[AVX-512] Replace subvector broadcast builtins with shufflevectors and selects.

Verified that the backend codegens this equally well.

llvm-svn: 292329

4 files changed, 133 insertions, 151 deletions

diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index ae44b98a949..4fd1add7735 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -995,51 +995,50 @@ _mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A)
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_broadcast_f32x8 (__m256 __A)
+_mm512_broadcast_f32x8(__m256 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
-                _mm512_undefined_ps(),
-                (__mmask16) -1);
+  return (__m512)__builtin_shufflevector((__v8sf)__A, (__v8sf)__A,
+                                         0, 1, 2, 3, 4, 5, 6, 7,
+                                         0, 1, 2, 3, 4, 5, 6, 7);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f32x8 (__m512 __O, __mmask16 __M, __m256 __A)
+_mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
-                (__v16sf)__O,
-                __M);
+  return (__m512)__builtin_ia32_selectps_512((__mmask8)__M,
+                                           (__v16sf)_mm512_broadcast_f32x8(__A),
+                                           (__v16sf)__O);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f32x8 (__mmask16 __M, __m256 __A)
+_mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
-                (__v16sf)_mm512_setzero_ps (),
-                __M);
+  return (__m512)__builtin_ia32_selectps_512((__mmask8)__M,
+                                           (__v16sf)_mm512_broadcast_f32x8(__A),
+                                           (__v16sf)_mm512_setzero_ps());
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_broadcast_f64x2 (__m128d __A)
+_mm512_broadcast_f64x2(__m128d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
-                 (__v8df)_mm512_undefined_pd(),
-                 (__mmask8) -1);
+  return (__m512d)__builtin_shufflevector((__v2df)__A, (__v2df)__A,
+                                          0, 1, 0, 1, 0, 1, 0, 1);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f64x2 (__m512d __O, __mmask8 __M, __m128d __A)
+_mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
-                 (__v8df)
-                 __O, __M);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+                                            (__v8df)_mm512_broadcast_f64x2(__A),
+                                            (__v8df)__O);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A)
+_mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
-                 (__v8df)_mm512_setzero_ps (),
-                 __M);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+                                            (__v8df)_mm512_broadcast_f64x2(__A),
+                                            (__v8df)_mm512_setzero_pd());
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -1067,52 +1066,50 @@ _mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A)
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i32x8 (__m256i __A)
+_mm512_broadcast_i32x8(__m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
-                 (__v16si)_mm512_setzero_si512(),
-                 (__mmask16) -1);
+  return (__m512i)__builtin_shufflevector((__v8si)__A, (__v8si)__A,
+                                          0, 1, 2, 3, 4, 5, 6, 7,
+                                          0, 1, 2, 3, 4, 5, 6, 7);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i32x8 (__m512i __O, __mmask16 __M, __m256i __A)
+_mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
-                 (__v16si)__O,
-                 __M);
+  return (__m512i)__builtin_ia32_selectd_512((__mmask8)__M,
+                                           (__v16si)_mm512_broadcast_i32x8(__A),
+                                           (__v16si)__O);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i32x8 (__mmask16 __M, __m256i __A)
+_mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
-                 (__v16si)
-                 _mm512_setzero_si512 (),
-                 __M);
+  return (__m512i)__builtin_ia32_selectd_512((__mmask8)__M,
+                                           (__v16si)_mm512_broadcast_i32x8(__A),
+                                           (__v16si)_mm512_setzero_si512());
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i64x2 (__m128i __A)
+_mm512_broadcast_i64x2(__m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
-                 (__v8di)_mm512_setzero_si512(),
-                 (__mmask8) -1);
+  return (__m512i)__builtin_shufflevector((__v2di)__A, (__v2di)__A,
+                                          0, 1, 0, 1, 0, 1, 0, 1);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i64x2 (__m512i __O, __mmask8 __M, __m128i __A)
+_mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
-                 (__v8di)
-                 __O, __M);
+  return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+                                            (__v8di)_mm512_broadcast_i64x2(__A),
+                                            (__v8di)__O);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
+_mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
-                 (__v8di)_mm512_setzero_si512 (),
-                 __M);
+  return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+                                            (__v8di)_mm512_broadcast_i64x2(__A),
+                                            (__v8di)_mm512_setzero_si512());
 }
 
 #define _mm512_extractf32x8_ps(A, imm) __extension__ ({ \
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index e6a7217c896..7d9a52fdf23 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -7278,107 +7278,97 @@ _mm_maskz_sqrt_ss (__mmask8 __U, __m128 __A, __m128 __B)
                                            (__mmask8)(U), (int)(R)); })
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_broadcast_f32x4 (__m128 __A)
+_mm512_broadcast_f32x4(__m128 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
-                 (__v16sf)
-                 _mm512_undefined_ps (),
-                 (__mmask16) -1);
+  return (__m512)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A,
+                                         0, 1, 2, 3, 0, 1, 2, 3,
+                                         0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f32x4 (__m512 __O, __mmask16 __M, __m128 __A)
+_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, __m128 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
-                 (__v16sf) __O,
-                 __M);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+                                           (__v16sf)_mm512_broadcast_f32x4(__A),
+                                           (__v16sf)__O);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f32x4 (__mmask16 __M, __m128 __A)
+_mm512_maskz_broadcast_f32x4(__mmask16 __M, __m128 __A)
 {
-  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
-                 (__v16sf)
-                 _mm512_setzero_ps (),
-                 __M);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+                                           (__v16sf)_mm512_broadcast_f32x4(__A),
+                                           (__v16sf)_mm512_setzero_ps());
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_broadcast_f64x4 (__m256d __A)
+_mm512_broadcast_f64x4(__m256d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
-                  (__v8df)
-                  _mm512_undefined_pd (),
-                  (__mmask8) -1);
+  return (__m512d)__builtin_shufflevector((__v4df)__A, (__v4df)__A,
+                                          0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f64x4 (__m512d __O, __mmask8 __M, __m256d __A)
+_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, __m256d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
-                  (__v8df) __O,
-                  __M);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+                                            (__v8df)_mm512_broadcast_f64x4(__A),
+                                            (__v8df)__O);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f64x4 (__mmask8 __M, __m256d __A)
+_mm512_maskz_broadcast_f64x4(__mmask8 __M, __m256d __A)
 {
-  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
-                  (__v8df)
-                  _mm512_setzero_pd (),
-                  __M);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+                                            (__v8df)_mm512_broadcast_f64x4(__A),
+                                            (__v8df)_mm512_setzero_pd());
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i32x4 (__m128i __A)
+_mm512_broadcast_i32x4(__m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
-                  (__v16si)
-                  _mm512_undefined_epi32 (),
-                  (__mmask16) -1);
+  return (__m512i)__builtin_shufflevector((__v4si)__A, (__v4si)__A,
+                                          0, 1, 2, 3, 0, 1, 2, 3,
+                                          0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i32x4 (__m512i __O, __mmask16 __M, __m128i __A)
+_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
-                  (__v16si) __O,
-                  __M);
+  return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+                                           (__v16si)_mm512_broadcast_i32x4(__A),
+                                           (__v16si)__O);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i32x4 (__mmask16 __M, __m128i __A)
+_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
-                  (__v16si)
-                  _mm512_setzero_si512 (),
-                  __M);
+  return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+                                           (__v16si)_mm512_broadcast_i32x4(__A),
+                                           (__v16si)_mm512_setzero_si512());
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i64x4 (__m256i __A)
+_mm512_broadcast_i64x4(__m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
-                  (__v8di)
-                  _mm512_undefined_epi32 (),
-                  (__mmask8) -1);
+  return (__m512i)__builtin_shufflevector((__v4di)__A, (__v4di)__A,
+                                          0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i64x4 (__m512i __O, __mmask8 __M, __m256i __A)
+_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
-                  (__v8di) __O,
-                  __M);
+  return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+                                            (__v8di)_mm512_broadcast_i64x4(__A),
+                                            (__v8di)__O);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i64x4 (__mmask8 __M, __m256i __A)
+_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i __A)
 {
-  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
-                  (__v8di)
-                  _mm512_setzero_si512 (),
-                  __M);
+  return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+                                            (__v8di)_mm512_broadcast_i64x4(__A),
+                                            (__v8di)_mm512_setzero_si512());
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vldqintrin.h b/clang/lib/Headers/avx512vldqintrin.h
index cd9da437056..aecd7df34d0 100644
--- a/clang/lib/Headers/avx512vldqintrin.h
+++ b/clang/lib/Headers/avx512vldqintrin.h
@@ -1000,27 +1000,26 @@ _mm256_maskz_broadcast_f32x2 (__mmask8 __M, __m128 __A)
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_broadcast_f64x2 (__m128d __A)
+_mm256_broadcast_f64x2(__m128d __A)
 {
-  return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df) __A,
-                 (__v4df)_mm256_undefined_pd(),
-                 (__mmask8) -1);
+  return (__m256d)__builtin_shufflevector((__v2df)__A, (__v2df)__A,
+                                          0, 1, 0, 1);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_f64x2 (__m256d __O, __mmask8 __M, __m128d __A)
+_mm256_mask_broadcast_f64x2(__m256d __O, __mmask8 __M, __m128d __A)
 {
-  return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df) __A,
-                 (__v4df) __O,
-                 __M);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__M,
+                                            (__v4df)_mm256_broadcast_f64x2(__A),
+                                            (__v4df)__O);
 }
 
 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);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__M,
+                                            (__v4df)_mm256_broadcast_f64x2(__A),
+                                            (__v4df)_mm256_setzero_pd());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
@@ -1072,27 +1071,26 @@ _mm256_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A)
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_broadcast_i64x2 (__m128i __A)
+_mm256_broadcast_i64x2(__m128i __A)
 {
-  return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di) __A,
-                 (__v4di)_mm256_undefined_si256(),
-                 (__mmask8) -1);
+  return (__m256i)__builtin_shufflevector((__v2di)__A, (__v2di)__A,
+                                          0, 1, 0, 1);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_i64x2 (__m256i __O, __mmask8 __M, __m128i __A)
+_mm256_mask_broadcast_i64x2(__m256i __O, __mmask8 __M, __m128i __A)
 {
-  return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di) __A,
-                 (__v4di) __O,
-                 __M);
+  return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+                                            (__v4di)_mm256_broadcast_i64x2(__A),
+                                            (__v4di)__O);
 }
 
 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);
+  return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+                                            (__v4di)_mm256_broadcast_i64x2(__A),
+                                            (__v4di)_mm256_setzero_si256());
 }
 
 #define _mm256_extractf64x2_pd(A, imm) __extension__ ({ \
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index f3744da6ab8..99bb050de4d 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -7189,52 +7189,49 @@ _mm256_maskz_rsqrt14_ps (__mmask8 __U, __m256 __A)
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_broadcast_f32x4 (__m128 __A)
+_mm256_broadcast_f32x4(__m128 __A)
 {
-  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
-                (__v8sf)_mm256_undefined_pd (),
-                (__mmask8) -1);
+  return (__m256)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A,
+                                         0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_f32x4 (__m256 __O, __mmask8 __M, __m128 __A)
+_mm256_mask_broadcast_f32x4(__m256 __O, __mmask8 __M, __m128 __A)
 {
-  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
-                (__v8sf) __O,
-                __M);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__M,
+                                            (__v8sf)_mm256_broadcast_f32x4(__A),
+                                            (__v8sf)__O);
 }
 
 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);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__M,
+                                            (__v8sf)_mm256_broadcast_f32x4(__A),
+                                            (__v8sf)_mm256_setzero_ps());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_broadcast_i32x4 (__m128i __A)
+_mm256_broadcast_i32x4(__m128i __A)
 {
-  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si) __A,
-                 (__v8si)_mm256_undefined_si256 (),
-                 (__mmask8) -1);
+  return (__m256i)__builtin_shufflevector((__v4si)__A, (__v4si)__A,
+                                          0, 1, 2, 3, 0, 1, 2, 3);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_i32x4 (__m256i __O, __mmask8 __M, __m128i __A)
+_mm256_mask_broadcast_i32x4(__m256i __O, __mmask8 __M, __m128i __A)
 {
-  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si) __A,
-                 (__v8si)
-                 __O, __M);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                            (__v8si)_mm256_broadcast_i32x4(__A),
+                                            (__v8si)__O);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_broadcast_i32x4 (__mmask8 __M, __m128i __A)
+_mm256_maskz_broadcast_i32x4(__mmask8 __M, __m128i __A)
 {
-  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
-                 __A,
-                 (__v8si) _mm256_setzero_si256 (),
-                 __M);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                            (__v8si)_mm256_broadcast_i32x4(__A),
+                                            (__v8si)_mm256_setzero_si256());
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS