Index: include/clang/Basic/BuiltinsX86.def
===================================================================
--- include/clang/Basic/BuiltinsX86.def
+++ include/clang/Basic/BuiltinsX86.def
@@ -1794,6 +1794,14 @@
 TARGET_BUILTIN(__builtin_ia32_loadupd256_mask, "V4dV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups128_mask, "V4fV4f*V4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups256_mask, "V8fV8f*V8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_sqrtsd_round_mask, "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_sqrtss_round_mask, "V4fV4fV4fV4fUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddss3_mask,  "V4fV4fV4fV4fUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddss3_maskz, "V4fV4fV4fV4fUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddss3_mask3, "V4fV4fV4fV4fUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_mask,  "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_maskz, "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_mask3, "V2dV2dV2dV2dUcIi", "", "avx512f")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN
Index: lib/Headers/avx512fintrin.h
===================================================================
--- lib/Headers/avx512fintrin.h
+++ lib/Headers/avx512fintrin.h
@@ -931,6 +931,24 @@
              (__mmask8) -1);
 }
 
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_rsqrt14_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_rsqrt14ss_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_rsqrt14_ss (__mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_rsqrt14ss_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) _mm_setzero_ps (),
+          (__mmask8) __U);
+}
+
 static  __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_rsqrt14_sd(__m128d __A, __m128d __B)
 {
@@ -941,6 +959,24 @@
               (__mmask8) -1);
 }
 
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_rsqrt14_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_rsqrt14sd_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_rsqrt14_sd (__mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_rsqrt14sd_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) _mm_setzero_pd (),
+          (__mmask8) __U);
+}
+
 static  __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_rcp14_pd(__m512d __A)
 {
@@ -968,6 +1004,24 @@
                  (__mmask8) -1);
 }
 
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_rcp14_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_rcp14ss_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_rcp14_ss (__mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_rcp14ss_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) _mm_setzero_ps (),
+          (__mmask8) __U);
+}
+
 static  __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_rcp14_sd(__m128d __A, __m128d __B)
 {
@@ -978,6 +1032,24 @@
             (__mmask8) -1);
 }
 
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_rcp14_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_rcp14sd_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_rcp14_sd (__mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_rcp14sd_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) _mm_setzero_pd (),
+          (__mmask8) __U);
+}
+
 static __inline __m512 __DEFAULT_FN_ATTRS
 _mm512_floor_ps(__m512 __A)
 {
@@ -3918,6 +3990,42 @@
                  (__v2df) __B, (__v2df) _mm_setzero_pd(), (__mmask8) -1, _MM_FROUND_CUR_DIRECTION);
 }
 
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_getexp_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_getexpsd128_round_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_getexp_round_sd( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_getexpsd128_round_mask ((__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_getexp_sd (__mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_getexpsd128_round_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) _mm_setzero_pd (),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_getexp_round_sd( __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_getexpsd128_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) _mm_setzero_pd (),\
+          (__mmask8) __U,\
+          __R);\
+})
+
 #define _mm_getexp_round_ss( __A, __B, __R) __extension__ ({ \
 __builtin_ia32_getexpss128_round_mask ((__v4sf)( __A),\
                 (__v4sf)( __B), (__v4sf)  _mm_setzero_ps(), (__mmask8) -1,\
@@ -3931,6 +4039,42 @@
                 (__v4sf) __B, (__v4sf)  _mm_setzero_ps(), (__mmask8) -1, _MM_FROUND_CUR_DIRECTION);
 }
 
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_getexp_ss (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_getexpss128_round_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_getexp_round_ss( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_getexpss128_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_getexp_ss (__mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_getexpss128_round_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) _mm_setzero_pd (),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_getexp_round_ss( __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_getexpss128_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (__v4sf) _mm_setzero_ps (),\
+          (__mmask8) __U,\
+          __R);\
+})
+
 #define _mm_getmant_round_sd( __A, __B, __C, __D, __R) __extension__ ({ \
 __builtin_ia32_getmantsd_round_mask ((__v2df)( __A),\
               (__v2df)( __B),\
@@ -3945,6 +4089,42 @@
               _MM_FROUND_CUR_DIRECTION);\
 })
 
+#define _mm_mask_getmant_sd( __W, __U, __A, __B, __C, __D) __extension__ ({\
+__builtin_ia32_getmantsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (( __D) << 2) |( __C),\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_mask_getmant_round_sd( __W, __U, __A, __B, __C, __D, __R)({\
+__builtin_ia32_getmantsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (( __D) << 2) |( __C),\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+#define _mm_maskz_getmant_sd( __U, __A, __B, __C, __D) __extension__ ({\
+__builtin_ia32_getmantsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (( __D) << 2) |( __C),\
+          (__v2df) _mm_setzero_pd (),\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_maskz_getmant_round_sd( __U, __A, __B, __C, __D, __R) __extension__ ({\
+__builtin_ia32_getmantsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (( __D) << 2) |( __C),\
+          (__v2df) _mm_setzero_pd (),\
+          (__mmask8) __U,\
+          __R);\
+})
+
 #define _mm_getmant_round_ss( __A, __B, __C, __D, __R) __extension__ ({ \
 __builtin_ia32_getmantss_round_mask ((__v4sf)( __A),\
               (__v4sf)( __B),\
@@ -3959,6 +4139,41 @@
               _MM_FROUND_CUR_DIRECTION);\
 })
 
+#define _mm_mask_getmant_ss( __W, __U, __A, __B, __C, __D) __extension__ ({\
+__builtin_ia32_getmantss_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (( __D) << 2) |( __C),\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_mask_getmant_round_ss( __W, __U, __A, __B, __C, __D, __R)({\
+__builtin_ia32_getmantss_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (( __D) << 2) |( __C),\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+#define _mm_maskz_getmant_ss( __U, __A, __B, __C, __D) __extension__ ({\
+__builtin_ia32_getmantss_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (( __D) << 2) |( __C),\
+          (__v4sf) _mm_setzero_pd (),\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_maskz_getmant_round_ss( __U, __A, __B, __C, __D, __R) __extension__ ({\
+__builtin_ia32_getmantss_round_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (( __D) << 2) |( __C),\
+          (__v4sf) _mm_setzero_ps (),\
+          (__mmask8) __U,\
+          __R);\
+})
 
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS
 _mm512_kmov (__mmask16 __A)
@@ -3974,6 +4189,524 @@
 __builtin_ia32_vcomiss ((__v4sf) (__A), (__v4sf) (__B), ( __P), ( __R));\
 })
 
+#define _mm_sqrt_round_ss( __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtss_round_mask ((__v4sf) __A,\
+        (__v4sf) __B,\
+        (__v4sf) _mm_setzero_ps (),\
+        (__mmask8) -1,\
+        __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_sqrt_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_sqrtss_round_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_sqrt_round_ss( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtss_round_mask ((__v4sf) __A,\
+        (__v4sf) __B,\
+        (__v4sf) __W,\
+        (__mmask8) __U,\
+        __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_sqrt_ss (__mmask8 __U, __m128 __A, __m128 __B) 
+{
+return (__m128) __builtin_ia32_sqrtss_round_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) _mm_setzero_ps(),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_sqrt_round_ss( __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtss_round_mask ((__v4sf) __A,\
+        (__v4sf) __B,\
+        (__v4sf) _mm_setzero_ps(),\
+        (__mmask8) __U,\
+        __R);\
+})
+
+#define _mm_sqrt_round_sd( __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtsd_round_mask ((__v2df) __A,\
+        (__v2df) __B,\
+        (__v2df) _mm_setzero_pd (),\
+        (__mmask8) -1,\
+        __R);\
+})
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_sqrt_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_sqrtsd_round_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_sqrt_round_sd( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_sqrt_sd (__mmask8 __U, __m128d __A, __m128d __B) 
+{
+ return (__m128d) __builtin_ia32_sqrtsd_round_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) _mm_setzero_pd (),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_sqrt_round_sd( __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_sqrtsd_round_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) _mm_setzero_pd (),\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fmadd_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fmadd_round_ss( __W, __U, __A, __B, __R) __extension__({\
+__builtin_ia32_vfmaddss3_mask ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fmadd_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_maskz ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fmadd_round_ss( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_maskz ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (__v4sf) __C,\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fmadd_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,
+          (__v4sf) __X,
+          (__v4sf) __Y,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fmadd_round_ss( __W, __X, __Y, __U, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,\
+          (__v4sf) __X,\
+          (__v4sf) __Y,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fmsub_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask ((__v4sf) __A,
+          (__v4sf) (-(__B)),
+          (__v4sf) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fmsub_round_ss( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_mask ((__v4sf) __A,\
+          (__v4sf) (-(__B)),\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fmsub_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_maskz ((__v4sf) __A,
+          (__v4sf) __B,
+          (__v4sf) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fmsub_round_ss( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_maskz ((__v4sf) __A,\
+          (__v4sf) __B,\
+          (__v4sf) __C,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fmsub_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,
+          (__v4sf) __X,
+          (__v4sf) (-__Y),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fmsub_round_ss( __W, __X, __Y, __U, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,\
+          (__v4sf) __X,\
+          (__v4sf) - (__Y),\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fnmadd_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask ((__v4sf) -(__A),
+          (__v4sf) __B,
+          (__v4sf) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fnmadd_round_ss( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_mask ((__v4sf) -(__A),\
+          (__v4sf) __B,\
+          (__v4sf) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fnmadd_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_maskz ((__v4sf) -__A,
+          (__v4sf) __B,
+          (__v4sf) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fnmadd_round_ss( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_maskz ((__v4sf) -(__A),\
+          (__v4sf) __B,\
+          (__v4sf) __C,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fnmadd_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask3 ((__v4sf) -(__W),
+          (__v4sf) __X,
+          (__v4sf) __Y,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fnmadd_round_ss( __W,  __X, __Y, __U, __R) __extension__({\
+__builtin_ia32_vfmaddss3_mask3 ((__v4sf) -(__W),\
+          (__v4sf) __X,\
+          (__v4sf) __Y,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fnmsub_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask ((__v4sf) -(__A),
+          (__v4sf) __B,
+          (__v4sf) -(__W),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fnmsub_round_ss( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_mask ((__v4sf) -(__A),\
+          (__v4sf) __B,\
+          (__v4sf) -(__W),\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fnmsub_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_maskz ((__v4sf) -(__A),
+          (__v4sf) __B,
+          (__v4sf) -(__C),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fnmsub_round_ss( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddss3_maskz((__v4sf) -(__A),\
+          (__v4sf) __B,\
+          (__v4sf) -(__C),\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fnmsub_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,
+          (__v4sf) __X,
+          (__v4sf) __Y,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fnmsub_round_ss( __W, __X, __Y, __U, __R) __extension__({\
+__builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,\
+          (__v4sf) __X,\
+          (__v4sf) __Y,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fmadd_sd (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fmadd_round_sd( __W, __U, __A, __B, __R) __extension__({\
+__builtin_ia32_vfmaddsd3_mask ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fmadd_sd (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_maskz ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fmadd_round_sd( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_maskz ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) __C,\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fmadd_sd (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask3 ((__v2df) __W,
+          (__v2df) __X,
+          (__v2df) __Y,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fmadd_round_sd( __W, __X, __Y, __U, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_mask3 ((__v2df) __W,\
+          (__v2df) __X,\
+          (__v2df) __Y,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fmsub_sd (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask ( (__v2df) __A,
+          (__v2df) (-(__B)),
+          (__v2df) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fmsub_round_sd( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_mask ( (__v2df) __A,\
+          (__v2df) (-(__B)),\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fmsub_sd (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_maskz ( (__v2df) __A,
+          (__v2df) __B,
+          (__v2df) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fmsub_round_sd( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_maskz ( (__v2df) __A,\
+          (__v2df) __B,\
+          (__v2df) __C,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fmsub_sd (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask3 ((__v2df) __W,
+          (__v2df) __X,
+          (__v2df) (-__Y),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fmsub_round_sd( __W, __X, __Y, __U, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_mask3 ((__v2df) __W,\
+          (__v2df) __X,\
+          (__v2df) -(__Y),\
+          (__mmask8) __U, __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fnmadd_sd (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask ( (__v2df) -(__A),
+          (__v2df) __B,
+          (__v2df) __W,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fnmadd_round_sd( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_mask ( (__v2df) -(__A),\
+          (__v2df) __B,\
+          (__v2df) __W,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fnmadd_sd (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_maskz ( (__v2df) -__A,
+          (__v2df) __B,
+          (__v2df) __C,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fnmadd_round_sd( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_maskz ( (__v2df) -(__A),\
+          (__v2df) __B,\
+          (__v2df) __C,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fnmadd_sd (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask3 ((__v2df) -(__W),
+          (__v2df) __X,
+          (__v2df) __Y,
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fnmadd_round_sd( __W,  __X, __Y, __U, __R) __extension__({\
+__builtin_ia32_vfmaddsd3_mask3 ((__v2df) -(__W),\
+          (__v2df) __X,\
+          (__v2df) __Y,\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_fnmsub_sd (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask ( (__v2df) -(__A),
+          (__v2df) __B,
+          (__v2df) -(__W),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask_fnmsub_round_sd( __W, __U, __A, __B, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_mask ( (__v2df) -(__A),\
+          (__v2df) __B,\
+          (__v2df) -(__W),\
+          (__mmask8) __U,\
+          __R);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_fnmsub_sd (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_maskz ( (__v2df) -(__A),
+          (__v2df) __B,
+          (__v2df) -(__C),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_maskz_fnmsub_round_sd( __U, __A, __B, __C, __R) __extension__ ({\
+__builtin_ia32_vfmaddsd3_maskz( (__v2df) -(__A),\
+          (__v2df) __B,\
+          (__v2df) -(__C),\
+          (__mmask8) __U,\
+          _MM_FROUND_CUR_DIRECTION);\
+})
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask3_fnmsub_sd (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U) 
+{
+ return (__m128) __builtin_ia32_vfmaddsd3_mask3 ((__v2df) -(__W),
+          (__v2df) __X,
+          (__v2df) -(__Y),
+          (__mmask8) __U,
+          _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_mask3_fnmsub_round_sd( __W, __X, __Y, __U, __R) __extension__({\
+__builtin_ia32_vfmaddsd3_mask3 ((__v2df) -(__W),\
+          (__v2df) __X,\
+          (__v2df) -(__Y),\
+          (__mmask8) __U,\
+          __R);\
+})
 
 #undef __DEFAULT_FN_ATTRS
 
Index: test/CodeGen/avx512f-builtins.c
===================================================================
--- test/CodeGen/avx512f-builtins.c
+++ test/CodeGen/avx512f-builtins.c
@@ -2576,5 +2576,494 @@
   return _mm512_kmov(__A); 
 }
 
+__m128 test_mm_sqrt_round_ss(__m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_sqrt_round_ss
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ss
+  return _mm_sqrt_round_ss(__A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_sqrt_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_sqrt_ss
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ss
+  return _mm_mask_sqrt_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_sqrt_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_sqrt_round_ss
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ss
+  return _mm_mask_sqrt_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_sqrt_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_sqrt_ss
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ss
+  return _mm_maskz_sqrt_ss(__U, __A, __B);
+}
+
+__m128 test_mm_maskz_sqrt_round_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_sqrt_round_ss
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ss
+  return _mm_maskz_sqrt_round_ss(__U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
 
+__m128d test_mm_sqrt_round_sd(__m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_sqrt_round_sd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.sd
+  return _mm_sqrt_round_sd(__A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_sqrt_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_sqrt_sd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.sd
+  return _mm_mask_sqrt_sd(__W, __U, __A, __B);
+}
 
+__m128d test_mm_mask_sqrt_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_sqrt_round_sd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.sd
+  return _mm_mask_sqrt_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_sqrt_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_sqrt_sd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.sd
+  return _mm_maskz_sqrt_sd(__U, __A, __B);
+}
+
+__m128d test_mm_maskz_sqrt_round_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_sqrt_round_sd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.sd
+  return _mm_maskz_sqrt_round_sd(__U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_fmadd_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fmadd_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fmadd_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_fmadd_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fmadd_round_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fmadd_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_fmadd_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fmadd_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fmadd_ss(__U, __A, __B, __C);
+}
+
+__m128 test_mm_maskz_fmadd_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fmadd_round_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fmadd_round_ss(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask3_fmadd_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmadd_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fmadd_ss(__W, __X, __Y, __U);
+}
+
+__m128 test_mm_mask3_fmadd_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmadd_round_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fmadd_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_fmsub_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fmsub_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fmsub_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_fmsub_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fmsub_round_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fmsub_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_fmsub_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fmsub_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fmsub_ss(__U, __A, __B, __C);
+}
+
+__m128 test_mm_maskz_fmsub_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fmsub_round_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fmsub_round_ss(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask3_fmsub_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmsub_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fmsub_ss(__W, __X, __Y, __U);
+}
+
+__m128 test_mm_mask3_fmsub_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmsub_round_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fmsub_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_fnmadd_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fnmadd_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fnmadd_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_fnmadd_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fnmadd_round_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fnmadd_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_fnmadd_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmadd_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fnmadd_ss(__U, __A, __B, __C);
+}
+
+__m128 test_mm_maskz_fnmadd_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmadd_round_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fnmadd_round_ss(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask3_fnmadd_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmadd_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fnmadd_ss(__W, __X, __Y, __U);
+}
+
+__m128 test_mm_mask3_fnmadd_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmadd_round_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fnmadd_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_fnmsub_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fnmsub_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fnmsub_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_fnmsub_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_fnmsub_round_ss
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.ss
+  return _mm_mask_fnmsub_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_fnmsub_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmsub_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fnmsub_ss(__U, __A, __B, __C);
+}
+
+__m128 test_mm_maskz_fnmsub_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128 __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmsub_round_ss
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.ss
+  return _mm_maskz_fnmsub_round_ss(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask3_fnmsub_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmsub_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fnmsub_ss(__W, __X, __Y, __U);
+}
+
+__m128 test_mm_mask3_fnmsub_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmsub_round_ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  return _mm_mask3_fnmsub_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_fmadd_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fmadd_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fmadd_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_mask_fmadd_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fmadd_round_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fmadd_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_fmadd_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fmadd_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fmadd_sd(__U, __A, __B, __C);
+}
+
+__m128d test_mm_maskz_fmadd_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fmadd_round_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fmadd_round_sd(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask3_fmadd_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmadd_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fmadd_sd(__W, __X, __Y, __U);
+}
+
+__m128d test_mm_mask3_fmadd_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmadd_round_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fmadd_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_fmsub_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fmsub_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fmsub_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_mask_fmsub_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fmsub_round_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fmsub_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_fmsub_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fmsub_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fmsub_sd(__U, __A, __B, __C);
+}
+
+__m128d test_mm_maskz_fmsub_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fmsub_round_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fmsub_round_sd(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask3_fmsub_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmsub_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fmsub_sd(__W, __X, __Y, __U);
+}
+
+__m128d test_mm_mask3_fmsub_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fmsub_round_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fmsub_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_fnmadd_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fnmadd_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fnmadd_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_mask_fnmadd_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fnmadd_round_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fnmadd_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_fnmadd_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmadd_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fnmadd_sd(__U, __A, __B, __C);
+}
+
+__m128d test_mm_maskz_fnmadd_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmadd_round_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fnmadd_round_sd(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask3_fnmadd_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmadd_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fnmadd_sd(__W, __X, __Y, __U);
+}
+
+__m128d test_mm_mask3_fnmadd_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmadd_round_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fnmadd_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_fnmsub_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fnmsub_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fnmsub_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_mask_fnmsub_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_fnmsub_round_sd
+  // CHECK: @llvm.x86.avx512.mask.vfmadd.sd
+  return _mm_mask_fnmsub_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_fnmsub_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmsub_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fnmsub_sd(__U, __A, __B, __C);
+}
+
+__m128d test_mm_maskz_fnmsub_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __m128d __C){
+  // CHECK-LABEL: @test_mm_maskz_fnmsub_round_sd
+  // CHECK: @llvm.x86.avx512.maskz.vfmadd.sd
+  return _mm_maskz_fnmsub_round_sd(__U, __A, __B, __C, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask3_fnmsub_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmsub_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fnmsub_sd(__W, __X, __Y, __U);
+}
+
+__m128d test_mm_mask3_fnmsub_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
+  // CHECK-LABEL: @test_mm_mask3_fnmsub_round_sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  return _mm_mask3_fnmsub_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_rsqrt14_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_rsqrt14_sd
+  // CHECK: @llvm.x86.avx512.rsqrt14.sd
+  return _mm_mask_rsqrt14_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_maskz_rsqrt14_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_rsqrt14_sd
+  // CHECK: @llvm.x86.avx512.rsqrt14.sd
+  return _mm_maskz_rsqrt14_sd(__U, __A, __B);
+}
+
+__m128 test_mm_mask_rsqrt14_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_rsqrt14_ss
+  // CHECK: @llvm.x86.avx512.rsqrt14.ss
+  return _mm_mask_rsqrt14_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_maskz_rsqrt14_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_rsqrt14_ss
+  // CHECK: @llvm.x86.avx512.rsqrt14.ss
+  return _mm_maskz_rsqrt14_ss(__U, __A, __B);
+}
+
+__m128d test_mm_mask_rcp14_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_rcp14_sd
+  // CHECK: @llvm.x86.avx512.rcp14.sd
+  return _mm_mask_rcp14_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_maskz_rcp14_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_rcp14_sd
+  // CHECK: @llvm.x86.avx512.rcp14.sd
+  return _mm_maskz_rcp14_sd(__U, __A, __B);
+}
+
+__m128 test_mm_mask_rcp14_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_rcp14_ss
+  // CHECK: @llvm.x86.avx512.rcp14.ss
+  return _mm_mask_rcp14_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_maskz_rcp14_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_rcp14_ss
+  // CHECK: @llvm.x86.avx512.rcp14.ss
+  return _mm_maskz_rcp14_ss(__U, __A, __B);
+}
+
+__m128d test_mm_mask_getexp_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_getexp_sd
+  // CHECK: @llvm.x86.avx512.mask.getexp.sd
+  return _mm_mask_getexp_sd(__W, __U, __A, __B);
+}
+
+__m128d test_mm_mask_getexp_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_getexp_round_sd
+  // CHECK: @llvm.x86.avx512.mask.getexp.sd
+  return _mm_mask_getexp_round_sd(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_getexp_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_getexp_sd
+  // CHECK: @llvm.x86.avx512.mask.getexp.sd
+  return _mm_maskz_getexp_sd(__U, __A, __B);
+}
+
+__m128d test_mm_maskz_getexp_round_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_getexp_round_sd
+  // CHECK: @llvm.x86.avx512.mask.getexp.sd
+  return _mm_maskz_getexp_round_sd(__U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_getexp_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_getexp_ss
+  // CHECK: @llvm.x86.avx512.mask.getexp.ss
+  return _mm_mask_getexp_ss(__W, __U, __A, __B);
+}
+
+__m128 test_mm_mask_getexp_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_getexp_round_ss
+  // CHECK: @llvm.x86.avx512.mask.getexp.ss
+  return _mm_mask_getexp_round_ss(__W, __U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_getexp_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_getexp_ss
+  // CHECK: @llvm.x86.avx512.mask.getexp.ss
+  return _mm_maskz_getexp_ss(__U, __A, __B);
+}
+
+__m128 test_mm_maskz_getexp_round_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_getexp_round_ss
+  // CHECK: @llvm.x86.avx512.mask.getexp.ss
+  return _mm_maskz_getexp_round_ss(__U, __A, __B, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_mask_getmant_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_getmant_sd
+  // CHECK: @llvm.x86.avx512.mask.getmant.sd
+  return _mm_mask_getmant_sd(__W, __U, __A, __B, 1, 2);
+}
+
+__m128d test_mm_mask_getmant_round_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_mask_getmant_round_sd
+  // CHECK: @llvm.x86.avx512.mask.getmant.sd
+  return _mm_mask_getmant_round_sd(__W, __U, __A, __B, 1, 2, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128d test_mm_maskz_getmant_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_getmant_sd
+  // CHECK: @llvm.x86.avx512.mask.getmant.sd
+  return _mm_maskz_getmant_sd(__U, __A, __B, 1, 2);
+}
+
+__m128d test_mm_maskz_getmant_round_sd(__mmask8 __U, __m128d __A, __m128d __B){
+  // CHECK-LABEL: @test_mm_maskz_getmant_round_sd
+  // CHECK: @llvm.x86.avx512.mask.getmant.sd
+  return _mm_maskz_getmant_round_sd(__U, __A, __B, 1, 2, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_mask_getmant_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_getmant_ss
+  // CHECK: @llvm.x86.avx512.mask.getmant.ss
+  return _mm_mask_getmant_ss(__W, __U, __A, __B, 1, 2);
+}
+
+__m128 test_mm_mask_getmant_round_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_mask_getmant_round_ss
+  // CHECK: @llvm.x86.avx512.mask.getmant.ss
+  return _mm_mask_getmant_round_ss(__W, __U, __A, __B, 1, 2, _MM_FROUND_CUR_DIRECTION);
+}
+
+__m128 test_mm_maskz_getmant_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_getmant_ss
+  // CHECK: @llvm.x86.avx512.mask.getmant.ss
+  return _mm_maskz_getmant_ss(__U, __A, __B, 1, 2);
+}
+
+__m128 test_mm_maskz_getmant_round_ss(__mmask8 __U, __m128 __A, __m128 __B){
+  // CHECK-LABEL: @test_mm_maskz_getmant_round_ss
+  // CHECK: @llvm.x86.avx512.mask.getmant.ss
+  return _mm_maskz_getmant_round_ss(__U, __A, __B, 1, 2, _MM_FROUND_CUR_DIRECTION);
+}