llvm-runtime-resources-darwin 0.0.0 → 23.1.0

package/include/endian.h

@@ -0,0 +1,91 @@
+//===-- Definition of macros from endian.h --------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLANG_ENDIAN_H
+#define __CLANG_ENDIAN_H
+
+// If the system has an endian.h, let's use that instead.
+#if __has_include_next(<endian.h>)
+#include_next <endian.h>
+#else
+
+#include <stdint.h>
+
+// Implementation taken from llvm libc endian-macros.h.
+#ifdef __cplusplus
+#define __CLANG_ENDIAN_CAST(cast, type, value) (cast<type>(value))
+#else
+#define __CLANG_ENDIAN_CAST(cast, type, value) ((type)(value))
+#endif
+
+#define LITTLE_ENDIAN __ORDER_LITTLE_ENDIAN__
+#define BIG_ENDIAN __ORDER_BIG_ENDIAN__
+#define PDP_ENDIAN __ORDER_PDP_ENDIAN__
+#define BYTE_ORDER __BYTE_ORDER__
+
+// Define some compatibility macros if they are not defined.
+#ifndef __BYTE_ORDER
+#define __BYTE_ORDER BYTE_ORDER
+#endif
+#ifndef __LITTLE_ENDIAN
+#define __LITTLE_ENDIAN LITTLE_ENDIAN
+#endif
+#ifndef __BIG_ENDIAN
+#define __BIG_ENDIAN BIG_ENDIAN
+#endif
+#ifndef __PDP_ENDIAN
+#define __PDP_ENDIAN PDP_ENDIAN
+#endif
+
+#if BYTE_ORDER == LITTLE_ENDIAN
+
+#define htobe16(x)                                                             \
+  __builtin_bswap16(__CLANG_ENDIAN_CAST(static_cast, uint16_t, x))
+#define htobe32(x)                                                             \
+  __builtin_bswap32(__CLANG_ENDIAN_CAST(static_cast, uint32_t, x))
+#define htobe64(x)                                                             \
+  __builtin_bswap64(__CLANG_ENDIAN_CAST(static_cast, uint64_t, x))
+#define htole16(x) __CLANG_ENDIAN_CAST(static_cast, uint16_t, x)
+#define htole32(x) __CLANG_ENDIAN_CAST(static_cast, uint32_t, x)
+#define htole64(x) __CLANG_ENDIAN_CAST(static_cast, uint64_t, x)
+#define be16toh(x)                                                             \
+  __builtin_bswap16(__CLANG_ENDIAN_CAST(static_cast, uint16_t, x))
+#define be32toh(x)                                                             \
+  __builtin_bswap32(__CLANG_ENDIAN_CAST(static_cast, uint32_t, x))
+#define be64toh(x)                                                             \
+  __builtin_bswap64(__CLANG_ENDIAN_CAST(static_cast, uint64_t, x))
+#define le16toh(x) __CLANG_ENDIAN_CAST(static_cast, uint16_t, x)
+#define le32toh(x) __CLANG_ENDIAN_CAST(static_cast, uint32_t, x)
+#define le64toh(x) __CLANG_ENDIAN_CAST(static_cast, uint64_t, x)
+
+#elif BYTE_ORDER == BIG_ENDIAN
+
+#define htobe16(x) __CLANG_ENDIAN_CAST(static_cast, uint16_t, x)
+#define htobe32(x) __CLANG_ENDIAN_CAST(static_cast, uint32_t, x)
+#define htobe64(x) __CLANG_ENDIAN_CAST(static_cast, uint64_t, x)
+#define htole16(x)                                                             \
+  __builtin_bswap16(__CLANG_ENDIAN_CAST(static_cast, uint16_t, x))
+#define htole32(x)                                                             \
+  __builtin_bswap32(__CLANG_ENDIAN_CAST(static_cast, uint32_t, x))
+#define htole64(x)                                                             \
+  __builtin_bswap64(__CLANG_ENDIAN_CAST(static_cast, uint64_t, x))
+#define be16toh(x) __CLANG_ENDIAN_CAST(static_cast, uint16_t, x)
+#define be32toh(x) __CLANG_ENDIAN_CAST(static_cast, uint32_t, x)
+#define be64toh(x) __CLANG_ENDIAN_CAST(static_cast, uint64_t, x)
+#define le16toh(x)                                                             \
+  __builtin_bswap16(__CLANG_ENDIAN_CAST(static_cast, uint16_t, x))
+#define le32toh(x)                                                             \
+  __builtin_bswap32(__CLANG_ENDIAN_CAST(static_cast, uint32_t, x))
+#define le64toh(x)                                                             \
+  __builtin_bswap64(__CLANG_ENDIAN_CAST(static_cast, uint64_t, x))
+
+#else
+#error "Unsupported endianness"
+#endif
+#endif // __has_include_next
+#endif // __CLANG_ENDIAN_H

package/include/enqcmdintrin.h

@@ -0,0 +1,63 @@
+/*===------------------ enqcmdintrin.h - enqcmd intrinsics -----------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __IMMINTRIN_H
+#error "Never use <enqcmdintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __ENQCMDINTRIN_H
+#define __ENQCMDINTRIN_H
+
+/* Define the default attributes for the functions in this file */
+#define _DEFAULT_FN_ATTRS \
+  __attribute__((__always_inline__, __nodebug__, __target__("enqcmd")))
+
+/// Reads 64-byte command pointed by \a __src, formats 64-byte enqueue store
+///    data, and performs 64-byte enqueue store to memory pointed by \a __dst.
+///    This intrinsics may only be used in User mode.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsics corresponds to the <c> ENQCMD </c> instruction.
+///
+/// \param __dst
+///    Pointer to the destination of the enqueue store.
+/// \param __src
+///    Pointer to 64-byte command data.
+/// \returns If the command data is successfully written to \a __dst then 0 is
+///    returned. Otherwise 1 is returned.
+static __inline__ int _DEFAULT_FN_ATTRS
+_enqcmd (void *__dst, const void *__src)
+{
+  return __builtin_ia32_enqcmd(__dst, __src);
+}
+
+/// Reads 64-byte command pointed by \a __src, formats 64-byte enqueue store
+///    data, and performs 64-byte enqueue store to memory pointed by \a __dst
+///    This intrinsic may only be used in Privileged mode.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsics corresponds to the <c> ENQCMDS </c> instruction.
+///
+/// \param __dst
+///    Pointer to the destination of the enqueue store.
+/// \param __src
+///    Pointer to 64-byte command data.
+/// \returns If the command data is successfully written to \a __dst then 0 is
+///    returned. Otherwise 1 is returned.
+static __inline__ int _DEFAULT_FN_ATTRS
+_enqcmds (void *__dst, const void *__src)
+{
+  return __builtin_ia32_enqcmds(__dst, __src);
+}
+
+#undef _DEFAULT_FN_ATTRS
+
+#endif /* __ENQCMDINTRIN_H */

package/include/f16cintrin.h

@@ -0,0 +1,176 @@
+/*===---- f16cintrin.h - F16C intrinsics -----------------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#if !defined __IMMINTRIN_H
+#error "Never use <f16cintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __F16CINTRIN_H
+#define __F16CINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS128                                                  \
+  __attribute__((__always_inline__, __nodebug__, __target__("f16c"),           \
+                 __min_vector_width__(128))) constexpr
+#define __DEFAULT_FN_ATTRS256                                                  \
+  __attribute__((__always_inline__, __nodebug__, __target__("f16c"),           \
+                 __min_vector_width__(256))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS128                                                  \
+  __attribute__((__always_inline__, __nodebug__, __target__("f16c"),           \
+                 __min_vector_width__(128)))
+#define __DEFAULT_FN_ATTRS256                                                  \
+  __attribute__((__always_inline__, __nodebug__, __target__("f16c"),           \
+                 __min_vector_width__(256)))
+#endif
+
+/* NOTE: Intel documents the 128-bit versions of these as being in emmintrin.h,
+ * but that's because icc can emulate these without f16c using a library call.
+ * Since we don't do that let's leave these in f16cintrin.h.
+ */
+
+/// Converts a 16-bit half-precision float value into a 32-bit float
+///    value.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> VCVTPH2PS </c> instruction.
+///
+/// \param __a
+///    A 16-bit half-precision float value.
+/// \returns The converted 32-bit float value.
+static __inline float __DEFAULT_FN_ATTRS128
+_cvtsh_ss(unsigned short __a)
+{
+  return (float)__builtin_bit_cast(__fp16, __a);
+}
+
+/// Converts a 32-bit single-precision float value to a 16-bit
+///    half-precision float value.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \code
+/// unsigned short _cvtss_sh(float a, const int imm);
+/// \endcode
+///
+/// This intrinsic corresponds to the <c> VCVTPS2PH </c> instruction.
+///
+/// \param a
+///    A 32-bit single-precision float value to be converted to a 16-bit
+///    half-precision float value.
+/// \param imm
+///    An immediate value controlling rounding using bits [2:0]: \n
+///    000: Nearest \n
+///    001: Down \n
+///    010: Up \n
+///    011: Truncate \n
+///    1XX: Use MXCSR.RC for rounding
+/// \returns The converted 16-bit half-precision float value.
+#define _cvtss_sh(a, imm) __extension__ ({ \
+  (unsigned short)(((__v8hi)__builtin_ia32_vcvtps2ph((__v4sf){a, 0, 0, 0}, \
+                                                     (imm)))[0]); })
+
+/// Converts a 128-bit vector containing 32-bit float values into a
+///    128-bit vector containing 16-bit half-precision float values.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \code
+/// __m128i _mm_cvtps_ph(__m128 a, const int imm);
+/// \endcode
+///
+/// This intrinsic corresponds to the <c> VCVTPS2PH </c> instruction.
+///
+/// \param a
+///    A 128-bit vector containing 32-bit float values.
+/// \param imm
+///    An immediate value controlling rounding using bits [2:0]: \n
+///    000: Nearest \n
+///    001: Down \n
+///    010: Up \n
+///    011: Truncate \n
+///    1XX: Use MXCSR.RC for rounding
+/// \returns A 128-bit vector containing converted 16-bit half-precision float
+///    values. The lower 64 bits are used to store the converted 16-bit
+///    half-precision floating-point values.
+#define _mm_cvtps_ph(a, imm) \
+  ((__m128i)__builtin_ia32_vcvtps2ph((__v4sf)(__m128)(a), (imm)))
+
+/// Converts a 128-bit vector containing 16-bit half-precision float
+///    values into a 128-bit vector containing 32-bit float values.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> VCVTPH2PS </c> instruction.
+///
+/// \param __a
+///    A 128-bit vector containing 16-bit half-precision float values. The lower
+///    64 bits are used in the conversion.
+/// \returns A 128-bit vector of [4 x float] containing converted float values.
+static __inline __m128 __DEFAULT_FN_ATTRS128
+_mm_cvtph_ps(__m128i __a)
+{
+  typedef __fp16 __v4fp16 __attribute__((__vector_size__(8)));
+
+  __v4hi __v = __builtin_shufflevector((__v8hi)__a, (__v8hi)__a, 0, 1, 2, 3);
+  return (__m128) __builtin_convertvector((__v4fp16)__v, __v4sf);
+}
+
+/// Converts a 256-bit vector of [8 x float] into a 128-bit vector
+///    containing 16-bit half-precision float values.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \code
+/// __m128i _mm256_cvtps_ph(__m256 a, const int imm);
+/// \endcode
+///
+/// This intrinsic corresponds to the <c> VCVTPS2PH </c> instruction.
+///
+/// \param a
+///    A 256-bit vector containing 32-bit single-precision float values to be
+///    converted to 16-bit half-precision float values.
+/// \param imm
+///    An immediate value controlling rounding using bits [2:0]: \n
+///    000: Nearest \n
+///    001: Down \n
+///    010: Up \n
+///    011: Truncate \n
+///    1XX: Use MXCSR.RC for rounding
+/// \returns A 128-bit vector containing the converted 16-bit half-precision
+///    float values.
+#define _mm256_cvtps_ph(a, imm) \
+ ((__m128i)__builtin_ia32_vcvtps2ph256((__v8sf)(__m256)(a), (imm)))
+
+/// Converts a 128-bit vector containing 16-bit half-precision float
+///    values into a 256-bit vector of [8 x float].
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> VCVTPH2PS </c> instruction.
+///
+/// \param __a
+///    A 128-bit vector containing 16-bit half-precision float values to be
+///    converted to 32-bit single-precision float values.
+/// \returns A vector of [8 x float] containing the converted 32-bit
+///    single-precision float values.
+static __inline __m256 __DEFAULT_FN_ATTRS256
+_mm256_cvtph_ps(__m128i __a)
+{
+  typedef __fp16 __v8fp16 __attribute__((__vector_size__(16), __aligned__(16)));
+
+  return (__m256) __builtin_convertvector((__v8fp16)__a, __v8sf);
+}
+
+#undef __DEFAULT_FN_ATTRS128
+#undef __DEFAULT_FN_ATTRS256
+
+#endif /* __F16CINTRIN_H */

package/include/float.h

@@ -0,0 +1,48 @@
+/*===---- float.h - Characteristics of floating point types ----------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#if defined(__MVS__) && __has_include_next(<float.h>)
+#include <__float_header_macro.h>
+#include_next <float.h>
+#else
+
+#if !defined(__need_infinity_nan)
+#define __need_float_float
+#if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 202311L) ||              \
+    !defined(__STRICT_ANSI__)
+#define __need_infinity_nan
+#endif
+#include <__float_header_macro.h>
+#endif
+
+#ifdef __need_float_float
+/* If we're on MinGW, fall back to the system's float.h, which might have
+ * additional definitions provided for Windows.
+ * For more details see http://msdn.microsoft.com/en-us/library/y0ybw9fy.aspx
+ *
+ * Also fall back on AIX to allow additional definitions and
+ * implementation-defined values.
+ */
+#if (defined(__MINGW32__) || defined(_MSC_VER) || defined(_AIX)) &&            \
+    __STDC_HOSTED__ && __has_include_next(<float.h>)
+
+#  include_next <float.h>
+
+#endif
+
+#include <__float_float.h>
+#undef __need_float_float
+#endif
+
+#ifdef __need_infinity_nan
+#include <__float_infinity_nan.h>
+#undef __need_infinity_nan
+#endif
+
+#endif /* __MVS__ */

package/include/fma4intrin.h

@@ -0,0 +1,220 @@
+/*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __X86INTRIN_H
+#error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead."
+#endif
+
+#ifndef __FMA4INTRIN_H
+#define __FMA4INTRIN_H
+
+#include <pmmintrin.h>
+
+/* Define the default attributes for the functions in this file. */
+#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)))
+
+#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_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_CONSTEXPR
+_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C) {
+  return _mm_set_ss(__builtin_elementwise_fma(__A[0], __B[0], __C[0]));
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C) {
+  return _mm_set_sd(__builtin_elementwise_fma(__A[0], __B[0], __C[0]));
+}
+
+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_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_CONSTEXPR
+_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C) {
+  return _mm_set_ss(__builtin_elementwise_fma(__A[0], __B[0], -__C[0]));
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C) {
+  return _mm_set_sd(__builtin_elementwise_fma(__A[0], __B[0], -__C[0]));
+}
+
+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_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_CONSTEXPR
+_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C) {
+  return _mm_set_ss(__builtin_elementwise_fma(-__A[0], __B[0], __C[0]));
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C) {
+  return _mm_set_sd(__builtin_elementwise_fma(-__A[0], __B[0], __C[0]));
+}
+
+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_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_CONSTEXPR
+_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C) {
+  return _mm_set_ss(__builtin_elementwise_fma(-__A[0], __B[0], -__C[0]));
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C) {
+  return _mm_set_sd(__builtin_elementwise_fma(-__A[0], __B[0], -__C[0]));
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS128
+_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
+{
+  return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128
+_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
+{
+  return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS128
+_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
+{
+  return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS128
+_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_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_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_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_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_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_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_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_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
+_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
+{
+  return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS256
+_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
+{
+  return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS256
+_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
+{
+  return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
+}
+
+static __inline__ __m256d __DEFAULT_FN_ATTRS256
+_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
+{
+  return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
+}
+
+#undef __DEFAULT_FN_ATTRS128
+#undef __DEFAULT_FN_ATTRS256
+#undef __DEFAULT_FN_ATTRS128_CONSTEXPR
+#undef __DEFAULT_FN_ATTRS256_CONSTEXPR
+
+#endif /* __FMA4INTRIN_H */