aboutsummaryrefslogtreecommitdiff
path: root/clang/lib/Headers/fma4intrin.h
diff options
context:
space:
mode:
Diffstat (limited to 'clang/lib/Headers/fma4intrin.h')
-rw-r--r--clang/lib/Headers/fma4intrin.h138
1 files changed, 74 insertions, 64 deletions
diff --git a/clang/lib/Headers/fma4intrin.h b/clang/lib/Headers/fma4intrin.h
index 694801b..e0a0e4c 100644
--- a/clang/lib/Headers/fma4intrin.h
+++ b/clang/lib/Headers/fma4intrin.h
@@ -20,16 +20,24 @@
#define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(128)))
#define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(256)))
-static __inline__ __m128 __DEFAULT_FN_ATTRS128
-_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
-{
- return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128 constexpr
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 constexpr
+#else
+#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256
+#endif
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C) {
+ return (__m128)__builtin_elementwise_fma((__v4sf)__A, (__v4sf)__B,
+ (__v4sf)__C);
}
-static __inline__ __m128d __DEFAULT_FN_ATTRS128
-_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
-{
- return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C) {
+ return (__m128d)__builtin_elementwise_fma((__v2df)__A, (__v2df)__B,
+ (__v2df)__C);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS128
@@ -44,16 +52,16 @@ _mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
}
-static __inline__ __m128 __DEFAULT_FN_ATTRS128
-_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
-{
- return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
+static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C) {
+ return (__m128)__builtin_elementwise_fma((__v4sf)__A, (__v4sf)__B,
+ -(__v4sf)__C);
}
-static __inline__ __m128d __DEFAULT_FN_ATTRS128
-_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
-{
- return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C) {
+ return (__m128d)__builtin_elementwise_fma((__v2df)__A, (__v2df)__B,
+ -(__v2df)__C);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS128
@@ -68,16 +76,16 @@ _mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
}
-static __inline__ __m128 __DEFAULT_FN_ATTRS128
-_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
-{
- return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C) {
+ return (__m128)__builtin_elementwise_fma(-(__v4sf)__A, (__v4sf)__B,
+ (__v4sf)__C);
}
-static __inline__ __m128d __DEFAULT_FN_ATTRS128
-_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
-{
- return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C) {
+ return (__m128d)__builtin_elementwise_fma(-(__v2df)__A, (__v2df)__B,
+ (__v2df)__C);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS128
@@ -92,16 +100,16 @@ _mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
}
-static __inline__ __m128 __DEFAULT_FN_ATTRS128
-_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
-{
- return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
+static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C) {
+ return (__m128)__builtin_elementwise_fma(-(__v4sf)__A, (__v4sf)__B,
+ -(__v4sf)__C);
}
-static __inline__ __m128d __DEFAULT_FN_ATTRS128
-_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
-{
- return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C) {
+ return (__m128d)__builtin_elementwise_fma(-(__v2df)__A, (__v2df)__B,
+ -(__v2df)__C);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS128
@@ -140,52 +148,52 @@ _mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
}
-static __inline__ __m256 __DEFAULT_FN_ATTRS256
-_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
-{
- return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C) {
+ return (__m256)__builtin_elementwise_fma((__v8sf)__A, (__v8sf)__B,
+ (__v8sf)__C);
}
-static __inline__ __m256d __DEFAULT_FN_ATTRS256
-_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
-{
- return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
+static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C) {
+ return (__m256d)__builtin_elementwise_fma((__v4df)__A, (__v4df)__B,
+ (__v4df)__C);
}
-static __inline__ __m256 __DEFAULT_FN_ATTRS256
-_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
-{
- return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
+static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C) {
+ return (__m256)__builtin_elementwise_fma((__v8sf)__A, (__v8sf)__B,
+ -(__v8sf)__C);
}
-static __inline__ __m256d __DEFAULT_FN_ATTRS256
-_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
-{
- return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
+static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C) {
+ return (__m256d)__builtin_elementwise_fma((__v4df)__A, (__v4df)__B,
+ -(__v4df)__C);
}
-static __inline__ __m256 __DEFAULT_FN_ATTRS256
-_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
-{
- return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C) {
+ return (__m256)__builtin_elementwise_fma(-(__v8sf)__A, (__v8sf)__B,
+ (__v8sf)__C);
}
-static __inline__ __m256d __DEFAULT_FN_ATTRS256
-_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
-{
- return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
+static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C) {
+ return (__m256d)__builtin_elementwise_fma(-(__v4df)__A, (__v4df)__B,
+ (__v4df)__C);
}
-static __inline__ __m256 __DEFAULT_FN_ATTRS256
-_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
-{
- return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
+static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C) {
+ return (__m256)__builtin_elementwise_fma(-(__v8sf)__A, (__v8sf)__B,
+ -(__v8sf)__C);
}
-static __inline__ __m256d __DEFAULT_FN_ATTRS256
-_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
-{
- return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
+static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C) {
+ return (__m256d)__builtin_elementwise_fma(-(__v4df)__A, (__v4df)__B,
+ -(__v4df)__C);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS256
@@ -214,5 +222,7 @@ _mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
#undef __DEFAULT_FN_ATTRS128
#undef __DEFAULT_FN_ATTRS256
+#undef __DEFAULT_FN_ATTRS128_CONSTEXPR
+#undef __DEFAULT_FN_ATTRS256_CONSTEXPR
#endif /* __FMA4INTRIN_H */
generated by cgit v1.1 at 2025-08-31 21:21:48 +0000