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

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.

Reviewers: zvi
Differential Revision: https://reviews.llvm.org/D34933

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@306973 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/avx512-cmp.ll b/test/CodeGen/X86/avx512-cmp.ll
index eae7b94f5135c083f3c69f8bb3e29d81e30ebb18..b5a13404a230459ca567c8f043c096abdc9e227f 100644 (file)
--- a/test/CodeGen/X86/avx512-cmp.ll
+++ b/test/CodeGen/X86/avx512-cmp.ll
@@ -14,6 +14,7 @@ define double @test1(double %a, double %b) nounwind {
 ; 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
 
@@ -36,6 +37,7 @@ define float @test2(float %a, float %b) nounwind {
 ; 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
 

diff --git a/test/CodeGen/X86/avx512-insert-extract.ll b/test/CodeGen/X86/avx512-insert-extract.ll
index 29a5325a0ae9856d2cec621477c7283165c145cd..6a38180b939e40ae5eca8af8a8dde9e9eabf9864 100644 (file)
--- a/test/CodeGen/X86/avx512-insert-extract.ll
+++ b/test/CodeGen/X86/avx512-insert-extract.ll
@@ -12,6 +12,7 @@ define <16 x float> @test1(<16 x float> %x, float* %br, float %y) nounwind {
 ; 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:
@@ -21,6 +22,7 @@ define <16 x float> @test1(<16 x float> %x, float* %br, float %y) nounwind {
 ; 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
@@ -36,6 +38,7 @@ define <8 x double> @test2(<8 x double> %x, double* %br, double %y) nounwind {
 ; 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:
@@ -45,6 +48,7 @@ define <8 x double> @test2(<8 x double> %x, double* %br, double %y) nounwind {
 ; 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
@@ -58,6 +62,7 @@ define <16 x float> @test3(<16 x float> %x) nounwind {
 ; 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:
@@ -65,6 +70,7 @@ define <16 x float> @test3(<16 x float> %x) nounwind {
 ; 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
@@ -78,6 +84,7 @@ define <8 x i64> @test4(<8 x i64> %x) nounwind {
 ; 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:
@@ -86,6 +93,7 @@ define <8 x i64> @test4(<8 x i64> %x) nounwind {
 ; 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
@@ -96,11 +104,13 @@ define i32 @test5(<4 x float> %x) nounwind {
 ; 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
@@ -111,11 +121,13 @@ define void @test6(<4 x float> %x, float* %out) nounwind {
 ; 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
@@ -135,6 +147,7 @@ define float @test7(<16 x float> %x, i32 %ind) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test7:
 ; SKX:       ## BB#0:
@@ -150,6 +163,7 @@ define float @test7(<16 x float> %x, i32 %ind) nounwind {
 ; 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
 }
@@ -168,6 +182,7 @@ define double @test8(<8 x double> %x, i32 %ind) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test8:
 ; SKX:       ## BB#0:
@@ -183,6 +198,7 @@ define double @test8(<8 x double> %x, i32 %ind) nounwind {
 ; 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
 }
@@ -201,6 +217,7 @@ define float @test9(<8 x float> %x, i32 %ind) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test9:
 ; SKX:       ## BB#0:
@@ -216,6 +233,7 @@ define float @test9(<8 x float> %x, i32 %ind) nounwind {
 ; 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
 }
@@ -234,6 +252,7 @@ define i32 @test10(<16 x i32> %x, i32 %ind) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test10:
 ; SKX:       ## BB#0:
@@ -249,6 +268,7 @@ define i32 @test10(<16 x i32> %x, i32 %ind) nounwind {
 ; 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
 }

diff --git a/test/CodeGen/X86/avx512-vec-cmp.ll b/test/CodeGen/X86/avx512-vec-cmp.ll
index 2b04b9229b3d2a7d862be88405fac3cbfd5d157e..b3fbceea80a94f641887057a90993bd4871b149c 100644 (file)
--- a/test/CodeGen/X86/avx512-vec-cmp.ll
+++ b/test/CodeGen/X86/avx512-vec-cmp.ll
@@ -8,6 +8,7 @@ define <16 x float> @test1(<16 x float> %x, <16 x float> %y) nounwind {
 ; 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
@@ -19,6 +20,7 @@ define <8 x double> @test2(<8 x double> %x, <8 x double> %y) nounwind {
 ; 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
@@ -30,6 +32,7 @@ define <16 x i32> @test3(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %yp) nounwin
 ; 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
@@ -42,6 +45,7 @@ define <16 x i32> @test4_unsigned(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1)
 ; 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
@@ -53,6 +57,7 @@ define <8 x i64> @test5(<8 x i64> %x, <8 x i64> %y) nounwind {
 ; 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
@@ -64,6 +69,7 @@ define <8 x i64> @test6_unsigned(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1) noun
 ; 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
@@ -117,12 +123,14 @@ define <8 x i32> @test9(<8 x i32> %x, <8 x i32> %y) nounwind {
 ; 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
@@ -137,12 +145,14 @@ define <8 x float> @test10(<8 x float> %x, <8 x float> %y) nounwind {
 ; 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
@@ -154,6 +164,7 @@ define <8 x i32> @test11_unsigned(<8 x i32> %x, <8 x i32> %y) nounwind {
 ; 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
@@ -168,6 +179,7 @@ define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind {
 ; 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:
@@ -178,6 +190,7 @@ define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind {
 ; 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
@@ -330,6 +343,7 @@ define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test12_v32i32:
 ; SKX:       ## BB#0:
@@ -339,6 +353,7 @@ define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
 ; 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
@@ -642,6 +657,7 @@ define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind {
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    retq
+; KNL-NEXT:    ## -- End function
 ;
 ; SKX-LABEL: test12_v64i16:
 ; SKX:       ## BB#0:
@@ -651,6 +667,7 @@ define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind {
 ; 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
@@ -704,6 +721,7 @@ define <16 x i32> @test16(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind
 ; 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
@@ -715,6 +733,7 @@ define <16 x i32> @test17(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou
 ; 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
@@ -727,6 +746,7 @@ define <16 x i32> @test18(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou
 ; 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
@@ -739,6 +759,7 @@ define <16 x i32> @test19(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou
 ; 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
@@ -752,6 +773,7 @@ define <16 x i32> @test20(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i3
 ; 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
@@ -766,6 +788,7 @@ define <8 x i64> @test21(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y
 ; 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
@@ -780,6 +803,7 @@ define <8 x i64> @test22(<8 x i64> %x, <8 x i64>* %y.ptr, <8 x i64> %x1, <8 x i6
 ; 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
@@ -795,6 +819,7 @@ define <16 x i32> @test23(<16 x i32> %x, <16 x i32>* %y.ptr, <16 x i32> %x1, <16
 ; 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
@@ -809,6 +834,7 @@ define <8 x i64> @test24(<8 x i64> %x, <8 x i64> %x1, i64* %yb.ptr) nounwind {
 ; 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
@@ -823,6 +849,7 @@ define <16 x i32> @test25(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1) nounwind
 ; 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
@@ -838,6 +865,7 @@ define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32
 ; 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
@@ -855,6 +883,7 @@ define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y
 ; 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
@@ -920,12 +949,14 @@ define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind {
 ; 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
@@ -938,12 +969,14 @@ define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp
 ; 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
@@ -957,12 +990,14 @@ define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp
 ; 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
@@ -976,6 +1011,7 @@ define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp
 ; 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
@@ -988,12 +1024,14 @@ define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) no
 ; 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
@@ -1010,12 +1048,14 @@ define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) no
 ; 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
@@ -1029,6 +1069,7 @@ define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp
 ; 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
@@ -1041,6 +1082,7 @@ define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nou
 ; 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
@@ -1058,12 +1100,14 @@ define <4 x double> @test38(<4 x double> %x, <4 x double> %x1, double* %ptr) nou
 ; 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
@@ -1081,12 +1125,14 @@ define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nou
 ; 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
@@ -1104,6 +1150,7 @@ define <16  x float> @test40(<16  x float> %x, <16  x float> %x1, float* %ptr) n
 ; 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
@@ -1124,12 +1171,14 @@ define <8  x float> @test41(<8  x float> %x, <8  x float> %x1, float* %ptr) noun
 ; 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
@@ -1147,12 +1196,14 @@ define <4  x float> @test42(<4  x float> %x, <4  x float> %x1, float* %ptr) noun
 ; 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
@@ -1172,6 +1223,7 @@ define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x
 ; 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:
@@ -1180,6 +1232,7 @@ define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x
 ; 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