[X86][AVX512] Split AVX512F and AVX512BW bool-vector bitcast tests

llvm-svn: 317020

1 files changed, 236 insertions, 106 deletions

diff --git a/llvm/test/CodeGen/X86/bitcast-setcc-128.ll b/llvm/test/CodeGen/X86/bitcast-setcc-128.ll
index 773dc0b0d62..b8a176a71d7 100644
--- a/llvm/test/CodeGen/X86/bitcast-setcc-128.ll
+++ b/llvm/test/CodeGen/X86/bitcast-setcc-128.ll
@@ -3,7 +3,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512 --check-prefixes=AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512 --check-prefixes=AVX512BW
 
 define i8 @v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-SSSE3-LABEL: v8i16:
@@ -22,12 +23,23 @@ define i8 @v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v8i16:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v8i16:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmovsxwq %xmm0, %zmm0
+; AVX512F-NEXT:    vpsllq $63, %zmm0, %zmm0
+; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v8i16:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <8 x i16> %a, %b
   %res = bitcast <8 x i1> %x to i8
   ret i8 %res
@@ -48,13 +60,21 @@ define i4 @v4i32(<4 x i32> %a, <4 x i32> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v4i32:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v4i32:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v4i32:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <4 x i32> %a, %b
   %res = bitcast <4 x i1> %x to i4
   ret i4 %res
@@ -75,13 +95,21 @@ define i4 @v4f32(<4 x float> %a, <4 x float> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v4f32:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vcmpltps %xmm0, %xmm1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v4f32:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vcmpltps %xmm0, %xmm1, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v4f32:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vcmpltps %xmm0, %xmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = fcmp ogt <4 x float> %a, %b
   %res = bitcast <4 x i1> %x to i4
   ret i4 %res
@@ -102,12 +130,23 @@ define i16 @v16i8(<16 x i8> %a, <16 x i8> %b) {
 ; AVX12-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v16i8:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v16i8:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
+; AVX512F-NEXT:    vpslld $31, %zmm0, %zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v16i8:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <16 x i8> %a, %b
   %res = bitcast <16 x i1> %x to i16
   ret i16 %res
@@ -179,17 +218,29 @@ define i2 @v2i8(<2 x i8> %a, <2 x i8> %b) {
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: v2i8:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsllq $56, %xmm1, %xmm1
-; AVX512-NEXT:    vpsraq $56, %xmm1, %xmm1
-; AVX512-NEXT:    vpsllq $56, %xmm0, %xmm0
-; AVX512-NEXT:    vpsraq $56, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v2i8:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpsllq $56, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsraq $56, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsllq $56, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsraq $56, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v2i8:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpsllq $56, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsraq $56, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsllq $56, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsraq $56, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <2 x i8> %a, %b
   %res = bitcast <2 x i1> %x to i2
   ret i2 %res
@@ -261,17 +312,29 @@ define i2 @v2i16(<2 x i16> %a, <2 x i16> %b) {
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: v2i16:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsllq $48, %xmm1, %xmm1
-; AVX512-NEXT:    vpsraq $48, %xmm1, %xmm1
-; AVX512-NEXT:    vpsllq $48, %xmm0, %xmm0
-; AVX512-NEXT:    vpsraq $48, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v2i16:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpsllq $48, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsraq $48, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsllq $48, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsraq $48, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v2i16:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpsllq $48, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsraq $48, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsllq $48, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsraq $48, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <2 x i16> %a, %b
   %res = bitcast <2 x i1> %x to i2
   ret i2 %res
@@ -335,17 +398,29 @@ define i2 @v2i32(<2 x i32> %a, <2 x i32> %b) {
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: v2i32:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsllq $32, %xmm1, %xmm1
-; AVX512-NEXT:    vpsraq $32, %xmm1, %xmm1
-; AVX512-NEXT:    vpsllq $32, %xmm0, %xmm0
-; AVX512-NEXT:    vpsraq $32, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v2i32:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpsllq $32, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsraq $32, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsllq $32, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsraq $32, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v2i32:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpsllq $32, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsraq $32, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsllq $32, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsraq $32, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <2 x i32> %a, %b
   %res = bitcast <2 x i1> %x to i2
   ret i2 %res
@@ -376,13 +451,21 @@ define i2 @v2i64(<2 x i64> %a, <2 x i64> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v2i64:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v2i64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v2i64:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <2 x i64> %a, %b
   %res = bitcast <2 x i1> %x to i2
   ret i2 %res
@@ -403,13 +486,21 @@ define i2 @v2f64(<2 x double> %a, <2 x double> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v2f64:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v2f64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vcmpltpd %xmm0, %xmm1, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v2f64:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vcmpltpd %xmm0, %xmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = fcmp ogt <2 x double> %a, %b
   %res = bitcast <2 x i1> %x to i2
   ret i2 %res
@@ -438,17 +529,29 @@ define i4 @v4i8(<4 x i8> %a, <4 x i8> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v4i8:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpslld $24, %xmm1, %xmm1
-; AVX512-NEXT:    vpsrad $24, %xmm1, %xmm1
-; AVX512-NEXT:    vpslld $24, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrad $24, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v4i8:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpslld $24, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsrad $24, %xmm1, %xmm1
+; AVX512F-NEXT:    vpslld $24, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsrad $24, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v4i8:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpslld $24, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsrad $24, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpslld $24, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsrad $24, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <4 x i8> %a, %b
   %res = bitcast <4 x i1> %x to i4
   ret i4 %res
@@ -477,17 +580,29 @@ define i4 @v4i16(<4 x i16> %a, <4 x i16> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v4i16:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpslld $16, %xmm1, %xmm1
-; AVX512-NEXT:    vpsrad $16, %xmm1, %xmm1
-; AVX512-NEXT:    vpslld $16, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX512-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v4i16:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpslld $16, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsrad $16, %xmm1, %xmm1
+; AVX512F-NEXT:    vpslld $16, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsrad $16, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v4i16:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpslld $16, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsrad $16, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpslld $16, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsrad $16, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; AVX512BW-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <4 x i16> %a, %b
   %res = bitcast <4 x i1> %x to i4
   ret i4 %res
@@ -518,16 +633,31 @@ define i8 @v8i8(<8 x i8> %a, <8 x i8> %b) {
 ; AVX12-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX12-NEXT:    retq
 ;
-; AVX512-LABEL: v8i8:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsllw $8, %xmm1, %xmm1
-; AVX512-NEXT:    vpsraw $8, %xmm1, %xmm1
-; AVX512-NEXT:    vpsllw $8, %xmm0, %xmm0
-; AVX512-NEXT:    vpsraw $8, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: v8i8:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpsllw $8, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsraw $8, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsllw $8, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsraw $8, %xmm0, %xmm0
+; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmovsxwq %xmm0, %zmm0
+; AVX512F-NEXT:    vpsllq $63, %zmm0, %zmm0
+; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: v8i8:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpsllw $8, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsraw $8, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsllw $8, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsraw $8, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX512BW-NEXT:    retq
   %x = icmp sgt <8 x i8> %a, %b
   %res = bitcast <8 x i1> %x to i8
   ret i8 %res