(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind {
 ; CHECK-LABEL: test17:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpgtd (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4
   %mask = icmp sgt <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test18(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind {
 ; CHECK-LABEL: test18:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpled (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4
   %mask = icmp sle <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test19(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind {
 ; CHECK-LABEL: test19:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpleud (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4
   %mask = icmp ule <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test20(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) nounwind {
 ; CHECK-LABEL: test20:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpcmpeqd %zmm3, %zmm2, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp eq <16 x i32> %x1, %y1
   %mask0 = icmp eq <16 x i32> %x, %y
   %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %y
   ret <16 x i32> %max
 }
 
 define <8 x i64> @test21(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) nounwind {
 ; CHECK-LABEL: test21:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpleq %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpcmpleq %zmm2, %zmm3, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmq %zmm0, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp sge <8 x i64> %x1, %y1
   %mask0 = icmp sle <8 x i64> %x, %y
   %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer
   %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1
   ret <8 x i64> %max
 }
 
 define <8 x i64> @test22(<8 x i64> %x, <8 x i64>* %y.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind {
 ; CHECK-LABEL: test22:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpgtq %zmm2, %zmm1, %k1
 ; CHECK-NEXT:    vpcmpgtq (%rdi), %zmm0, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmq %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp sgt <8 x i64> %x1, %y1
   %y = load <8 x i64>, <8 x i64>* %y.ptr, align 4
   %mask0 = icmp sgt <8 x i64> %x, %y
   %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer
   %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1
   ret <8 x i64> %max
 }
 
 define <16 x i32> @test23(<16 x i32> %x, <16 x i32>* %y.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind {
 ; CHECK-LABEL: test23:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpled %zmm1, %zmm2, %k1
 ; CHECK-NEXT:    vpcmpleud (%rdi), %zmm0, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp sge <16 x i32> %x1, %y1
   %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4
   %mask0 = icmp ule <16 x i32> %x, %y
   %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <8 x i64> @test24(<8 x i64> %x, <8 x i64> %x1, i64* %yb.ptr) nounwind {
 ; CHECK-LABEL: test24:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpeqq (%rdi){1to8}, %zmm0, %k1
 ; CHECK-NEXT:    vpblendmq %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %yb = load i64, i64* %yb.ptr, align 4
   %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0
   %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer
   %mask = icmp eq <8 x i64> %x, %y
   %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1
   ret <8 x i64> %max
 }
 
 define <16 x i32> @test25(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1) nounwind {
 ; CHECK-LABEL: test25:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpled (%rdi){1to16}, %zmm0, %k1
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %yb = load i32, i32* %yb.ptr, align 4
   %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0
   %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer
   %mask = icmp sle <16 x i32> %x, %y
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind {
 ; CHECK-LABEL: test26:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpled %zmm1, %zmm2, %k1
 ; CHECK-NEXT:    vpcmpgtd (%rdi){1to16}, %zmm0, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp sge <16 x i32> %x1, %y1
   %yb = load i32, i32* %yb.ptr, align 4
   %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0
   %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer
   %mask0 = icmp sgt <16 x i32> %x, %y
   %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer
   %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1
   ret <16 x i32> %max
 }
 
 define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind {
 ; CHECK-LABEL: test27:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpcmpleq %zmm1, %zmm2, %k1
 ; CHECK-NEXT:    vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1}
 ; CHECK-NEXT:    vpblendmq %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %mask1 = icmp sge <8 x i64> %x1, %y1
   %yb = load i64, i64* %yb.ptr, align 4
   %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0
   %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer
   %mask0 = icmp sle <8 x i64> %x, %y
   %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer
   %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1
   ret <8 x i64> %max

 }
 
 define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind {
 ; KNL-LABEL: test30:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm2
 ; KNL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test30:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
 ; SKX-NEXT:    vblendmpd %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %mask = fcmp oeq <4 x double> %x, %y
   %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %y
   ret <4 x double> %max
 }
 
 define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp) nounwind {
 ; KNL-LABEL: test31:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcmpltpd (%rdi), %xmm0, %xmm2
 ; KNL-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test31:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltpd (%rdi), %xmm0, %k1
 ; SKX-NEXT:    vblendmpd %xmm0, %xmm1, %xmm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %y = load <2 x double>, <2 x double>* %yp, align 4
   %mask = fcmp olt <2 x double> %x, %y
   %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1
   ret <2 x double> %max
 }
 
 define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp) nounwind {
 ; KNL-LABEL: test32:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcmpltpd (%rdi), %ymm0, %ymm2
 ; KNL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test32:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltpd (%rdi), %ymm0, %k1
 ; SKX-NEXT:    vblendmpd %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %y = load <4 x double>, <4 x double>* %yp, align 4
   %mask = fcmp ogt <4 x double> %y, %x
   %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1
   ret <4 x double> %max
 }
 
 define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp) nounwind {
 ; CHECK-LABEL: test33:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmpltpd (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vblendmpd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <8 x double>, <8 x double>* %yp, align 4
   %mask = fcmp olt <8 x double> %x, %y
   %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1
   ret <8 x double> %max
 }
 
 define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) nounwind {
 ; KNL-LABEL: test34:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcmpltps (%rdi), %xmm0, %xmm2
 ; KNL-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test34:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltps (%rdi), %xmm0, %k1
 ; SKX-NEXT:    vblendmps %xmm0, %xmm1, %xmm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
   %y = load <4 x float>, <4 x float>* %yp, align 4
   %mask = fcmp olt <4 x float> %x, %y
   %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1
   ret <4 x float> %max
 }
 
 define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) nounwind {
 ; KNL-LABEL: test35:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    ## kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vmovups (%rdi), %ymm2
 ; KNL-NEXT:    vcmpltps %zmm2, %zmm0, %k1
 ; KNL-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test35:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltps (%rdi), %ymm0, %k1
 ; SKX-NEXT:    vblendmps %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %y = load <8 x float>, <8 x float>* %yp, align 4
   %mask = fcmp ogt <8 x float> %y, %x
   %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1
   ret <8 x float> %max
 }
 
 define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp) nounwind {
 ; CHECK-LABEL: test36:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmpltps (%rdi), %zmm0, %k1
 ; CHECK-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
   %y = load <16 x float>, <16 x float>* %yp, align 4
   %mask = fcmp olt <16 x float> %x, %y
   %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1
   ret <16 x float> %max
 }
 
 define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nounwind {
 ; CHECK-LABEL: test37:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmpltpd (%rdi){1to8}, %zmm0, %k1
 ; CHECK-NEXT:    vblendmpd %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
 
   %a = load double, double* %ptr
   %v = insertelement <8 x double> undef, double %a, i32 0
   %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer
 
   %mask = fcmp ogt <8 x double> %shuffle, %x
   %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1
   ret <8 x double> %max
 }
 
 define <4 x double> @test38(<4 x double> %x, <4 x double> %x1, double* %ptr) nounwind {
 ; KNL-LABEL: test38:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vbroadcastsd (%rdi), %ymm2
 ; KNL-NEXT:    vcmpltpd %ymm2, %ymm0, %ymm2
 ; KNL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test38:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltpd (%rdi){1to4}, %ymm0, %k1
 ; SKX-NEXT:    vblendmpd %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %a = load double, double* %ptr
   %v = insertelement <4 x double> undef, double %a, i32 0
   %shuffle = shufflevector <4 x double> %v, <4 x double> undef, <4 x i32> zeroinitializer
 
   %mask = fcmp ogt <4 x double> %shuffle, %x
   %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1
   ret <4 x double> %max
 }
 
 define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nounwind {
 ; KNL-LABEL: test39:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vmovddup {{.*#+}} xmm2 = mem[0,0]
 ; KNL-NEXT:    vcmpltpd %xmm2, %xmm0, %xmm2
 ; KNL-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test39:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltpd (%rdi){1to2}, %xmm0, %k1
 ; SKX-NEXT:    vblendmpd %xmm0, %xmm1, %xmm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %a = load double, double* %ptr
   %v = insertelement <2 x double> undef, double %a, i32 0
   %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0>
 
   %mask = fcmp ogt <2 x double> %shuffle, %x
   %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1
   ret <2 x double> %max
 }
 
 
 define <16  x float> @test40(<16  x float> %x, <16  x float> %x1, float* %ptr) nounwind {
 ; CHECK-LABEL: test40:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcmpltps (%rdi){1to16}, %zmm0, %k1
 ; CHECK-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:    ## -- End function
 
   %a = load float, float* %ptr
   %v = insertelement <16  x float> undef, float %a, i32 0
   %shuffle = shufflevector <16  x float> %v, <16  x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
 
   %mask = fcmp ogt <16  x float> %shuffle, %x
   %max = select <16 x i1> %mask, <16  x float> %x, <16  x float> %x1
   ret <16  x float> %max
 }
 
 define <8  x float> @test41(<8  x float> %x, <8  x float> %x1, float* %ptr) nounwind {
 ; KNL-LABEL: test41:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    ## kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vbroadcastss (%rdi), %ymm2
 ; KNL-NEXT:    vcmpltps %zmm2, %zmm0, %k1
 ; KNL-NEXT:    vblendmps %zmm0, %zmm1, %zmm0 {%k1}
 ; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test41:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltps (%rdi){1to8}, %ymm0, %k1
 ; SKX-NEXT:    vblendmps %ymm0, %ymm1, %ymm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %a = load float, float* %ptr
   %v = insertelement <8  x float> undef, float %a, i32 0
   %shuffle = shufflevector <8  x float> %v, <8  x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
 
   %mask = fcmp ogt <8  x float> %shuffle, %x
   %max = select <8 x i1> %mask, <8  x float> %x, <8  x float> %x1
   ret <8  x float> %max
 }
 
 define <4  x float> @test42(<4  x float> %x, <4  x float> %x1, float* %ptr) nounwind {
 ; KNL-LABEL: test42:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vbroadcastss (%rdi), %xmm2
 ; KNL-NEXT:    vcmpltps %xmm2, %xmm0, %xmm2
 ; KNL-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test42:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vcmpltps (%rdi){1to4}, %xmm0, %k1
 ; SKX-NEXT:    vblendmps %xmm0, %xmm1, %xmm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %a = load float, float* %ptr
   %v = insertelement <4  x float> undef, float %a, i32 0
   %shuffle = shufflevector <4  x float> %v, <4  x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
 
   %mask = fcmp ogt <4  x float> %shuffle, %x
   %max = select <4 x i1> %mask, <4  x float> %x, <4  x float> %x1
   ret <4  x float> %max
 }
 
 define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x i1> %mask_in) nounwind {
 ; KNL-LABEL: test43:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vpmovsxwq %xmm2, %zmm2
 ; KNL-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; KNL-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; KNL-NEXT:    vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1}
 ; KNL-NEXT:    vblendmpd %zmm0, %zmm1, %zmm0 {%k1}
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test43:
 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vpsllw $15, %xmm2, %xmm2
 ; SKX-NEXT:    vpmovw2m %xmm2, %k1
 ; SKX-NEXT:    vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1}
 ; SKX-NEXT:    vblendmpd %zmm0, %zmm1, %zmm0 {%k1}
 ; SKX-NEXT:    retq
+; SKX-NEXT:    ## -- End function
 
   %a = load double, double* %ptr
   %v = insertelement <8 x double> undef, double %a, i32 0
   %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer
 
   %mask_cmp = fcmp ogt <8 x double> %shuffle, %x
   %mask = and <8 x i1> %mask_cmp, %mask_in
   %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1