[Clang][Avx512][BuiltIn] Adding brodcast builtin to clang

[igorb]

include/clang/Basic/BuiltinsX86.def

1981

it should be "avx512bw" ?

lib/Headers/avx512bwintrin.h

2065

could you please put casting on the same line with argument

[igorb]

lib/Headers/avx512bwintrin.h

2080

could you please put casting on the same line with argument ( the same regarding the code below)

[igorb]

lib/Headers/avx512dqintrin.h

175

could you please put casting on the same line with argument

214

could you please put casting on the same line with argument ( the same regarding the code below)

252

could you please put casting on the same line with argument

288

could you please put casting on the same line with argument

lib/Headers/avx512vldqintrin.h

170

could you please put casting on the same line with argument ( the same regarding the code below)

[igorb]

Thanks!

LGTM

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_shufps128_mask, "V4fV4fV4fIiV4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_shufps256_mask, "V8fV8fV8fIiV8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_sqrtsd_round_mask, "V2dV2dV2dV2dUcIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_sqrtss_round_mask, "V4fV4fV4fV4fUcIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_rsqrt14pd128_mask, "V2dV2dV2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_rsqrt14pd256_mask, "V4dV4dV4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_rsqrt14ps128_mask, "V4fV4fV4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_rsqrt14ps256_mask, "V8fV8fV8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_512_mask, "V16fV4fV16fUs","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcastf32x8_512_mask, "V16fV8fV16fUs","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_512_mask, "V8dV2dV8dUc","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_512_mask, "V16iV4iV16iUs","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x8_512_mask, "V16iV8iV16iUs","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcasti64x2_512_mask, "V8LLiV2LLiV8LLiUc","","avx512dq")
+TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_256_mask, "V8fV4fV8fUc","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_256_mask, "V4dV2dV4dUc","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_128_mask, "V4iV4iV4iUc","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_256_mask, "V8iV4iV8iUc","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcasti64x2_256_mask, "V4LLiV2LLiV4LLiUc","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastf32x4_256_mask, "V8fV4fV8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x4_256_mask, "V8iV4iV8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastsd256_mask, "V4dV2dV4dUc","","")
+TARGET_BUILTIN(__builtin_ia32_broadcastss128_mask, "V4fV4fV4fUc","","")
+TARGET_BUILTIN(__builtin_ia32_broadcastss256_mask, "V8fV4fV8fUc","","")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw512_mask, "V32sV8sV32sUi","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastb128_mask, "V16cV16cV16cUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastb256_mask, "V32cV16cV32cUi","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw128_mask, "V8sV8sV8sUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw256_mask, "V16sV8sV16sUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastd128_mask, "V4iV4iV4iUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastd256_mask, "V8iV4iV8iUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastq128_mask, "V2LLiV2LLiV2LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastq256_mask, "V4LLiV2LLiV4LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastb512_mask, "V64cV16cV64cULLi","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw512_gpr_mask, "V32shV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw256_gpr_mask, "V16shV16sUs","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pbroadcastw128_gpr_mask, "V8ssV8sUc","","avx512bw,avx512vl")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

lib/Headers/avx512bwintrin.h

 
 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
 _mm512_mask_testn_epi16_mask (__mmask32 __U, __m512i __A, __m512i __B)
 {
   return (__mmask32) __builtin_ia32_ptestnmw512 ((__v32hi) __A,
              (__v32hi) __B, __U);
 }
 
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcastb_epi8 (__m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
+                   (__v64qi)_mm512_setzero_si512(),
+                   (__mmask64) -
+                   1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcastb_epi8 (__m512i __O, __mmask64 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
+                   (__v64qi) __O,
+                   __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcastb_epi8 (__mmask64 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
+                   (__v64qi)
+                   _mm512_setzero_qi(),
+                   __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_set1_epi16 (__m512i __O, __mmask32 __M, short __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastw512_gpr_mask (__A,
+                 (__v32hi) __O,
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_set1_epi16 (__mmask32 __M, short __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastw512_gpr_mask (__A,
+                 (__v32hi)
+                 _mm512_setzero_hi(),
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcastw_epi16 (__m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
+                   (__v32hi)_mm512_setzero_si512(),
+                   (__mmask32)-1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcastw_epi16 (__m512i __O, __mmask32 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
+                   (__v32hi) __O,
+                   __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcastw_epi16 (__mmask32 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
+                   (__v32hi)
+                   _mm512_setzero_hi(),
+                   __M);
+}
 
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512dqintrin.h

 _mm512_and_ps (__m512 __A, __m512 __B) {
   return (__m512) ((__v16si) __A & (__v16si) __B);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_mask_and_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
   return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf) __W,
             (__mmask16) __U);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_maskz_and_ps (__mmask16 __U, __m512 __A, __m512 __B) {
   return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf)

 _mm512_maskz_andnot_pd (__mmask8 __U, __m512d __A, __m512d __B) {
   return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
               (__v8df) __B,
               (__v8df)
               _mm512_setzero_pd (),
               (__mmask8) __U);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_andnot_ps (__m512 __A, __m512 __B) {
   return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
              (__v16sf) __B,
              (__v16sf)
              _mm512_setzero_ps (),
              (__mmask16) -1);
 }
 
return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A,
                 (__mmask8) -1,
                 _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtpd_epi64 (__m512i __W, __mmask8 __U, __m512d __A) {
   return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A,
                 (__v8di) __W,
                 (__mmask8) __U,
                 _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtpd_epi64 (__mmask8 __U, __m512d __A) {
   return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A,
                 (__v8di) _mm512_setzero_si512(),
                 (__mmask8) __U,
return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A,
                  (__mmask8) -1,
                  _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtpd_epu64 (__m512i __W, __mmask8 __U, __m512d __A) {
   return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A,
                  (__v8di) __W,
                  (__mmask8) __U,
                  _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtpd_epu64 (__mmask8 __U, __m512d __A) {
   return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A,
                  (__v8di) _mm512_setzero_si512(),
                  (__mmask8) __U,

 #define _mm512_mask_reduce_round_ps(__W, __U, __A, __B, __R) __extension__ ({\
   (__m512) __builtin_ia32_reduceps512_mask ((__v16sf) __A, __B,      \
                (__v16sf) __W, (__mmask16) __U, __R);})
 
 #define _mm512_maskz_reduce_round_ps(__U, __A, __B, __R) __extension__ ({\
   (__m512) __builtin_ia32_reduceps512_mask ((__v16sf) __A, __B,      \
                (__v16sf) _mm512_setzero_ps(), (__mmask16) __U, __R);})
 
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_broadcast_f32x2 (__m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A,
+                (__v16sf)_mm512_undefined_ps(),
+                (__mmask16) - 1);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_f32x2 (__m512 __O, __mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A,
+                (__v16sf)
+                __O, __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A,
+                (__v16sf)
+                _mm512_setzero_ps (),
+                __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_broadcast_f32x8 (__m256 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
+                _mm512_undefined_ps(),
+                (__mmask16) - 1);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_f32x8 (__m512 __O, __mmask16 __M, __m256 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
+                (__v16sf)__O,
+                __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_f32x8 (__mmask16 __M, __m256 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
+                (__v16sf)
+                _mm512_setzero_ps (),
+                __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_broadcast_f64x2 (__m128d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df)
+                 __A,
+                 _mm512_undefined_pd(),
+                 (__mmask8) - 1);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_f64x2 (__m512d __O, __mmask8 __M, __m128d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df)
+                 __A,
+                 (__v8df)
+                 __O, __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df)
+                 __A,
+                 (__v8df)
+                 _mm512_setzero_ps (),
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcast_i32x2 (__m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si)
+                 __A,
+                 (__v16si)_mm512_setzero_si512(),
+                 (__mmask16) - 1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_i32x2 (__m512i __O, __mmask16 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si)
+                 __A,
+                 (__v16si)
+                 __O, __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si)
+                 __A,
+                 (__v16si)
+                 _mm512_setzero_si512 (),
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcast_i32x8 (__m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si)
+                 __A,
+                 (__v16si)_mm512_setzero_si512(),
+                 (__mmask16) - 1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_i32x8 (__m512i __O, __mmask16 __M, __m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si)
+                 __A,
+                 (__v16si)__O,
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_i32x8 (__mmask16 __M, __m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si)
+                 __A,
+                 (__v16si)
+                 _mm512_setzero_si512 (),
+                 __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcast_i64x2 (__m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di)
+                 __A,
+                 _mm512_setzero_si512(),
+                 (__mmask8) - 1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_i64x2 (__m512i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di)
+                 __A,
+                 (__v8di)
+                 __O, __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di)
+                 __A,
+                 (__v8di)
+                 _mm512_setzero_si512 (),
+                 __M);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512vlbwintrin.h

 
 #define _mm256_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \
 __builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\
               (__v16hi)\
               _mm256_setzero_si256 (),\
               (__mmask16)( __U));\
 })
 
-
-
-
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_mask_mov_epi16 (__m128i __W, __mmask8 __U, __m128i __A)
 {
   return (__m128i) __builtin_ia32_movdquhi128_mask ((__v8hi) __A,
                 (__v8hi) __W,
                 (__mmask8) __U);
 }
 

 
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS
 _mm256_mask_testn_epi16_mask (__mmask16 __U, __m256i __A, __m256i __B)
 {
   return (__mmask16) __builtin_ia32_ptestnmw256 ((__v16hi) __A,
              (__v16hi) __B, __U);
 }
 
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_broadcastb_epi8 (__m128i __O, __mmask16 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastb128_mask ((__v16qi) __A,
+                   (__v16qi) __O,
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_broadcastb_epi8 (__mmask16 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastb128_mask ((__v16qi) __A,
+                   (__v16qi)
+                   _mm_setzero_si128 (),
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastb_epi8 (__m256i __O, __mmask32 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastb256_mask ((__v16qi) __A,
+                   (__v32qi) __O,
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastb_epi8 (__mmask32 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastb256_mask ((__v16qi) __A,
+                   (__v32qi)
+                   _mm256_setzero_si256 (),
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_broadcastw_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastw128_mask ((__v8hi) __A,
+                   (__v8hi) __O,
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_broadcastw_epi16 (__mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastw128_mask ((__v8hi) __A,
+                   (__v8hi)
+                   _mm_setzero_si128 (),
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastw_epi16 (__m256i __O, __mmask16 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastw256_mask ((__v8hi) __A,
+                   (__v16hi) __O,
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastw_epi16 (__mmask16 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastw256_mask ((__v8hi) __A,
+                   (__v16hi)
+                   _mm256_setzero_si256 (),
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_set1_epi16 (__m256i __O, __mmask16 __M, short __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastw256_gpr_mask (__A,
+                 (__v16hi) __O,
+                 __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_set1_epi16 (__mmask16 __M, short __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastw256_gpr_mask (__A,
+                 (__v16hi)
+                 _mm256_setzero_si256 (),
+                 __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_set1_epi16 (__m128i __O, __mmask8 __M, short __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastw128_gpr_mask (__A,
+                 (__v8hi) __O,
+                 __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_set1_epi16 (__mmask8 __M, short __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastw128_gpr_mask (__A,
+                 (__v8hi)
+                 _mm_setzero_si128 (),
+                 __M);
+}
+
 
 
 #undef __DEFAULT_FN_ATTRS
 
 #endif /* __AVX512VLBWINTRIN_H */

lib/Headers/avx512vldqintrin.h

 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask_and_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
   return (__m128d) __builtin_ia32_andpd128_mask ((__v2df) __A,
              (__v2df) __B,
              (__v2df) __W,
              (__mmask8) __U);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_maskz_and_pd (__mmask8 __U, __m128d __A, __m128d __B) {
   return (__m128d) __builtin_ia32_andpd128_mask ((__v2df) __A,
              (__v2df) __B,
              (__v2df)
              _mm_setzero_pd (),
              (__mmask8) __U);
 }

 #define _mm256_mask_reduce_ps(__W, __U, __A, __B) __extension__ ({ \
   (__m256) __builtin_ia32_reduceps256_mask ((__v8sf) __A, __B,     \
                 (__v8sf) __W, (__mmask8) __U); })
 
 #define _mm256_maskz_reduce_ps(__U, __A, __B) __extension__ ({     \
   (__m256) __builtin_ia32_reduceps256_mask ((__v8sf) __A, __B,     \
                 (__v8sf) _mm256_setzero_ps(), (__mmask8) __U); })
 
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_broadcast_f32x2 (__m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A,
+                      (__v8sf)_mm256_undefined_ps(),
+                (__mmask8) -
+                1);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_f32x2 (__m256 __O, __mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A,
+                (__v8sf) __O,
+                __M);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_f32x2 (__mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A,
+                (__v8sf)
+                _mm256_setzero_ps (),
+                __M);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS
+_mm256_broadcast_f64x2 (__m128d __A)
+{
+  return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df)
+                 __A,
+                       (__v4df)_mm256_undefined_pd(),
+                 (__mmask8) -
+                 1);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_f64x2 (__m256d __O, __mmask8 __M, __m128d __A)
+{
+  return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df)
+                 __A,
+                 (__v4df)
+                 __O, __M);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A)
+{
+  return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df)
+                 __A,
+                 (__v4df)
+                 _mm256_setzero_ps (),
+                 __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_broadcast_i32x2 (__m128i __A)
+{
+  return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si)
+                 __A,
+                      (__v4si)_mm_undefined_si128(),
+                 (__mmask8) -
+                 1);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_broadcast_i32x2 (__m128i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si)
+                 __A,
+                 (__v4si)
+                 __O, __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si)
+                 __A,
+                 (__v4si)
+                 _mm_setzero_si128 (),
+                 __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_broadcast_i32x2 (__m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si)
+                 __A,
+                      (__v8si)_mm256_undefined_si256(),
+                 (__mmask8) -
+                 1);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_i32x2 (__m256i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si)
+                 __A,
+                 (__v8si)
+                 __O, __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si)
+                 __A,
+                 (__v8si)
+                 _mm256_setzero_si256 (),
+                 __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_broadcast_i64x2 (__m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di)
+                 __A,
+                       (__v4di)_mm256_undefined_si256(),
+                 (__mmask8) -
+                 1);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_i64x2 (__m256i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di)
+                 __A,
+                 (__v4di)
+                 __O, __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di)
+                 __A,
+                 (__v4di)
+                 _mm256_setzero_si256 (),
+                 __M);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512vlintrin.h

 _mm256_maskz_rsqrt14_ps (__mmask8 __U, __m256 __A)
 {
   return (__m256) __builtin_ia32_rsqrt14ps256_mask ((__v8sf) __A,
                 (__v8sf)
                 _mm256_setzero_ps (),
                 (__mmask8) __U);
 }
 
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_broadcast_f32x4 (__m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
+                      (__v8sf)_mm256_undefined_pd (),
+                (__mmask8) - 1);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_f32x4 (__m256 __O, __mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
+                (__v8sf) __O,
+                __M);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_f32x4 (__mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
+                (__v8sf)
+                _mm256_setzero_ps (),
+                __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_broadcast_i32x4 (__m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
+                 __A,
+                       (__v8si)_mm256_undefined_si256 (),
+                 (__mmask8) - 1);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcast_i32x4 (__m256i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
+                 __A,
+                 (__v8si)
+                 __O, __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcast_i32x4 (__mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
+                 __A,
+                 (__v8si)
+                 _mm256_setzero_si256 (),
+                 __M);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastsd_pd (__m256d __O, __mmask8 __M, __m128d __A)
+{
+  return (__m256d) __builtin_ia32_broadcastsd256_mask ((__v2df) __A,
+                   (__v4df) __O,
+                   __M);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastsd_pd (__mmask8 __M, __m128d __A)
+{
+  return (__m256d) __builtin_ia32_broadcastsd256_mask ((__v2df) __A,
+                   (__v4df)
+                   _mm256_setzero_pd (),
+                   __M);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_broadcastss_ps (__m128 __O, __mmask8 __M, __m128 __A)
+{
+  return (__m128) __builtin_ia32_broadcastss128_mask ((__v4sf) __A,
+                  (__v4sf) __O,
+                  __M);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_broadcastss_ps (__mmask8 __M, __m128 __A)
+{
+  return (__m128) __builtin_ia32_broadcastss128_mask ((__v4sf) __A,
+                  (__v4sf)
+                  _mm_setzero_ps (),
+                  __M);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastss_ps (__m256 __O, __mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastss256_mask ((__v4sf) __A,
+                  (__v8sf) __O,
+                  __M);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastss_ps (__mmask8 __M, __m128 __A)
+{
+  return (__m256) __builtin_ia32_broadcastss256_mask ((__v4sf) __A,
+                  (__v8sf)
+                  _mm256_setzero_ps (),
+                  __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_broadcastd_epi32 (__m128i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastd128_mask ((__v4si) __A,
+                   (__v4si) __O,
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_broadcastd_epi32 (__mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastd128_mask ((__v4si) __A,
+                   (__v4si)
+                   _mm_setzero_si128 (),
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastd_epi32 (__m256i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastd256_mask ((__v4si) __A,
+                   (__v8si) __O,
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastd_epi32 (__mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastd256_mask ((__v4si) __A,
+                   (__v8si)
+                   _mm256_setzero_si256 (),
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_broadcastq_epi64 (__m128i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastq128_mask ((__v2di) __A,
+                   (__v2di) __O,
+                   __M);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_broadcastq_epi64 (__mmask8 __M, __m128i __A)
+{
+  return (__m128i) __builtin_ia32_pbroadcastq128_mask ((__v2di) __A,
+                   (__v2di)
+                   _mm_setzero_si128 (),
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_broadcastq_epi64 (__m256i __O, __mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastq256_mask ((__v2di) __A,
+                   (__v4di) __O,
+                   __M);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_broadcastq_epi64 (__mmask8 __M, __m128i __A)
+{
+  return (__m256i) __builtin_ia32_pbroadcastq256_mask ((__v2di) __A,
+                   (__v4di)
+                   _mm256_setzero_si256 (),
+                   __M);
+}
+
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS_BOTH
 
 #endif /* __AVX512VLINTRIN_H */

test/CodeGen/avx512bw-builtins.c

 }
 
 __mmask32 test_mm512_mask_testn_epi16_mask(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_testn_epi16_mask
   // CHECK: @llvm.x86.avx512.ptestnm.w.
   return _mm512_mask_testn_epi16_mask(__U, __A, __B); 
 }
 
+__m512i test_mm512_broadcastb_epi8(__m128i __A) {
+  // CHECK-LABEL: @test_mm512_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.512
+  return _mm512_broadcastb_epi8(__A); 
+}
+
+__m512i test_mm512_mask_broadcastb_epi8(__m512i __O, __mmask64 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.512
+  return _mm512_mask_broadcastb_epi8(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcastb_epi8(__mmask64 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.512
+  return _mm512_maskz_broadcastb_epi8(__M, __A); 
+}
+
+__m512i test_mm512_broadcastw_epi16(__m128i __A) {
+  // CHECK-LABEL: @test_mm512_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.512
+  return _mm512_broadcastw_epi16(__A); 
+}
+
+__m512i test_mm512_mask_broadcastw_epi16(__m512i __O, __mmask32 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.512
+  return _mm512_mask_broadcastw_epi16(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcastw_epi16(__mmask32 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.512
+  return _mm512_maskz_broadcastw_epi16(__M, __A); 
+}
+
+__m128i test_mm512_mask_set1_epi16(__m512i __O, __mmask32 __M, short __A) {
+  // CHECK-LABEL: @test_mm_mask_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.512
+  return _mm512_mask_set1_epi16(__O, __M, __A); 
+}
+
+__m128i test_mm512_mask_set1_epi16(__mmask32 __M, short __A) {
+  // CHECK-LABEL: @test_mm_mask_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.512
+  return _mm512_maskz_set1_epi16(__M, __A); 
+}

test/CodeGen/avx512dq-builtins.c

 }
 
 __m512 test_mm512_maskz_reduce_round_ps(__mmask16 __U, __m512 __A) {
   // CHECK-LABEL: @test_mm512_maskz_reduce_round_ps
   // CHECK: @llvm.x86.avx512.mask.reduce.ps.512
   return _mm512_maskz_reduce_round_ps(__U, __A, 4, 8); 
 }
 
+__m512 test_mm512_broadcast_f32x2(__m128 __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm512_broadcast_f32x2(__A); 
+}
+
+__m512 test_mm512_mask_broadcast_f32x2(__m512 __O, __mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm512_mask_broadcast_f32x2(__O, __M, __A); 
+}
+
+__m512 test_mm512_maskz_broadcast_f32x2(__mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm512_maskz_broadcast_f32x2(__M, __A); 
+}
+
+__m512 test_mm512_broadcast_f32x8(__m256 __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_f32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
+  return _mm512_broadcast_f32x8(__A); 
+}
+
+__m512 test_mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_f32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
+  return _mm512_mask_broadcast_f32x8(__O, __M, __A); 
+}
+
+__m512 test_mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_f32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
+  return _mm512_maskz_broadcast_f32x8(__M, __A); 
+}
+
+__m512d test_mm512_broadcast_f64x2(__m128d __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm512_broadcast_f64x2(__A); 
+}
+
+__m512d test_mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm512_mask_broadcast_f64x2(__O, __M, __A); 
+}
+
+__m512d test_mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm512_maskz_broadcast_f64x2(__M, __A); 
+}
+
+__m512i test_mm512_broadcast_i32x2(__m128i __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm512_broadcast_i32x2(__A); 
+}
+
+__m512i test_mm512_mask_broadcast_i32x2(__m512i __O, __mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm512_mask_broadcast_i32x2(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcast_i32x2(__mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm512_maskz_broadcast_i32x2(__M, __A); 
+}
+
+__m512i test_mm512_broadcast_i32x8(__m256i __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_i32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
+  return _mm512_broadcast_i32x8(__A); 
+}
+
+__m512i test_mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_i32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
+  return _mm512_mask_broadcast_i32x8(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_i32x8
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
+  return _mm512_maskz_broadcast_i32x8(__M, __A); 
+}
+
+__m512i test_mm512_broadcast_i64x2(__m128i __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm512_broadcast_i64x2(__A); 
+}
+
+__m512i test_mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm512_mask_broadcast_i64x2(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm512_maskz_broadcast_i64x2(__M, __A); 
+}

test/CodeGen/avx512vl-builtins.c

__m256 test_mm256_mask_rsqrt14_ps(__m256 __W, __mmask8 __U, __m256 __A) {
   return _mm256_mask_rsqrt14_ps(__W, __U, __A); 
 }
 
 __m256 test_mm256_maskz_rsqrt14_ps(__mmask8 __U, __m256 __A) {
   // CHECK-LABEL: @test_mm256_maskz_rsqrt14_ps
   // CHECK: @llvm.x86.avx512.rsqrt14.ps.256
   return _mm256_maskz_rsqrt14_ps(__U, __A); 
 }
+
+__m256 test_mm256_broadcast_f32x4(__m128 __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm256_broadcast_f32x4(__A); 
+}
+
+__m256 test_mm256_mask_broadcast_f32x4(__m256 __O, __mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm256_mask_broadcast_f32x4(__O, __M, __A); 
+}
+
+__m256 test_mm256_maskz_broadcast_f32x4(__mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm256_maskz_broadcast_f32x4(__M, __A); 
+}
+
+__m256i test_mm256_broadcast_i32x4(__m128i __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm256_broadcast_i32x4(__A); 
+}
+
+__m256i test_mm256_mask_broadcast_i32x4(__m256i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm256_mask_broadcast_i32x4(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcast_i32x4(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm256_maskz_broadcast_i32x4(__M, __A); 
+}
+
+__m256d test_mm256_mask_broadcastsd_pd(__m256d __O, __mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastsd_pd
+  // CHECK: @llvm.x86.avx512.mask.broadcast.sd.pd.256
+  return _mm256_mask_broadcastsd_pd(__O, __M, __A); 
+}
+
+__m256d test_mm256_maskz_broadcastsd_pd(__mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastsd_pd
+  // CHECK: @llvm.x86.avx512.mask.broadcast.sd.pd.256
+  return _mm256_maskz_broadcastsd_pd(__M, __A); 
+}
+
+__m128 test_mm_mask_broadcastss_ps(__m128 __O, __mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.128
+  return _mm_mask_broadcastss_ps(__O, __M, __A); 
+}
+
+__m128 test_mm_maskz_broadcastss_ps(__mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.128
+  return _mm_maskz_broadcastss_ps(__M, __A); 
+}
+
+__m256 test_mm256_mask_broadcastss_ps(__m256 __O, __mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.256
+  return _mm256_mask_broadcastss_ps(__O, __M, __A); 
+}
+
+__m256 test_mm256_maskz_broadcastss_ps(__mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.256
+  return _mm256_maskz_broadcastss_ps(__M, __A); 
+}
+
+__m128i test_mm_mask_broadcastd_epi32(__m128i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcastd_epi32
+  // CHECK: @llvm.x86.avx512.pbroadcastd.128
+  return _mm_mask_broadcastd_epi32(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_broadcastd_epi32(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcastd_epi32
+  // CHECK: @llvm.x86.avx512.pbroadcastd.128
+  return _mm_maskz_broadcastd_epi32(__M, __A); 
+}
+
+__m256i test_mm256_mask_broadcastd_epi32(__m256i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastd_epi32
+  // CHECK: @llvm.x86.avx512.pbroadcastd.256
+  return _mm256_mask_broadcastd_epi32(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcastd_epi32(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastd_epi32
+  // CHECK: @llvm.x86.avx512.pbroadcastd.256
+  return _mm256_maskz_broadcastd_epi32(__M, __A); 
+}
+
+__m128i test_mm_mask_broadcastq_epi64(__m128i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcastq_epi64
+  // CHECK: @llvm.x86.avx512.pbroadcastq.128
+  return _mm_mask_broadcastq_epi64(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_broadcastq_epi64(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcastq_epi64
+  // CHECK: @llvm.x86.avx512.pbroadcastq.128
+  return _mm_maskz_broadcastq_epi64(__M, __A); 
+}
+
+__m256i test_mm256_mask_broadcastq_epi64(__m256i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastq_epi64
+  // CHECK: @llvm.x86.avx512.pbroadcastq.256
+  return _mm256_mask_broadcastq_epi64(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcastq_epi64(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastq_epi64
+  // CHECK: @llvm.x86.avx512.pbroadcastq.256
+  return _mm256_maskz_broadcastq_epi64(__M, __A); 
+}

test/CodeGen/avx512vlbw-builtins.c

 }
 
 __mmask16 test_mm256_mask_testn_epi16_mask(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_testn_epi16_mask
   // CHECK: @llvm.x86.avx512.ptestnm.w.256
   return _mm256_mask_testn_epi16_mask(__U, __A, __B); 
 }
 
+__m128i test_mm_mask_broadcastb_epi8(__m128i __O, __mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.128
+  return _mm_mask_broadcastb_epi8(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_broadcastb_epi8(__mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.128
+  return _mm_maskz_broadcastb_epi8(__M, __A); 
+}
+
+__m256i test_mm256_mask_broadcastb_epi8(__m256i __O, __mmask32 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.256
+  return _mm256_mask_broadcastb_epi8(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcastb_epi8(__mmask32 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastb_epi8
+  // CHECK: @llvm.x86.avx512.pbroadcastb.256
+  return _mm256_maskz_broadcastb_epi8(__M, __A); 
+}
+
+__m128i test_mm_mask_broadcastw_epi16(__m128i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.128
+  return _mm_mask_broadcastw_epi16(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_broadcastw_epi16(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.128
+  return _mm_maskz_broadcastw_epi16(__M, __A); 
+}
+
+__m256i test_mm256_mask_broadcastw_epi16(__m256i __O, __mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.256
+  return _mm256_mask_broadcastw_epi16(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcastw_epi16(__mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcastw_epi16
+  // CHECK: @llvm.x86.avx512.pbroadcastw.256
+  return _mm256_maskz_broadcastw_epi16(__M, __A); 
+}
+
+__m256i test_mm256_mask_set1_epi16(__m256i __O, __mmask16 __M, short __A) {
+  // CHECK-LABEL: @test_mm256_mask_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.256
+  return _mm256_mask_set1_epi16(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_set1_epi16(__mmask16 __M, short __A) {
+  // CHECK-LABEL: @test_mm256_maskz_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.256
+  return _mm256_maskz_set1_epi16(__M, __A); 
+}
+
+__m128i test_mm_mask_set1_epi16(__m128i __O, __mmask8 __M, short __A) {
+  // CHECK-LABEL: @test_mm_mask_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.128
+  return _mm_mask_set1_epi16(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_set1_epi16(__mmask8 __M, short __A) {
+  // CHECK-LABEL: @test_mm_maskz_set1_epi16
+  // CHECK: @llvm.x86.avx512.mask.pbroadcast.w.gpr.128
+  return _mm_maskz_set1_epi16(__M, __A); 
+}

test/CodeGen/avx512vldq-builtins.c

__m256 test_mm256_mask_reduce_ps(__m256 __W, __mmask8 __U, __m256 __A) {
   return _mm256_mask_reduce_ps(__W, __U, __A, 4); 
 }
 
 __m256 test_mm256_maskz_reduce_ps(__mmask8 __U, __m256 __A) {
   // CHECK-LABEL: @test_mm256_maskz_reduce_ps
   // CHECK: @llvm.x86.avx512.mask.reduce.ps.256
   return _mm256_maskz_reduce_ps(__U, __A, 4); 
 }
+
+__m256 test_mm256_broadcast_f32x2(__m128 __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm256_broadcast_f32x2(__A); 
+}
+
+__m256 test_mm256_mask_broadcast_f32x2(__m256 __O, __mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm256_mask_broadcast_f32x2(__O, __M, __A); 
+}
+
+__m256 test_mm256_maskz_broadcast_f32x2(__mmask8 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_f32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x2
+  return _mm256_maskz_broadcast_f32x2(__M, __A); 
+}
+
+__m256d test_mm256_broadcast_f64x2(__m128d __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm256_broadcast_f64x2(__A); 
+}
+
+__m256d test_mm256_mask_broadcast_f64x2(__m256d __O, __mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm256_mask_broadcast_f64x2(__O, __M, __A); 
+}
+
+__m256d test_mm256_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_f64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
+  return _mm256_maskz_broadcast_f64x2(__M, __A); 
+}
+
+__m128i test_mm_broadcast_i32x2(__m128i __A) {
+  // CHECK-LABEL: @test_mm_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm_broadcast_i32x2(__A); 
+}
+
+__m128i test_mm_mask_broadcast_i32x2(__m128i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm_mask_broadcast_i32x2(__O, __M, __A); 
+}
+
+__m128i test_mm_maskz_broadcast_i32x2(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm_maskz_broadcast_i32x2(__M, __A); 
+}
+
+__m256i test_mm256_broadcast_i32x2(__m128i __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm256_broadcast_i32x2(__A); 
+}
+
+__m256i test_mm256_mask_broadcast_i32x2(__m256i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm256_mask_broadcast_i32x2(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcast_i32x2(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_i32x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x2
+  return _mm256_maskz_broadcast_i32x2(__M, __A); 
+}
+
+__m256i test_mm256_broadcast_i64x2(__m128i __A) {
+  // CHECK-LABEL: @test_mm256_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm256_broadcast_i64x2(__A); 
+}
+
+__m256i test_mm256_mask_broadcast_i64x2(__m256i __O, __mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_mask_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm256_mask_broadcast_i64x2(__O, __M, __A); 
+}
+
+__m256i test_mm256_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_broadcast_i64x2
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
+  return _mm256_maskz_broadcast_i64x2(__M, __A); 
+}