[X86] Add kortest intrinsics for 8, 32, and 64 bit masks. Add new intrinsic names...

This matches gcc and icc despite not being documented in the Intel Intrinsics Guide.

llvm-svn: 340798

diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index fc5096d..ce2288e 100644 (file)
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1749,8 +1749,14 @@ TARGET_BUILTIN(__builtin_ia32_korqi, "UcUcUc", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_korhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_korsi, "UiUiUi", "nc", "avx512bw")
 TARGET_BUILTIN(__builtin_ia32_kordi, "ULLiULLiULLi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kortestcqi, "iUcUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_kortestzqi, "iUcUc", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_kortestchi, "iUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_kortestzhi, "iUsUs", "nc", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_kortestcsi, "iUiUi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kortestzsi, "iUiUi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kortestcdi, "iULLiULLi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kortestzdi, "iULLiULLi", "nc", "avx512bw")
 TARGET_BUILTIN(__builtin_ia32_kunpckhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_kxnorqi, "UcUcUc", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_kxnorhi, "UsUsUs", "nc", "avx512f")

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 46c073e..116863a 100644 (file)
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -10012,14 +10012,21 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     return EmitX86MaskedCompare(*this, CC, false, Ops);
   }
 
+  case X86::BI__builtin_ia32_kortestcqi:
   case X86::BI__builtin_ia32_kortestchi:
-  case X86::BI__builtin_ia32_kortestzhi: {
+  case X86::BI__builtin_ia32_kortestcsi:
+  case X86::BI__builtin_ia32_kortestcdi: {
     Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
-    Value *C;
-    if (BuiltinID == X86::BI__builtin_ia32_kortestchi)
-      C = llvm::Constant::getAllOnesValue(Builder.getInt16Ty());
-    else
-      C = llvm::Constant::getNullValue(Builder.getInt16Ty());
+    Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType());
+    Value *Cmp = Builder.CreateICmpEQ(Or, C);
+    return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
+  }
+  case X86::BI__builtin_ia32_kortestzqi:
+  case X86::BI__builtin_ia32_kortestzhi:
+  case X86::BI__builtin_ia32_kortestzsi:
+  case X86::BI__builtin_ia32_kortestzdi: {
+    Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
+    Value *C = llvm::Constant::getNullValue(Ops[0]->getType());
     Value *Cmp = Builder.CreateICmpEQ(Or, C);
     return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
   }

diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 6f39b2e..a08833b 100644 (file)
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -107,6 +107,42 @@ _kxor_mask64(__mmask64 __A, __mmask64 __B)
   return (__mmask64)__builtin_ia32_kxordi((__mmask64)__A, (__mmask64)__B);
 }
 
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestc_mask32_u8(__mmask32 __A, __mmask32 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestcsi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestz_mask32_u8(__mmask32 __A, __mmask32 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestzsi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortest_mask32_u8(__mmask32 __A, __mmask32 __B, unsigned char *__C) {
+  *__C = (unsigned char)__builtin_ia32_kortestcsi(__A, __B);
+  return (unsigned char)__builtin_ia32_kortestzsi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestc_mask64_u8(__mmask64 __A, __mmask64 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestcdi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestz_mask64_u8(__mmask64 __A, __mmask64 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {
+  *__C = (unsigned char)__builtin_ia32_kortestcdi(__A, __B);
+  return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
+}
+
 /* Integer compare */
 
 #define _mm512_cmp_epi8_mask(a, b, p) \

diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index a887078..563e94b 100644 (file)
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -68,6 +68,24 @@ _kxor_mask8(__mmask8 __A, __mmask8 __B)
   return (__mmask8)__builtin_ia32_kxorqi((__mmask8)__A, (__mmask8)__B);
 }
 
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestc_mask8_u8(__mmask8 __A, __mmask8 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestcqi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestz_mask8_u8(__mmask8 __A, __mmask8 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestzqi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortest_mask8_u8(__mmask8 __A, __mmask8 __B, unsigned char *__C) {
+  *__C = (unsigned char)__builtin_ia32_kortestcqi(__A, __B);
+  return (unsigned char)__builtin_ia32_kortestzqi(__A, __B);
+}
+
 static __inline__ __m512i __DEFAULT_FN_ATTRS512
 _mm512_mullo_epi64 (__m512i __A, __m512i __B) {
   return (__m512i) ((__v8du) __A * (__v8du) __B);

diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 47cc86a..b5ef07a 100644 (file)
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8351,6 +8351,24 @@ _mm512_kortestz (__mmask16 __A, __mmask16 __B)
   return __builtin_ia32_kortestzhi ((__mmask16) __A, (__mmask16) __B);
 }
 
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestc_mask16_u8(__mmask16 __A, __mmask16 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestchi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortestz_mask16_u8(__mmask16 __A, __mmask16 __B)
+{
+  return (unsigned char)__builtin_ia32_kortestzhi(__A, __B);
+}
+
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
+_kortest_mask16_u8(__mmask16 __A, __mmask16 __B, unsigned char *__C) {
+  *__C = (unsigned char)__builtin_ia32_kortestchi(__A, __B);
+  return (unsigned char)__builtin_ia32_kortestzhi(__A, __B);
+}
+
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS
 _mm512_kunpackb (__mmask16 __A, __mmask16 __B)
 {

diff --git a/clang/test/CodeGen/avx512bw-builtins.c b/clang/test/CodeGen/avx512bw-builtins.c
index 7a3a7d0..808b1c0 100644 (file)
--- a/clang/test/CodeGen/avx512bw-builtins.c
+++ b/clang/test/CodeGen/avx512bw-builtins.c
@@ -134,6 +134,98 @@ __mmask64 test_kxor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
                                                    __E, __F);
 }
 
+unsigned char test_kortestz_mask32_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestz_mask32_u8
+  // CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[OR:%.*]] = or <32 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <32 x i1> [[OR]] to i32
+  // CHECK: [[CMP:%.*]] = icmp eq i32 [[CAST]], 0
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestz_mask32_u8(_mm512_cmpneq_epu16_mask(__A, __B),
+                             _mm512_cmpneq_epu16_mask(__C, __D));
+}
+
+unsigned char test_kortestc_mask32_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestc_mask32_u8
+  // CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[OR:%.*]] = or <32 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <32 x i1> [[OR]] to i32
+  // CHECK: [[CMP:%.*]] = icmp eq i32 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestc_mask32_u8(_mm512_cmpneq_epu16_mask(__A, __B),
+                             _mm512_cmpneq_epu16_mask(__C, __D));
+}
+
+unsigned char test_kortest_mask32_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, unsigned char *CF) {
+  // CHECK-LABEL: @test_kortest_mask32_u8
+  // CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[OR:%.*]] = or <32 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <32 x i1> [[OR]] to i32
+  // CHECK: [[CMP:%.*]] = icmp eq i32 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  // CHECK: [[LHS2:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RHS2:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[OR2:%.*]] = or <32 x i1> [[LHS2]], [[RHS2]]
+  // CHECK: [[CAST2:%.*]] = bitcast <32 x i1> [[OR2]] to i32
+  // CHECK: [[CMP2:%.*]] = icmp eq i32 [[CAST2]], 0
+  // CHECK: [[ZEXT2:%.*]] = zext i1 [[CMP2]] to i32
+  // CHECK: trunc i32 [[ZEXT2]] to i8
+  return _kortest_mask32_u8(_mm512_cmpneq_epu16_mask(__A, __B),
+                            _mm512_cmpneq_epu16_mask(__C, __D), CF);
+}
+
+unsigned char test_kortestz_mask64_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestz_mask64_u8
+  // CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[OR:%.*]] = or <64 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <64 x i1> [[OR]] to i64
+  // CHECK: [[CMP:%.*]] = icmp eq i64 [[CAST]], 0
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestz_mask64_u8(_mm512_cmpneq_epu8_mask(__A, __B),
+                             _mm512_cmpneq_epu8_mask(__C, __D));
+}
+
+unsigned char test_kortestc_mask64_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestc_mask64_u8
+  // CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[OR:%.*]] = or <64 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <64 x i1> [[OR]] to i64
+  // CHECK: [[CMP:%.*]] = icmp eq i64 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestc_mask64_u8(_mm512_cmpneq_epu8_mask(__A, __B),
+                             _mm512_cmpneq_epu8_mask(__C, __D));
+}
+
+unsigned char test_kortest_mask64_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, unsigned char *CF) {
+  // CHECK-LABEL: @test_kortest_mask64_u8
+  // CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[OR:%.*]] = or <64 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <64 x i1> [[OR]] to i64
+  // CHECK: [[CMP:%.*]] = icmp eq i64 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  // CHECK: [[LHS2:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RHS2:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[OR2:%.*]] = or <64 x i1> [[LHS2]], [[RHS2]]
+  // CHECK: [[CAST2:%.*]] = bitcast <64 x i1> [[OR2]] to i64
+  // CHECK: [[CMP2:%.*]] = icmp eq i64 [[CAST2]], 0
+  // CHECK: [[ZEXT2:%.*]] = zext i1 [[CMP2]] to i32
+  // CHECK: trunc i32 [[ZEXT2]] to i8
+  return _kortest_mask64_u8(_mm512_cmpneq_epu8_mask(__A, __B),
+                            _mm512_cmpneq_epu8_mask(__C, __D), CF);
+}
+
 __mmask64 test_mm512_cmpeq_epi8_mask(__m512i __a, __m512i __b) {
   // CHECK-LABEL: @test_mm512_cmpeq_epi8_mask
   // CHECK: icmp eq <64 x i8> %{{.*}}, %{{.*}}

diff --git a/clang/test/CodeGen/avx512dq-builtins.c b/clang/test/CodeGen/avx512dq-builtins.c
index 0dd0049..ee339a0 100644 (file)
--- a/clang/test/CodeGen/avx512dq-builtins.c
+++ b/clang/test/CodeGen/avx512dq-builtins.c
@@ -68,6 +68,52 @@ __mmask8 test_kxor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
                                                    __E, __F);
 }
 
+unsigned char test_kortestz_mask8_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestz_mask8_u8
+  // CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[OR:%.*]] = or <8 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <8 x i1> [[OR]] to i8
+  // CHECK: [[CMP:%.*]] = icmp eq i8 [[CAST]], 0
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestz_mask8_u8(_mm512_cmpneq_epu64_mask(__A, __B),
+                            _mm512_cmpneq_epu64_mask(__C, __D));
+}
+
+unsigned char test_kortestc_mask8_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestc_mask8_u8
+  // CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[OR:%.*]] = or <8 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <8 x i1> [[OR]] to i8
+  // CHECK: [[CMP:%.*]] = icmp eq i8 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestc_mask8_u8(_mm512_cmpneq_epu64_mask(__A, __B),
+                            _mm512_cmpneq_epu64_mask(__C, __D));
+}
+
+unsigned char test_kortest_mask8_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, unsigned char *CF) {
+  // CHECK-LABEL: @test_kortest_mask8_u8
+  // CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[OR:%.*]] = or <8 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <8 x i1> [[OR]] to i8
+  // CHECK: [[CMP:%.*]] = icmp eq i8 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  // CHECK: [[LHS2:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RHS2:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[OR2:%.*]] = or <8 x i1> [[LHS2]], [[RHS2]]
+  // CHECK: [[CAST2:%.*]] = bitcast <8 x i1> [[OR2]] to i8
+  // CHECK: [[CMP2:%.*]] = icmp eq i8 [[CAST2]], 0
+  // CHECK: [[ZEXT2:%.*]] = zext i1 [[CMP2]] to i32
+  // CHECK: trunc i32 [[ZEXT2]] to i8
+  return _kortest_mask8_u8(_mm512_cmpneq_epu64_mask(__A, __B),
+                           _mm512_cmpneq_epu64_mask(__C, __D), CF);
+}
+
 __m512i test_mm512_mullo_epi64 (__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mullo_epi64
   // CHECK: mul <8 x i64>

diff --git a/clang/test/CodeGen/avx512f-builtins.c b/clang/test/CodeGen/avx512f-builtins.c
index 326d6d4..39ac42f 100644 (file)
--- a/clang/test/CodeGen/avx512f-builtins.c
+++ b/clang/test/CodeGen/avx512f-builtins.c
@@ -8149,6 +8149,52 @@ int test_mm512_kortestz(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
                          _mm512_cmpneq_epu32_mask(__C, __D));
 }
 
+unsigned char test_kortestz_mask16_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestz_mask16_u8
+  // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[OR:%.*]] = or <16 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <16 x i1> [[OR]] to i16
+  // CHECK: [[CMP:%.*]] = icmp eq i16 [[CAST]], 0
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestz_mask16_u8(_mm512_cmpneq_epu32_mask(__A, __B),
+                             _mm512_cmpneq_epu32_mask(__C, __D));
+}
+
+unsigned char test_kortestc_mask16_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
+  // CHECK-LABEL: @test_kortestc_mask16_u8
+  // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[OR:%.*]] = or <16 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <16 x i1> [[OR]] to i16
+  // CHECK: [[CMP:%.*]] = icmp eq i16 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  return _kortestc_mask16_u8(_mm512_cmpneq_epu32_mask(__A, __B),
+                             _mm512_cmpneq_epu32_mask(__C, __D));
+}
+
+unsigned char test_kortest_mask16_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, unsigned char *CF) {
+  // CHECK-LABEL: @test_kortest_mask16_u8
+  // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[OR:%.*]] = or <16 x i1> [[LHS]], [[RHS]]
+  // CHECK: [[CAST:%.*]] = bitcast <16 x i1> [[OR]] to i16
+  // CHECK: [[CMP:%.*]] = icmp eq i16 [[CAST]], -1
+  // CHECK: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
+  // CHECK: trunc i32 [[ZEXT]] to i8
+  // CHECK: [[LHS2:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RHS2:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[OR2:%.*]] = or <16 x i1> [[LHS2]], [[RHS2]]
+  // CHECK: [[CAST2:%.*]] = bitcast <16 x i1> [[OR2]] to i16
+  // CHECK: [[CMP2:%.*]] = icmp eq i16 [[CAST2]], 0
+  // CHECK: [[ZEXT2:%.*]] = zext i1 [[CMP2]] to i32
+  // CHECK: trunc i32 [[ZEXT2]] to i8
+  return _kortest_mask16_u8(_mm512_cmpneq_epu32_mask(__A, __B),
+                            _mm512_cmpneq_epu32_mask(__C, __D), CF);
+}
+
 __mmask16 test_mm512_kunpackb(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
   // CHECK-LABEL: @test_mm512_kunpackb
   // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>