[Clang][AVX512][Builtin] adding intrinsics for vf{n}madd{ss|sd} and vf{n}sub{ss|sd} instruction set

[igorb]

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_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

lib/Headers/avx512fintrin.h

 
 #define _mm_maskz_sqrt_round_ss( __U, __A, __B, __R) __extension__ ({ \
 __builtin_ia32_sqrtss_round_mask ((__v4sf)( __B),\
                  (__v4sf)( __A),(__v4sf) _mm_setzero_ps(),\
                  (__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
 
 #endif // __AVX512FINTRIN_H

test/CodeGen/avx512f-builtins.c

// 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: @llvm.x86.avx512.mask.sqrt.ss
     return _mm_maskz_sqrt_round_ss(__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);
+}