diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -12479,8 +12479,8 @@
                                         const CallExpr *E,
                                         ArrayRef<Value *> Ops) {
   llvm::Type *Int32Ty = CGF.Builder.getInt32Ty();
-  Value *ZeroExt = CGF.Builder.CreateZExt(Ops[0], Int32Ty);
-  Value *Shl = CGF.Builder.CreateShl(ZeroExt, 16);
+  Value *SignExt = CGF.Builder.CreateSExt(Ops[0], Int32Ty);
+  Value *Shl = CGF.Builder.CreateShl(SignExt, 16);
   llvm::Type *ResultType = CGF.ConvertType(E->getType());
   Value *BitCast = CGF.Builder.CreateBitCast(Shl, ResultType);
   return BitCast;
diff --git a/clang/lib/Headers/avx512bf16intrin.h b/clang/lib/Headers/avx512bf16intrin.h
--- a/clang/lib/Headers/avx512bf16intrin.h
+++ b/clang/lib/Headers/avx512bf16intrin.h
@@ -15,7 +15,6 @@
 
 typedef short __m512bh __attribute__((__vector_size__(64), __aligned__(64)));
 typedef short __m256bh __attribute__((__vector_size__(32), __aligned__(32)));
-typedef unsigned short __bfloat16;
 
 #define __DEFAULT_FN_ATTRS512 \
   __attribute__((__always_inline__, __nodebug__, __target__("avx512bf16"), \
@@ -33,7 +32,7 @@
 ///    A bfloat data.
 /// \returns A float data whose sign field and exponent field keep unchanged,
 ///    and fraction field is extended to 23 bits.
-static __inline__ float __DEFAULT_FN_ATTRS _mm_cvtsbh_ss(__bfloat16 __A) {
+static __inline__ float __DEFAULT_FN_ATTRS _mm_cvtsbh_ss(short __A) {
   return __builtin_ia32_cvtsbf162ss_32(__A);
 }
 
diff --git a/clang/lib/Headers/avx512vlbf16intrin.h b/clang/lib/Headers/avx512vlbf16intrin.h
--- a/clang/lib/Headers/avx512vlbf16intrin.h
+++ b/clang/lib/Headers/avx512vlbf16intrin.h
@@ -413,7 +413,7 @@
 ///    A float data.
 /// \returns A bf16 data whose sign field and exponent field keep unchanged,
 ///    and fraction field is truncated to 7 bits.
-static __inline__ __bfloat16 __DEFAULT_FN_ATTRS128 _mm_cvtness_sbh(float __A) {
+static __inline__ short __DEFAULT_FN_ATTRS128 _mm_cvtness_sbh(float __A) {
   __v4sf __V = {__A, 0, 0, 0};
   __v8hi __R = __builtin_ia32_cvtneps2bf16_128_mask(
       (__v4sf)__V, (__v8hi)_mm_undefined_si128(), (__mmask8)-1);
diff --git a/clang/test/CodeGen/X86/avx512bf16-builtins.c b/clang/test/CodeGen/X86/avx512bf16-builtins.c
--- a/clang/test/CodeGen/X86/avx512bf16-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bf16-builtins.c
@@ -4,9 +4,9 @@
 
 #include <immintrin.h>
 
-float test_mm_cvtsbh_ss(__bfloat16 A) {
+float test_mm_cvtsbh_ss(short A) {
   // CHECK-LABEL: @test_mm_cvtsbh_ss
-  // CHECK: zext i16 %{{.*}} to i32
+  // CHECK: sext i16 %{{.*}} to i32
   // CHECK: shl i32 %{{.*}}, 16
   // CHECK: bitcast i32 %{{.*}} to float
   // CHECK: ret float %{{.*}}
diff --git a/clang/test/CodeGen/X86/avx512vlbf16-builtins.c b/clang/test/CodeGen/X86/avx512vlbf16-builtins.c
--- a/clang/test/CodeGen/X86/avx512vlbf16-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vlbf16-builtins.c
@@ -162,7 +162,7 @@
   return _mm256_mask_dpbf16_ps(D, U, A, B);
 }
 
-__bfloat16 test_mm_cvtness_sbh(float A) {
+short test_mm_cvtness_sbh(float A) {
   // CHECK-LABEL: @test_mm_cvtness_sbh
   // CHECK: @llvm.x86.avx512bf16.mask.cvtneps2bf16.128
   // CHECK: ret i16 %{{.*}}