[Clang][Intrinsics][avx512] Adding round intrinsics fot max/min/sqrt instruction set to clang

cfe/trunk/lib/Headers/avx512fintrin.h

 {
   return (__m512i) __builtin_ia32_psubd512_mask ((__v16si) __A,
              (__v16si) __B,
              (__v16si)
              _mm512_setzero_si512 (),
              (__mmask16) __U);
 }
 
+#define _mm512_mask_max_round_pd( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)( __W),\
+            (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_max_round_pd( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)\
+            _mm512_setzero_pd (),\
+            (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_max_round_pd( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)\
+            _mm512_undefined_pd (),\
+            (__mmask8) -1,( __R));\
+})
+
 static  __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_max_pd(__m512d __A, __m512d __B)
 {
   return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
              (__v8df) __B,
              (__v8df)
              _mm512_setzero_pd (),
              (__mmask8) -1,
_mm512_maskz_max_pd (__mmask8 __U, __m512d __A, __m512d __B)
   return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
                   (__v8df) __B,
                   (__v8df)
                   _mm512_setzero_pd (),
                   (__mmask8) __U,
                   _MM_FROUND_CUR_DIRECTION);
 }
 
+#define _mm512_mask_max_round_ps( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)( __W),\
+           (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_max_round_ps( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)\
+           _mm512_setzero_ps (),\
+           (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_max_round_ps( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)\
+           _mm512_undefined_ps (),\
+           (__mmask16) -1,( __R));\
+})
+
 static  __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_max_ps(__m512 __A, __m512 __B)
 {
   return (__m512) __builtin_ia32_maxps512_mask ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf)
             _mm512_setzero_ps (),
             (__mmask16) -1,

 {
   return (__m512i) __builtin_ia32_pmaxuq512_mask ((__v8di) __A,
                    (__v8di) __B,
                    (__v8di)
                    _mm512_setzero_si512 (),
                    __M);
 }
 
+#define _mm512_mask_min_round_pd( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)( __W),\
+            (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_min_round_pd( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)\
+            _mm512_setzero_pd (),\
+            (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_min_round_pd( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+            (__v8df)( __B),\
+            (__v8df)\
+            _mm512_undefined_pd (),\
+            (__mmask8) -1,( __R));\
+})
+
 static  __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_min_pd(__m512d __A, __m512d __B)
 {
   return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
              (__v8df) __B,
              (__v8df)
              _mm512_setzero_pd (),
              (__mmask8) -1,
              _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_mask_min_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
 {
   return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
                   (__v8df) __B,
                   (__v8df) __W,
                   (__mmask8) __U,
                   _MM_FROUND_CUR_DIRECTION);
 }
 
+#define _mm512_mask_min_round_ps( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)( __W),\
+           (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_min_round_ps( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)\
+           _mm512_setzero_ps (),\
+           (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_min_round_ps( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+           (__v16sf)( __B),\
+           (__v16sf)\
+           _mm512_undefined_ps (),\
+           (__mmask16) -1,( __R));\
+})
+
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_maskz_min_pd (__mmask8 __U, __m512d __A, __m512d __B)
 {
   return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
                   (__v8df) __B,
                   (__v8df)
                   _mm512_setzero_pd (),
                   (__mmask8) __U,

 static __inline __m512i __DEFAULT_FN_ATTRS
 _mm512_mask_mullo_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
 {
   return (__m512i) __builtin_ia32_pmulld512_mask ((__v16si) __A,
               (__v16si) __B,
               (__v16si) __W, __M);
 }
 
+#define _mm512_mask_sqrt_round_pd( __W, __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+             (__v8df)( __W),\
+             (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_sqrt_round_pd( __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+             (__v8df)\
+             _mm512_setzero_pd (),\
+             (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_sqrt_round_pd( __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+             (__v8df)\
+             _mm512_undefined_pd (),\
+             (__mmask8) -1,( __R));\
+})
+
 static  __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_sqrt_pd(__m512d __a)
 {
   return (__m512d)__builtin_ia32_sqrtpd512_mask((__v8df)__a,
                                                 (__v8df) _mm512_setzero_pd (),
                                                 (__mmask8) -1,
                                                 _MM_FROUND_CUR_DIRECTION);
 }

 {
   return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
                    (__v8df)
                    _mm512_setzero_pd (),
                    (__mmask8) __U,
                    _MM_FROUND_CUR_DIRECTION);
 }
 
+#define _mm512_mask_sqrt_round_ps( __W, __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+            (__v16sf)( __W),\
+            (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_sqrt_round_ps( __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+            (__v16sf)\
+            _mm512_setzero_ps (),\
+            (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_sqrt_round_ps( __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+            (__v16sf)\
+            _mm512_undefined_ps (),\
+            (__mmask16) -1,( __R));\
+})
+
 static  __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_sqrt_ps(__m512 __a)
 {
   return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)__a,
                                                (__v16sf) _mm512_setzero_ps (),
                                                (__mmask16) -1,
                                                _MM_FROUND_CUR_DIRECTION);
 }

 _mm512_mask_floor_ps (__m512 __W, __mmask16 __U, __m512 __A)
 {
   return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
                    _MM_FROUND_FLOOR,
                    (__v16sf) __W, __U,
                    _MM_FROUND_CUR_DIRECTION);
 }
 
-
 static __inline __m512d __DEFAULT_FN_ATTRS
 _mm512_floor_pd(__m512d __A)
 {
   return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
                                                    _MM_FROUND_FLOOR,
                                                    (__v8df) __A, -1,
                                                    _MM_FROUND_CUR_DIRECTION);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_mask_floor_pd (__m512d __W, __mmask8 __U, __m512d __A)
 {
   return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
                 _MM_FROUND_FLOOR,
                 (__v8df) __W, __U,
                 _MM_FROUND_CUR_DIRECTION);
 }
 
-
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_mask_ceil_ps (__m512 __W, __mmask16 __U, __m512 __A)
 {
   return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
                    _MM_FROUND_CEIL,
                    (__v16sf) __W, __U,
                    _MM_FROUND_CUR_DIRECTION);
 }
#define _mm512_mask_cvtps_ph(U, W, A, I) __extension__ ({ \
   (__m256i)__builtin_ia32_vcvtps2ph512_mask((__v16sf)(__m512)(A), (int)(I), \
                                             (__v16hi)(__m256i)(U), \
                                             (__mmask16)(W)); })
 
 #define _mm512_maskz_cvtps_ph(W, A, I) __extension__ ({\
   (__m256i)__builtin_ia32_vcvtps2ph512_mask((__v16sf)(__m512)(A), (int)(I), \
                                             (__v16hi)_mm256_setzero_si256(), \
                                             (__mmask16)(W)); })
                    
-
 static  __inline __m512 __DEFAULT_FN_ATTRS
 _mm512_cvtph_ps(__m256i __A)
 {
   return (__m512) __builtin_ia32_vcvtph2ps512_mask ((__v16hi) __A,
                 (__v16sf)
                 _mm512_setzero_ps (),
                 (__mmask16) -1,
                 _MM_FROUND_CUR_DIRECTION);

cfe/trunk/test/CodeGen/avx512f-builtins.c

 
 __m512d test_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
 {
   // CHECK-LABEL: @test_mm512_maskz_sqrt_pd 
   // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
   return _mm512_maskz_sqrt_pd (__U,__A);
 }
 
+__m512d test_mm512_mask_sqrt_round_pd(__m512d __W,__mmask8 __U,__m512d __A)
+{
+  // CHECK-LABEL: @test_mm512_mask_sqrt_round_pd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+  return _mm512_mask_sqrt_round_pd(__W,__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_sqrt_round_pd(__mmask8 __U,__m512d __A)
+{
+  // CHECK-LABEL: @test_mm512_maskz_sqrt_round_pd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+  return _mm512_maskz_sqrt_round_pd(__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_sqrt_round_pd(__m512d __A)
+{
+  // CHECK-LABEL: @test_mm512_sqrt_round_pd
+  // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+  return _mm512_sqrt_round_pd(__A,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512 test_mm512_sqrt_ps(__m512 a)
 {
   // CHECK-LABEL: @test_mm512_sqrt_ps
   // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
   return _mm512_sqrt_ps(a);
 }
 
 __m512 test_mm512_mask_sqrt_ps(__m512 __W, __mmask8 __U, __m512 __A)
 {
   // CHECK-LABEL: @test_mm512_mask_sqrt_ps
   // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
   return _mm512_mask_sqrt_ps( __W, __U, __A);
 }
 
 __m512 test_mm512_maskz_sqrt_ps( __mmask8 __U, __m512 __A)
 {
   // CHECK-LABEL: @test_mm512_maskz_sqrt_ps
   // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
   return _mm512_maskz_sqrt_ps(__U ,__A);
 }
 
+__m512 test_mm512_mask_sqrt_round_ps(__m512 __W,__mmask16 __U,__m512 __A)
+{
+  // CHECK-LABEL: @test_mm512_mask_sqrt_round_ps
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+  return _mm512_mask_sqrt_round_ps(__W,__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_sqrt_round_ps(__mmask16 __U,__m512 __A)
+{
+  // CHECK-LABEL: @test_mm512_maskz_sqrt_round_ps
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+  return _mm512_maskz_sqrt_round_ps(__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_sqrt_round_ps(__m512 __A)
+{
+  // CHECK-LABEL: @test_mm512_sqrt_round_ps
+  // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+  return _mm512_sqrt_round_ps(__A,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512d test_mm512_rsqrt14_pd(__m512d a)
 {
   // CHECK-LABEL: @test_mm512_rsqrt14_pd
   // CHECK: @llvm.x86.avx512.rsqrt14.pd.512
   return _mm512_rsqrt14_pd(a);
 }
 
 __m512d test_mm512_mask_rsqrt14_pd (__m512d __W, __mmask8 __U, __m512d __A)

 
 __m512 test_mm512_mask_max_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
 {
   // CHECK-LABEL: @test_mm512_mask_max_ps 
   // CHECK: @llvm.x86.avx512.mask.max.ps.512
   return _mm512_mask_max_ps (__W,__U,__A,__B);
 }
 
+__m512d test_mm512_mask_max_round_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_mask_max_round_pd
+  // CHECK: @llvm.x86.avx512.mask.max.pd.512
+  return _mm512_mask_max_round_pd(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_max_round_pd(__mmask8 __U,__m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_maskz_max_round_pd
+  // CHECK: @llvm.x86.avx512.mask.max.pd.512
+  return _mm512_maskz_max_round_pd(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_max_round_pd(__m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_max_round_pd
+  // CHECK: @llvm.x86.avx512.mask.max.pd.512
+  return _mm512_max_round_pd(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512 test_mm512_maskz_max_ps (__mmask16 __U, __m512 __A, __m512 __B)
 {
   // CHECK-LABEL: @test_mm512_maskz_max_ps 
   // CHECK: @llvm.x86.avx512.mask.max.ps.512
   return _mm512_maskz_max_ps (__U,__A,__B);
 }
 
+__m512 test_mm512_mask_max_round_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_mask_max_round_ps
+  // CHECK: @llvm.x86.avx512.mask.max.ps.512
+  return _mm512_mask_max_round_ps(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_max_round_ps(__mmask16 __U,__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_maskz_max_round_ps
+  // CHECK: @llvm.x86.avx512.mask.max.ps.512
+  return _mm512_maskz_max_round_ps(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_max_round_ps(__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_max_round_ps
+  // CHECK: @llvm.x86.avx512.mask.max.ps.512
+  return _mm512_max_round_ps(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512d test_mm512_mask_min_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
 {
   // CHECK-LABEL: @test_mm512_mask_min_pd 
   // CHECK: @llvm.x86.avx512.mask.min.pd.512
   return _mm512_mask_min_pd (__W,__U,__A,__B);
 }
 
 __m512d test_mm512_maskz_min_pd (__mmask8 __U, __m512d __A, __m512d __B)
 {
   // CHECK-LABEL: @test_mm512_maskz_min_pd 
   // CHECK: @llvm.x86.avx512.mask.min.pd.512
   return _mm512_maskz_min_pd (__U,__A,__B);
 }
 
+__m512d test_mm512_mask_min_round_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_mask_min_round_pd
+  // CHECK: @llvm.x86.avx512.mask.min.pd.512
+  return _mm512_mask_min_round_pd(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_min_round_pd(__mmask8 __U,__m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_maskz_min_round_pd
+  // CHECK: @llvm.x86.avx512.mask.min.pd.512
+  return _mm512_maskz_min_round_pd(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_min_round_pd( __m512d __A,__m512d __B)
+{
+  // CHECK-LABEL: @test_mm512_min_round_pd
+  // CHECK: @llvm.x86.avx512.mask.min.pd.512
+  return _mm512_min_round_pd(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512 test_mm512_mask_min_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
 {
   // CHECK-LABEL: @test_mm512_mask_min_ps 
   // CHECK: @llvm.x86.avx512.mask.min.ps.512
   return _mm512_mask_min_ps (__W,__U,__A,__B);
 }
 
 __m512 test_mm512_maskz_min_ps (__mmask16 __U, __m512 __A, __m512 __B)
 {
   // CHECK-LABEL: @test_mm512_maskz_min_ps 
   // CHECK: @llvm.x86.avx512.mask.min.ps.512
   return _mm512_maskz_min_ps (__U,__A,__B);
 }
 
+__m512 test_mm512_mask_min_round_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_mask_min_round_ps
+  // CHECK: @llvm.x86.avx512.mask.min.ps.512
+  return _mm512_mask_min_round_ps(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_min_round_ps(__mmask16 __U,__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_maskz_min_round_ps
+  // CHECK: @llvm.x86.avx512.mask.min.ps.512
+  return _mm512_maskz_min_round_ps(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_min_round_ps(__m512 __A,__m512 __B)
+{
+  // CHECK-LABEL: @test_mm512_min_round_ps
+  // CHECK: @llvm.x86.avx512.mask.min.ps.512
+  return _mm512_min_round_ps(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
 __m512 test_mm512_mask_floor_ps (__m512 __W, __mmask16 __U, __m512 __A)
 {
   // CHECK-LABEL: @test_mm512_mask_floor_ps 
   // CHECK: @llvm.x86.avx512.mask.rndscale.ps.512
   return _mm512_mask_floor_ps (__W,__U,__A);
 }
 
 __m512d test_mm512_mask_floor_pd (__m512d __W, __mmask8 __U, __m512d __A)