gcc/config/aarch64/arm_neon.h

/* ARM NEON intrinsics include file.

   Copyright (C) 2011-2021 Free Software Foundation, Inc.
   Contributed by ARM Ltd.

   This file is part of GCC.

   GCC is free software; you can redistribute it and/or modify it
   under the terms of the GNU General Public License as published
   by the Free Software Foundation; either version 3, or (at your
   option) any later version.

   GCC is distributed in the hope that it will be useful, but WITHOUT
   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
   License for more details.

   Under Section 7 of GPL version 3, you are granted additional
   permissions described in the GCC Runtime Library Exception, version
   3.1, as published by the Free Software Foundation.

   You should have received a copy of the GNU General Public License and
   a copy of the GCC Runtime Library Exception along with this program;
   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
   <http://www.gnu.org/licenses/>.  */

#ifndef _AARCH64_NEON_H_
#define _AARCH64_NEON_H_

#pragma GCC push_options
#pragma GCC target ("+nothing+simd")

#include <stdint.h>

#define __AARCH64_UINT64_C(__C) ((uint64_t) __C)
#define __AARCH64_INT64_C(__C) ((int64_t) __C)

typedef __Int8x8_t int8x8_t;
typedef __Int16x4_t int16x4_t;
typedef __Int32x2_t int32x2_t;
typedef __Int64x1_t int64x1_t;
typedef __Float16x4_t float16x4_t;
typedef __Float32x2_t float32x2_t;
typedef __Poly8x8_t poly8x8_t;
typedef __Poly16x4_t poly16x4_t;
typedef __Uint8x8_t uint8x8_t;
typedef __Uint16x4_t uint16x4_t;
typedef __Uint32x2_t uint32x2_t;
typedef __Float64x1_t float64x1_t;
typedef __Uint64x1_t uint64x1_t;
typedef __Int8x16_t int8x16_t;
typedef __Int16x8_t int16x8_t;
typedef __Int32x4_t int32x4_t;
typedef __Int64x2_t int64x2_t;
typedef __Float16x8_t float16x8_t;
typedef __Float32x4_t float32x4_t;
typedef __Float64x2_t float64x2_t;
typedef __Poly8x16_t poly8x16_t;
typedef __Poly16x8_t poly16x8_t;
typedef __Poly64x2_t poly64x2_t;
typedef __Poly64x1_t poly64x1_t;
typedef __Uint8x16_t uint8x16_t;
typedef __Uint16x8_t uint16x8_t;
typedef __Uint32x4_t uint32x4_t;
typedef __Uint64x2_t uint64x2_t;

typedef __Poly8_t poly8_t;
typedef __Poly16_t poly16_t;
typedef __Poly64_t poly64_t;
typedef __Poly128_t poly128_t;

typedef __fp16 float16_t;
typedef float float32_t;
typedef double float64_t;

typedef __Bfloat16x4_t bfloat16x4_t;
typedef __Bfloat16x8_t bfloat16x8_t;

typedef struct bfloat16x4x2_t
{
  bfloat16x4_t val[2];
} bfloat16x4x2_t;

typedef struct bfloat16x8x2_t
{
  bfloat16x8_t val[2];
} bfloat16x8x2_t;

typedef struct bfloat16x4x3_t
{
  bfloat16x4_t val[3];
} bfloat16x4x3_t;

typedef struct bfloat16x8x3_t
{
  bfloat16x8_t val[3];
} bfloat16x8x3_t;

typedef struct bfloat16x4x4_t
{
  bfloat16x4_t val[4];
} bfloat16x4x4_t;

typedef struct bfloat16x8x4_t
{
  bfloat16x8_t val[4];
} bfloat16x8x4_t;

typedef struct int8x8x2_t
{
  int8x8_t val[2];
} int8x8x2_t;

typedef struct int8x16x2_t
{
  int8x16_t val[2];
} int8x16x2_t;

typedef struct int16x4x2_t
{
  int16x4_t val[2];
} int16x4x2_t;

typedef struct int16x8x2_t
{
  int16x8_t val[2];
} int16x8x2_t;

typedef struct int32x2x2_t
{
  int32x2_t val[2];
} int32x2x2_t;

typedef struct int32x4x2_t
{
  int32x4_t val[2];
} int32x4x2_t;

typedef struct int64x1x2_t
{
  int64x1_t val[2];
} int64x1x2_t;

typedef struct int64x2x2_t
{
  int64x2_t val[2];
} int64x2x2_t;

typedef struct uint8x8x2_t
{
  uint8x8_t val[2];
} uint8x8x2_t;

typedef struct uint8x16x2_t
{
  uint8x16_t val[2];
} uint8x16x2_t;

typedef struct uint16x4x2_t
{
  uint16x4_t val[2];
} uint16x4x2_t;

typedef struct uint16x8x2_t
{
  uint16x8_t val[2];
} uint16x8x2_t;

typedef struct uint32x2x2_t
{
  uint32x2_t val[2];
} uint32x2x2_t;

typedef struct uint32x4x2_t
{
  uint32x4_t val[2];
} uint32x4x2_t;

typedef struct uint64x1x2_t
{
  uint64x1_t val[2];
} uint64x1x2_t;

typedef struct uint64x2x2_t
{
  uint64x2_t val[2];
} uint64x2x2_t;

typedef struct float16x4x2_t
{
  float16x4_t val[2];
} float16x4x2_t;

typedef struct float16x8x2_t
{
  float16x8_t val[2];
} float16x8x2_t;

typedef struct float32x2x2_t
{
  float32x2_t val[2];
} float32x2x2_t;

typedef struct float32x4x2_t
{
  float32x4_t val[2];
} float32x4x2_t;

typedef struct float64x2x2_t
{
  float64x2_t val[2];
} float64x2x2_t;

typedef struct float64x1x2_t
{
  float64x1_t val[2];
} float64x1x2_t;

typedef struct poly8x8x2_t
{
  poly8x8_t val[2];
} poly8x8x2_t;

typedef struct poly8x16x2_t
{
  poly8x16_t val[2];
} poly8x16x2_t;

typedef struct poly16x4x2_t
{
  poly16x4_t val[2];
} poly16x4x2_t;

typedef struct poly16x8x2_t
{
  poly16x8_t val[2];
} poly16x8x2_t;

typedef struct poly64x1x2_t
{
  poly64x1_t val[2];
} poly64x1x2_t;

typedef struct poly64x1x3_t
{
  poly64x1_t val[3];
} poly64x1x3_t;

typedef struct poly64x1x4_t
{
  poly64x1_t val[4];
} poly64x1x4_t;

typedef struct poly64x2x2_t
{
  poly64x2_t val[2];
} poly64x2x2_t;

typedef struct poly64x2x3_t
{
  poly64x2_t val[3];
} poly64x2x3_t;

typedef struct poly64x2x4_t
{
  poly64x2_t val[4];
} poly64x2x4_t;

typedef struct int8x8x3_t
{
  int8x8_t val[3];
} int8x8x3_t;

typedef struct int8x16x3_t
{
  int8x16_t val[3];
} int8x16x3_t;

typedef struct int16x4x3_t
{
  int16x4_t val[3];
} int16x4x3_t;

typedef struct int16x8x3_t
{
  int16x8_t val[3];
} int16x8x3_t;

typedef struct int32x2x3_t
{
  int32x2_t val[3];
} int32x2x3_t;

typedef struct int32x4x3_t
{
  int32x4_t val[3];
} int32x4x3_t;

typedef struct int64x1x3_t
{
  int64x1_t val[3];
} int64x1x3_t;

typedef struct int64x2x3_t
{
  int64x2_t val[3];
} int64x2x3_t;

typedef struct uint8x8x3_t
{
  uint8x8_t val[3];
} uint8x8x3_t;

typedef struct uint8x16x3_t
{
  uint8x16_t val[3];
} uint8x16x3_t;

typedef struct uint16x4x3_t
{
  uint16x4_t val[3];
} uint16x4x3_t;

typedef struct uint16x8x3_t
{
  uint16x8_t val[3];
} uint16x8x3_t;

typedef struct uint32x2x3_t
{
  uint32x2_t val[3];
} uint32x2x3_t;

typedef struct uint32x4x3_t
{
  uint32x4_t val[3];
} uint32x4x3_t;

typedef struct uint64x1x3_t
{
  uint64x1_t val[3];
} uint64x1x3_t;

typedef struct uint64x2x3_t
{
  uint64x2_t val[3];
} uint64x2x3_t;

typedef struct float16x4x3_t
{
  float16x4_t val[3];
} float16x4x3_t;

typedef struct float16x8x3_t
{
  float16x8_t val[3];
} float16x8x3_t;

typedef struct float32x2x3_t
{
  float32x2_t val[3];
} float32x2x3_t;

typedef struct float32x4x3_t
{
  float32x4_t val[3];
} float32x4x3_t;

typedef struct float64x2x3_t
{
  float64x2_t val[3];
} float64x2x3_t;

typedef struct float64x1x3_t
{
  float64x1_t val[3];
} float64x1x3_t;

typedef struct poly8x8x3_t
{
  poly8x8_t val[3];
} poly8x8x3_t;

typedef struct poly8x16x3_t
{
  poly8x16_t val[3];
} poly8x16x3_t;

typedef struct poly16x4x3_t
{
  poly16x4_t val[3];
} poly16x4x3_t;

typedef struct poly16x8x3_t
{
  poly16x8_t val[3];
} poly16x8x3_t;

typedef struct int8x8x4_t
{
  int8x8_t val[4];
} int8x8x4_t;

typedef struct int8x16x4_t
{
  int8x16_t val[4];
} int8x16x4_t;

typedef struct int16x4x4_t
{
  int16x4_t val[4];
} int16x4x4_t;

typedef struct int16x8x4_t
{
  int16x8_t val[4];
} int16x8x4_t;

typedef struct int32x2x4_t
{
  int32x2_t val[4];
} int32x2x4_t;

typedef struct int32x4x4_t
{
  int32x4_t val[4];
} int32x4x4_t;

typedef struct int64x1x4_t
{
  int64x1_t val[4];
} int64x1x4_t;

typedef struct int64x2x4_t
{
  int64x2_t val[4];
} int64x2x4_t;

typedef struct uint8x8x4_t
{
  uint8x8_t val[4];
} uint8x8x4_t;

typedef struct uint8x16x4_t
{
  uint8x16_t val[4];
} uint8x16x4_t;

typedef struct uint16x4x4_t
{
  uint16x4_t val[4];
} uint16x4x4_t;

typedef struct uint16x8x4_t
{
  uint16x8_t val[4];
} uint16x8x4_t;

typedef struct uint32x2x4_t
{
  uint32x2_t val[4];
} uint32x2x4_t;

typedef struct uint32x4x4_t
{
  uint32x4_t val[4];
} uint32x4x4_t;

typedef struct uint64x1x4_t
{
  uint64x1_t val[4];
} uint64x1x4_t;

typedef struct uint64x2x4_t
{
  uint64x2_t val[4];
} uint64x2x4_t;

typedef struct float16x4x4_t
{
  float16x4_t val[4];
} float16x4x4_t;

typedef struct float16x8x4_t
{
  float16x8_t val[4];
} float16x8x4_t;

typedef struct float32x2x4_t
{
  float32x2_t val[4];
} float32x2x4_t;

typedef struct float32x4x4_t
{
  float32x4_t val[4];
} float32x4x4_t;

typedef struct float64x2x4_t
{
  float64x2_t val[4];
} float64x2x4_t;

typedef struct float64x1x4_t
{
  float64x1_t val[4];
} float64x1x4_t;

typedef struct poly8x8x4_t
{
  poly8x8_t val[4];
} poly8x8x4_t;

typedef struct poly8x16x4_t
{
  poly8x16_t val[4];
} poly8x16x4_t;

typedef struct poly16x4x4_t
{
  poly16x4_t val[4];
} poly16x4x4_t;

typedef struct poly16x8x4_t
{
  poly16x8_t val[4];
} poly16x8x4_t;

/* __aarch64_vdup_lane internal macros.  */
#define __aarch64_vdup_lane_any(__size, __q, __a, __b) \
  vdup##__q##_n_##__size (__aarch64_vget_lane_any (__a, __b))

#define __aarch64_vdup_lane_f16(__a, __b) \
   __aarch64_vdup_lane_any (f16, , __a, __b)
#define __aarch64_vdup_lane_f32(__a, __b) \
   __aarch64_vdup_lane_any (f32, , __a, __b)
#define __aarch64_vdup_lane_f64(__a, __b) \
   __aarch64_vdup_lane_any (f64, , __a, __b)
#define __aarch64_vdup_lane_p8(__a, __b) \
   __aarch64_vdup_lane_any (p8, , __a, __b)
#define __aarch64_vdup_lane_p16(__a, __b) \
   __aarch64_vdup_lane_any (p16, , __a, __b)
#define __aarch64_vdup_lane_p64(__a, __b) \
   __aarch64_vdup_lane_any (p64, , __a, __b)
#define __aarch64_vdup_lane_s8(__a, __b) \
   __aarch64_vdup_lane_any (s8, , __a, __b)
#define __aarch64_vdup_lane_s16(__a, __b) \
   __aarch64_vdup_lane_any (s16, , __a, __b)
#define __aarch64_vdup_lane_s32(__a, __b) \
   __aarch64_vdup_lane_any (s32, , __a, __b)
#define __aarch64_vdup_lane_s64(__a, __b) \
  __aarch64_vdup_lane_any (s64, , __a, __b)
#define __aarch64_vdup_lane_u8(__a, __b) \
   __aarch64_vdup_lane_any (u8, , __a, __b)
#define __aarch64_vdup_lane_u16(__a, __b) \
   __aarch64_vdup_lane_any (u16, , __a, __b)
#define __aarch64_vdup_lane_u32(__a, __b) \
   __aarch64_vdup_lane_any (u32, , __a, __b)
#define __aarch64_vdup_lane_u64(__a, __b) \
   __aarch64_vdup_lane_any (u64, , __a, __b)

/* __aarch64_vdup_laneq internal macros.  */
#define __aarch64_vdup_laneq_f16(__a, __b) \
   __aarch64_vdup_lane_any (f16, , __a, __b)
#define __aarch64_vdup_laneq_f32(__a, __b) \
   __aarch64_vdup_lane_any (f32, , __a, __b)
#define __aarch64_vdup_laneq_f64(__a, __b) \
   __aarch64_vdup_lane_any (f64, , __a, __b)
#define __aarch64_vdup_laneq_p8(__a, __b) \
   __aarch64_vdup_lane_any (p8, , __a, __b)
#define __aarch64_vdup_laneq_p16(__a, __b) \
   __aarch64_vdup_lane_any (p16, , __a, __b)
#define __aarch64_vdup_laneq_p64(__a, __b) \
   __aarch64_vdup_lane_any (p64, , __a, __b)
#define __aarch64_vdup_laneq_s8(__a, __b) \
   __aarch64_vdup_lane_any (s8, , __a, __b)
#define __aarch64_vdup_laneq_s16(__a, __b) \
   __aarch64_vdup_lane_any (s16, , __a, __b)
#define __aarch64_vdup_laneq_s32(__a, __b) \
   __aarch64_vdup_lane_any (s32, , __a, __b)
#define __aarch64_vdup_laneq_s64(__a, __b) \
   __aarch64_vdup_lane_any (s64, , __a, __b)
#define __aarch64_vdup_laneq_u8(__a, __b) \
   __aarch64_vdup_lane_any (u8, , __a, __b)
#define __aarch64_vdup_laneq_u16(__a, __b) \
   __aarch64_vdup_lane_any (u16, , __a, __b)
#define __aarch64_vdup_laneq_u32(__a, __b) \
   __aarch64_vdup_lane_any (u32, , __a, __b)
#define __aarch64_vdup_laneq_u64(__a, __b) \
   __aarch64_vdup_lane_any (u64, , __a, __b)

/* __aarch64_vdupq_lane internal macros.  */
#define __aarch64_vdupq_lane_f16(__a, __b) \
   __aarch64_vdup_lane_any (f16, q, __a, __b)
#define __aarch64_vdupq_lane_f32(__a, __b) \
   __aarch64_vdup_lane_any (f32, q, __a, __b)
#define __aarch64_vdupq_lane_f64(__a, __b) \
   __aarch64_vdup_lane_any (f64, q, __a, __b)
#define __aarch64_vdupq_lane_p8(__a, __b) \
   __aarch64_vdup_lane_any (p8, q, __a, __b)
#define __aarch64_vdupq_lane_p16(__a, __b) \
   __aarch64_vdup_lane_any (p16, q, __a, __b)
#define __aarch64_vdupq_lane_p64(__a, __b) \
   __aarch64_vdup_lane_any (p64, q, __a, __b)
#define __aarch64_vdupq_lane_s8(__a, __b) \
   __aarch64_vdup_lane_any (s8, q, __a, __b)
#define __aarch64_vdupq_lane_s16(__a, __b) \
   __aarch64_vdup_lane_any (s16, q, __a, __b)
#define __aarch64_vdupq_lane_s32(__a, __b) \
   __aarch64_vdup_lane_any (s32, q, __a, __b)
#define __aarch64_vdupq_lane_s64(__a, __b) \
   __aarch64_vdup_lane_any (s64, q, __a, __b)
#define __aarch64_vdupq_lane_u8(__a, __b) \
   __aarch64_vdup_lane_any (u8, q, __a, __b)
#define __aarch64_vdupq_lane_u16(__a, __b) \
   __aarch64_vdup_lane_any (u16, q, __a, __b)
#define __aarch64_vdupq_lane_u32(__a, __b) \
   __aarch64_vdup_lane_any (u32, q, __a, __b)
#define __aarch64_vdupq_lane_u64(__a, __b) \
   __aarch64_vdup_lane_any (u64, q, __a, __b)

/* __aarch64_vdupq_laneq internal macros.  */
#define __aarch64_vdupq_laneq_f16(__a, __b) \
   __aarch64_vdup_lane_any (f16, q, __a, __b)
#define __aarch64_vdupq_laneq_f32(__a, __b) \
   __aarch64_vdup_lane_any (f32, q, __a, __b)
#define __aarch64_vdupq_laneq_f64(__a, __b) \
   __aarch64_vdup_lane_any (f64, q, __a, __b)
#define __aarch64_vdupq_laneq_p8(__a, __b) \
   __aarch64_vdup_lane_any (p8, q, __a, __b)
#define __aarch64_vdupq_laneq_p16(__a, __b) \
   __aarch64_vdup_lane_any (p16, q, __a, __b)
#define __aarch64_vdupq_laneq_p64(__a, __b) \
   __aarch64_vdup_lane_any (p64, q, __a, __b)
#define __aarch64_vdupq_laneq_s8(__a, __b) \
   __aarch64_vdup_lane_any (s8, q, __a, __b)
#define __aarch64_vdupq_laneq_s16(__a, __b) \
   __aarch64_vdup_lane_any (s16, q, __a, __b)
#define __aarch64_vdupq_laneq_s32(__a, __b) \
   __aarch64_vdup_lane_any (s32, q, __a, __b)
#define __aarch64_vdupq_laneq_s64(__a, __b) \
   __aarch64_vdup_lane_any (s64, q, __a, __b)
#define __aarch64_vdupq_laneq_u8(__a, __b) \
   __aarch64_vdup_lane_any (u8, q, __a, __b)
#define __aarch64_vdupq_laneq_u16(__a, __b) \
   __aarch64_vdup_lane_any (u16, q, __a, __b)
#define __aarch64_vdupq_laneq_u32(__a, __b) \
   __aarch64_vdup_lane_any (u32, q, __a, __b)
#define __aarch64_vdupq_laneq_u64(__a, __b) \
   __aarch64_vdup_lane_any (u64, q, __a, __b)

/* Internal macro for lane indices.  */

#define __AARCH64_NUM_LANES(__v) (sizeof (__v) / sizeof (__v[0]))
#define __AARCH64_LANE_CHECK(__vec, __idx)	\
	__builtin_aarch64_im_lane_boundsi (sizeof(__vec), sizeof(__vec[0]), __idx)

/* For big-endian, GCC's vector indices are the opposite way around
   to the architectural lane indices used by Neon intrinsics.  */
#ifdef __AARCH64EB__
#define __aarch64_lane(__vec, __idx) (__AARCH64_NUM_LANES (__vec) - 1 - __idx)
#else
#define __aarch64_lane(__vec, __idx) __idx
#endif

/* vget_lane internal macro.  */
#define __aarch64_vget_lane_any(__vec, __index)				\
  __extension__								\
  ({									\
    __AARCH64_LANE_CHECK (__vec, __index);				\
    __vec[__aarch64_lane (__vec, __index)];				\
  })

/* vset_lane and vld1_lane internal macro.  */
#define __aarch64_vset_lane_any(__elem, __vec, __index)			\
  __extension__								\
  ({									\
    __AARCH64_LANE_CHECK (__vec, __index);				\
    __vec[__aarch64_lane (__vec, __index)] = __elem;			\
    __vec;								\
  })

/* vadd  */
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_f32 (float32x2_t __a, float32x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_f64 (float64x1_t __a, float64x1_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vadd_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_f32 (float32x4_t __a, float32x4_t __b)
{
  return __a + __b;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_f64 (float64x2_t __a, float64x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a + __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_saddlv8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_saddlv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_saddlv2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uaddlv8qi ((int8x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uaddlv4hi ((int16x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t) __builtin_aarch64_uaddlv2si ((int32x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int16x8_t) __builtin_aarch64_saddl2v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int32x4_t) __builtin_aarch64_saddl2v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int64x2_t) __builtin_aarch64_saddl2v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uaddl2v16qi ((int8x16_t) __a,
						     (int8x16_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uaddl2v8hi ((int16x8_t) __a,
						    (int16x8_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddl_high_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint64x2_t) __builtin_aarch64_uaddl2v4si ((int32x4_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_s8 (int16x8_t __a, int8x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_saddwv8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_s16 (int32x4_t __a, int16x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_saddwv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_s32 (int64x2_t __a, int32x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_saddwv2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_u8 (uint16x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uaddwv8qi ((int16x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_u16 (uint32x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uaddwv4hi ((int32x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_u32 (uint64x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t) __builtin_aarch64_uaddwv2si ((int64x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_s8 (int16x8_t __a, int8x16_t __b)
{
  return (int16x8_t) __builtin_aarch64_saddw2v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_s16 (int32x4_t __a, int16x8_t __b)
{
  return (int32x4_t) __builtin_aarch64_saddw2v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_s32 (int64x2_t __a, int32x4_t __b)
{
  return (int64x2_t) __builtin_aarch64_saddw2v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_u8 (uint16x8_t __a, uint8x16_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uaddw2v16qi ((int16x8_t) __a,
						     (int8x16_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_u16 (uint32x4_t __a, uint16x8_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uaddw2v8hi ((int32x4_t) __a,
						    (int16x8_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddw_high_u32 (uint64x2_t __a, uint32x4_t __b)
{
  return (uint64x2_t) __builtin_aarch64_uaddw2v4si ((int64x2_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_shaddv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_shaddv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_shaddv2si (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_uhaddv8qi ((int8x8_t) __a,
						  (int8x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_uhaddv4hi ((int16x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_uhaddv2si ((int32x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t) __builtin_aarch64_shaddv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_shaddv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_shaddv4si (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t) __builtin_aarch64_uhaddv16qi ((int8x16_t) __a,
						    (int8x16_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uhaddv8hi ((int16x8_t) __a,
						   (int16x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uhaddv4si ((int32x4_t) __a,
						   (int32x4_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_srhaddv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_srhaddv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_srhaddv2si (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_urhaddv8qi ((int8x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_urhaddv4hi ((int16x4_t) __a,
						    (int16x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_urhaddv2si ((int32x2_t) __a,
						    (int32x2_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t) __builtin_aarch64_srhaddv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_srhaddv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_srhaddv4si (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t) __builtin_aarch64_urhaddv16qi ((int8x16_t) __a,
						     (int8x16_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_urhaddv8hi ((int16x8_t) __a,
						    (int16x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_urhaddv4si ((int32x4_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_addhnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_addhnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_addhnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_addhnv8hi ((int16x8_t) __a,
						  (int16x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_addhnv4si ((int32x4_t) __a,
						   (int32x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_addhnv2di ((int64x2_t) __a,
						   (int64x2_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_raddhnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_raddhnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_raddhnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_raddhnv8hi ((int16x8_t) __a,
						   (int16x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_raddhnv4si ((int32x4_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_raddhnv2di ((int64x2_t) __a,
						    (int64x2_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int8x16_t) __builtin_aarch64_addhn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int16x8_t) __builtin_aarch64_addhn2v4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return (int32x4_t) __builtin_aarch64_addhn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint8x16_t) __builtin_aarch64_addhn2v8hi ((int8x8_t) __a,
						    (int16x8_t) __b,
						    (int16x8_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint16x8_t) __builtin_aarch64_addhn2v4si ((int16x4_t) __a,
						    (int32x4_t) __b,
						    (int32x4_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return (uint32x4_t) __builtin_aarch64_addhn2v2di ((int32x2_t) __a,
						    (int64x2_t) __b,
						    (int64x2_t) __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int8x16_t) __builtin_aarch64_raddhn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int16x8_t) __builtin_aarch64_raddhn2v4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return (int32x4_t) __builtin_aarch64_raddhn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint8x16_t) __builtin_aarch64_raddhn2v8hi ((int8x8_t) __a,
						     (int16x8_t) __b,
						     (int16x8_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint16x8_t) __builtin_aarch64_raddhn2v4si ((int16x4_t) __a,
						     (int32x4_t) __b,
						     (int32x4_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vraddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return (uint32x4_t) __builtin_aarch64_raddhn2v2di ((int32x2_t) __a,
						     (int64x2_t) __b,
						     (int64x2_t) __c);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdiv_f32 (float32x2_t __a, float32x2_t __b)
{
  return __a / __b;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdiv_f64 (float64x1_t __a, float64x1_t __b)
{
  return __a / __b;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdivq_f32 (float32x4_t __a, float32x4_t __b)
{
  return __a / __b;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdivq_f64 (float64x2_t __a, float64x2_t __b)
{
  return __a / __b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a * __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a * __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a * __b;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_f32 (float32x2_t __a, float32x2_t __b)
{
  return __a * __b;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_f64 (float64x1_t __a, float64x1_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a * __b;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmul_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly8x8_t) __builtin_aarch64_pmulv8qi ((int8x8_t) __a,
						 (int8x8_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a * __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a * __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a * __b;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_f32 (float32x4_t __a, float32x4_t __b)
{
  return __a * __b;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_f64 (float64x2_t __a, float64x2_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a * __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a * __b;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmulq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (poly8x16_t) __builtin_aarch64_pmulv16qi ((int8x16_t) __a,
						   (int8x16_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vand_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a & __b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vandq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a & __b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorr_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a | __b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorrq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a | __b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veor_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
veorq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a ^ __b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbic_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbicq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a & ~__b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vorn_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vornq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a | ~__b;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_f32 (float32x2_t __a, float32x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_f64 (float64x1_t __a, float64x1_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_s64 (int64x1_t __a, int64x1_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsub_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_f32 (float32x4_t __a, float32x4_t __b)
{
  return __a - __b;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_f64 (float64x2_t __a, float64x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __a - __b;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __a - __b;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_ssublv8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_ssublv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_ssublv2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_usublv8qi ((int8x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_usublv4hi ((int16x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t) __builtin_aarch64_usublv2si ((int32x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int16x8_t) __builtin_aarch64_ssubl2v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int32x4_t) __builtin_aarch64_ssubl2v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int64x2_t) __builtin_aarch64_ssubl2v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint16x8_t) __builtin_aarch64_usubl2v16qi ((int8x16_t) __a,
						     (int8x16_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint32x4_t) __builtin_aarch64_usubl2v8hi ((int16x8_t) __a,
						    (int16x8_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubl_high_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint64x2_t) __builtin_aarch64_usubl2v4si ((int32x4_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_s8 (int16x8_t __a, int8x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_ssubwv8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_s16 (int32x4_t __a, int16x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_ssubwv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_s32 (int64x2_t __a, int32x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_ssubwv2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_u8 (uint16x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_usubwv8qi ((int16x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_u16 (uint32x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_usubwv4hi ((int32x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_u32 (uint64x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t) __builtin_aarch64_usubwv2si ((int64x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_s8 (int16x8_t __a, int8x16_t __b)
{
  return (int16x8_t) __builtin_aarch64_ssubw2v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_s16 (int32x4_t __a, int16x8_t __b)
{
  return (int32x4_t) __builtin_aarch64_ssubw2v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_s32 (int64x2_t __a, int32x4_t __b)
{
  return (int64x2_t) __builtin_aarch64_ssubw2v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_u8 (uint16x8_t __a, uint8x16_t __b)
{
  return (uint16x8_t) __builtin_aarch64_usubw2v16qi ((int16x8_t) __a,
						     (int8x16_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_u16 (uint32x4_t __a, uint16x8_t __b)
{
  return (uint32x4_t) __builtin_aarch64_usubw2v8hi ((int32x4_t) __a,
						    (int16x8_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubw_high_u32 (uint64x2_t __a, uint32x4_t __b)
{
  return (uint64x2_t) __builtin_aarch64_usubw2v4si ((int64x2_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_sqaddv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_sqaddv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_sqaddv2si (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t) {__builtin_aarch64_sqadddi (__a[0], __b[0])};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __builtin_aarch64_uqaddv8qi_uuu (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_aarch64_shsubv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_shsubv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_shsubv2si (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_uhsubv8qi ((int8x8_t) __a,
						  (int8x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_uhsubv4hi ((int16x4_t) __a,
						   (int16x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_uhsubv2si ((int32x2_t) __a,
						   (int32x2_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t) __builtin_aarch64_shsubv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_shsubv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_shsubv4si (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t) __builtin_aarch64_uhsubv16qi ((int8x16_t) __a,
						    (int8x16_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_uhsubv8hi ((int16x8_t) __a,
						   (int16x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vhsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_uhsubv4si ((int32x4_t) __a,
						   (int32x4_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_subhnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_subhnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_subhnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_subhnv8hi ((int16x8_t) __a,
						  (int16x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_subhnv4si ((int32x4_t) __a,
						   (int32x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_subhnv2di ((int64x2_t) __a,
						   (int64x2_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_rsubhnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_rsubhnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_rsubhnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t) __builtin_aarch64_rsubhnv8hi ((int16x8_t) __a,
						   (int16x8_t) __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t) __builtin_aarch64_rsubhnv4si ((int32x4_t) __a,
						    (int32x4_t) __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t) __builtin_aarch64_rsubhnv2di ((int64x2_t) __a,
						    (int64x2_t) __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int8x16_t) __builtin_aarch64_rsubhn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int16x8_t) __builtin_aarch64_rsubhn2v4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return (int32x4_t) __builtin_aarch64_rsubhn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint8x16_t) __builtin_aarch64_rsubhn2v8hi ((int8x8_t) __a,
						     (int16x8_t) __b,
						     (int16x8_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint16x8_t) __builtin_aarch64_rsubhn2v4si ((int16x4_t) __a,
						     (int32x4_t) __b,
						     (int32x4_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return (uint32x4_t) __builtin_aarch64_rsubhn2v2di ((int32x2_t) __a,
						     (int64x2_t) __b,
						     (int64x2_t) __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int8x16_t) __builtin_aarch64_subhn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int16x8_t) __builtin_aarch64_subhn2v4si (__a, __b, __c);;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return (int32x4_t) __builtin_aarch64_subhn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint8x16_t) __builtin_aarch64_subhn2v8hi ((int8x8_t) __a,
						    (int16x8_t) __b,
						    (int16x8_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint16x8_t) __builtin_aarch64_subhn2v4si ((int16x4_t) __a,
						    (int32x4_t) __b,
						    (int32x4_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return (uint32x4_t) __builtin_aarch64_subhn2v2di ((int32x2_t) __a,
						    (int64x2_t) __b,
						    (int64x2_t) __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __builtin_aarch64_uqaddv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __builtin_aarch64_uqaddv2si_uuu (__a, __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqadd_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t) {__builtin_aarch64_uqadddi_uuu (__a[0], __b[0])};
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t) __builtin_aarch64_sqaddv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_sqaddv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_sqaddv4si (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_sqaddv2di (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __builtin_aarch64_uqaddv16qi_uuu (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_uqaddv8hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_uqaddv4si_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqaddq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __builtin_aarch64_uqaddv2di_uuu (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t) __builtin_aarch64_sqsubv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_sqsubv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_sqsubv2si (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t) {__builtin_aarch64_sqsubdi (__a[0], __b[0])};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __builtin_aarch64_uqsubv8qi_uuu (__a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __builtin_aarch64_uqsubv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __builtin_aarch64_uqsubv2si_uuu (__a, __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsub_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t) {__builtin_aarch64_uqsubdi_uuu (__a[0], __b[0])};
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t) __builtin_aarch64_sqsubv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_sqsubv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_sqsubv4si (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t) __builtin_aarch64_sqsubv2di (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __builtin_aarch64_uqsubv16qi_uuu (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_uqsubv8hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_uqsubv4si_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqsubq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return __builtin_aarch64_uqsubv2di_uuu (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqneg_s8 (int8x8_t __a)
{
  return (int8x8_t) __builtin_aarch64_sqnegv8qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqneg_s16 (int16x4_t __a)
{
  return (int16x4_t) __builtin_aarch64_sqnegv4hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqneg_s32 (int32x2_t __a)
{
  return (int32x2_t) __builtin_aarch64_sqnegv2si (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqneg_s64 (int64x1_t __a)
{
  return (int64x1_t) {__builtin_aarch64_sqnegdi (__a[0])};
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqnegq_s8 (int8x16_t __a)
{
  return (int8x16_t) __builtin_aarch64_sqnegv16qi (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqnegq_s16 (int16x8_t __a)
{
  return (int16x8_t) __builtin_aarch64_sqnegv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqnegq_s32 (int32x4_t __a)
{
  return (int32x4_t) __builtin_aarch64_sqnegv4si (__a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabs_s8 (int8x8_t __a)
{
  return (int8x8_t) __builtin_aarch64_sqabsv8qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabs_s16 (int16x4_t __a)
{
  return (int16x4_t) __builtin_aarch64_sqabsv4hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabs_s32 (int32x2_t __a)
{
  return (int32x2_t) __builtin_aarch64_sqabsv2si (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabs_s64 (int64x1_t __a)
{
  return (int64x1_t) {__builtin_aarch64_sqabsdi (__a[0])};
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabsq_s8 (int8x16_t __a)
{
  return (int8x16_t) __builtin_aarch64_sqabsv16qi (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabsq_s16 (int16x8_t __a)
{
  return (int16x8_t) __builtin_aarch64_sqabsv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqabsq_s32 (int32x4_t __a)
{
  return (int32x4_t) __builtin_aarch64_sqabsv4si (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_sqdmulhv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_sqdmulhv2si (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_sqdmulhv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_sqdmulhv4si (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t) __builtin_aarch64_sqrdmulhv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t) __builtin_aarch64_sqrdmulhv2si (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t) __builtin_aarch64_sqrdmulhv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t) __builtin_aarch64_sqrdmulhv4si (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_s8 (uint64_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_s16 (uint64_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_s32 (uint64_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_s64 (uint64_t __a)
{
  return (int64x1_t) {__a};
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_f16 (uint64_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_f32 (uint64_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_u8 (uint64_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_u16 (uint64_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_u32 (uint64_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_u64 (uint64_t __a)
{
  return (uint64x1_t) {__a};
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_f64 (uint64_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_p8 (uint64_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_p16 (uint64_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcreate_p64 (uint64_t __a)
{
  return (poly64x1_t) __a;
}

/* vget_lane  */

__extension__ extern __inline float16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_f16 (float16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_f32 (float32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_f64 (float64x1_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_p8 (poly8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_p16 (poly16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_p64 (poly64x1_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_s8 (int8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_s16 (int16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_s32 (int32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_s64 (int64x1_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_u8 (uint8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_u16 (uint16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_u32 (uint32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_lane_u64 (uint64x1_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vgetq_lane  */

__extension__ extern __inline float16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_f16 (float16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_f32 (float32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_f64 (float64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_p8 (poly8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_p16 (poly16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_p64 (poly64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_s8 (int8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_s16 (int16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_s32 (int32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_s64 (int64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_u8 (uint8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_u16 (uint16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_u32 (uint32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vgetq_lane_u64 (uint64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vreinterpret  */

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_f16 (float16x4_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_f64 (float64x1_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_s8 (int8x8_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_s16 (int16x4_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_s32 (int32x2_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_s64 (int64x1_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_f32 (float32x2_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_u8 (uint8x8_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_u16 (uint16x4_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_u32 (uint32x2_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_u64 (uint64x1_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_p16 (poly16x4_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p8_p64 (poly64x1_t __a)
{
  return (poly8x8_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_f64 (float64x2_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_s8 (int8x16_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_s16 (int16x8_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_s32 (int32x4_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_s64 (int64x2_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_f16 (float16x8_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_f32 (float32x4_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_u8 (uint8x16_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_u16 (uint16x8_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_u32 (uint32x4_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_u64 (uint64x2_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_p16 (poly16x8_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_p64 (poly64x2_t __a)
{
  return (poly8x16_t) __a;
}

__extension__ extern __inline poly8x16_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p8_p128 (poly128_t __a)
{
  return (poly8x16_t)__a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_f16 (float16x4_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_f64 (float64x1_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_s8 (int8x8_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_s16 (int16x4_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_s32 (int32x2_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_s64 (int64x1_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_f32 (float32x2_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_u8 (uint8x8_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_u16 (uint16x4_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_u32 (uint32x2_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_u64 (uint64x1_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_p8 (poly8x8_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p16_p64 (poly64x1_t __a)
{
  return (poly16x4_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_f64 (float64x2_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_s8 (int8x16_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_s16 (int16x8_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_s32 (int32x4_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_s64 (int64x2_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_f16 (float16x8_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_f32 (float32x4_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_u8 (uint8x16_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_u16 (uint16x8_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_u32 (uint32x4_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_u64 (uint64x2_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_p8 (poly8x16_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_p64 (poly64x2_t __a)
{
  return (poly16x8_t) __a;
}

__extension__ extern __inline poly16x8_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p16_p128 (poly128_t __a)
{
  return (poly16x8_t)__a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_f16 (float16x4_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_f64 (float64x1_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_s8 (int8x8_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_s16 (int16x4_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_s32 (int32x2_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_s64 (int64x1_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_f32 (float32x2_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_u8 (uint8x8_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_u16 (uint16x4_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_u32 (uint32x2_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_u64 (uint64x1_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_p8 (poly8x8_t __a)
{
  return (poly64x1_t) __a;
}

__extension__ extern __inline poly64x1_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_p64_p16 (poly16x4_t __a)
{
  return (poly64x1_t)__a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_f64 (float64x2_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_s8 (int8x16_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_s16 (int16x8_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_s32 (int32x4_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_s64 (int64x2_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_f16 (float16x8_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_f32 (float32x4_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_p128 (poly128_t __a)
{
  return (poly64x2_t)__a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_u8 (uint8x16_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_u16 (uint16x8_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_p16 (poly16x8_t __a)
{
  return (poly64x2_t)__a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_u32 (uint32x4_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_u64 (uint64x2_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p64_p8 (poly8x16_t __a)
{
  return (poly64x2_t) __a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_p8 (poly8x16_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_p16 (poly16x8_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_f16 (float16x8_t __a)
{
  return (poly128_t) __a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_f32 (float32x4_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_p64 (poly64x2_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_s64 (int64x2_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_u64 (uint64x2_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_s8 (int8x16_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_s16 (int16x8_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_s32 (int32x4_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_u8 (uint8x16_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_u16 (uint16x8_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_u32 (uint32x4_t __a)
{
  return (poly128_t)__a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_f64 (float64x1_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_s8 (int8x8_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_s16 (int16x4_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_s32 (int32x2_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_s64 (int64x1_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_f32 (float32x2_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_u8 (uint8x8_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_u16 (uint16x4_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_u32 (uint32x2_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_u64 (uint64x1_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_p8 (poly8x8_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_p16 (poly16x4_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f16_p64 (poly64x1_t __a)
{
  return (float16x4_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_f64 (float64x2_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_s8 (int8x16_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_s16 (int16x8_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_s32 (int32x4_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_s64 (int64x2_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_f32 (float32x4_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_u8 (uint8x16_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_u16 (uint16x8_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_u32 (uint32x4_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_u64 (uint64x2_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_p8 (poly8x16_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_p128 (poly128_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_p16 (poly16x8_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f16_p64 (poly64x2_t __a)
{
  return (float16x8_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_f16 (float16x4_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_f64 (float64x1_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_s8 (int8x8_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_s16 (int16x4_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_s32 (int32x2_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_s64 (int64x1_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_u8 (uint8x8_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_u16 (uint16x4_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_u32 (uint32x2_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_u64 (uint64x1_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_p8 (poly8x8_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_p16 (poly16x4_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f32_p64 (poly64x1_t __a)
{
  return (float32x2_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_f16 (float16x8_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_f64 (float64x2_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_s8 (int8x16_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_s16 (int16x8_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_s32 (int32x4_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_s64 (int64x2_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_u8 (uint8x16_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_u16 (uint16x8_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_u32 (uint32x4_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_u64 (uint64x2_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_p8 (poly8x16_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_p16 (poly16x8_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_p64 (poly64x2_t __a)
{
  return (float32x4_t) __a;
}

__extension__ extern __inline float32x4_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f32_p128 (poly128_t __a)
{
  return (float32x4_t)__a;
}


__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_f16 (float16x4_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_f32 (float32x2_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_p8 (poly8x8_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_p16 (poly16x4_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_p64 (poly64x1_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_s8 (int8x8_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_s16 (int16x4_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_s32 (int32x2_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_s64 (int64x1_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_u8 (uint8x8_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_u16 (uint16x4_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_u32 (uint32x2_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_f64_u64 (uint64x1_t __a)
{
  return (float64x1_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_f16 (float16x8_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_f32 (float32x4_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_p8 (poly8x16_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_p16 (poly16x8_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_p64 (poly64x2_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_s8 (int8x16_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_s16 (int16x8_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_s32 (int32x4_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_s64 (int64x2_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_u8 (uint8x16_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_u16 (uint16x8_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_u32 (uint32x4_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_u64 (uint64x2_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_f16 (float16x4_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_s8 (int8x8_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_s16 (int16x4_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_s32 (int32x2_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_f32 (float32x2_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_u8 (uint8x8_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_u16 (uint16x4_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_u32 (uint32x2_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_u64 (uint64x1_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_p8 (poly8x8_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_p16 (poly16x4_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s64_p64 (poly64x1_t __a)
{
  return (int64x1_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_f64 (float64x2_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_s8 (int8x16_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_s16 (int16x8_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_s32 (int32x4_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_f16 (float16x8_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_f32 (float32x4_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_u8 (uint8x16_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_u16 (uint16x8_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_u32 (uint32x4_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_u64 (uint64x2_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_p8 (poly8x16_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_p16 (poly16x8_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_p64 (poly64x2_t __a)
{
  return (int64x2_t) __a;
}

__extension__ extern __inline int64x2_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s64_p128 (poly128_t __a)
{
  return (int64x2_t)__a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_f16 (float16x4_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_s8 (int8x8_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_s16 (int16x4_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_s32 (int32x2_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_s64 (int64x1_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_f32 (float32x2_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_u8 (uint8x8_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_u16 (uint16x4_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_u32 (uint32x2_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_p8 (poly8x8_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_p16 (poly16x4_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u64_p64 (poly64x1_t __a)
{
  return (uint64x1_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_f64 (float64x2_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_s8 (int8x16_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_s16 (int16x8_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_s32 (int32x4_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_s64 (int64x2_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_f16 (float16x8_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_f32 (float32x4_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_u8 (uint8x16_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_u16 (uint16x8_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_u32 (uint32x4_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_p8 (poly8x16_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_p16 (poly16x8_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_p64 (poly64x2_t __a)
{
  return (uint64x2_t) __a;
}

__extension__ extern __inline uint64x2_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u64_p128 (poly128_t __a)
{
  return (uint64x2_t)__a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_f16 (float16x4_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_f64 (float64x1_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_s16 (int16x4_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_s32 (int32x2_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_s64 (int64x1_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_f32 (float32x2_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_u8 (uint8x8_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_u16 (uint16x4_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_u32 (uint32x2_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_u64 (uint64x1_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_p8 (poly8x8_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_p16 (poly16x4_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s8_p64 (poly64x1_t __a)
{
  return (int8x8_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_f64 (float64x2_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_s16 (int16x8_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_s32 (int32x4_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_s64 (int64x2_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_f16 (float16x8_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_f32 (float32x4_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_u8 (uint8x16_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_u16 (uint16x8_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_u32 (uint32x4_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_u64 (uint64x2_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_p8 (poly8x16_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_p16 (poly16x8_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_p64 (poly64x2_t __a)
{
  return (int8x16_t) __a;
}

__extension__ extern __inline int8x16_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s8_p128 (poly128_t __a)
{
  return (int8x16_t)__a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_f16 (float16x4_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_f64 (float64x1_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_s8 (int8x8_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_s32 (int32x2_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_s64 (int64x1_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_f32 (float32x2_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_u8 (uint8x8_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_u16 (uint16x4_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_u32 (uint32x2_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_u64 (uint64x1_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_p8 (poly8x8_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_p16 (poly16x4_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s16_p64 (poly64x1_t __a)
{
  return (int16x4_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_f64 (float64x2_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_s8 (int8x16_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_s32 (int32x4_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_s64 (int64x2_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_f16 (float16x8_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_f32 (float32x4_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_u8 (uint8x16_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_u16 (uint16x8_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_u32 (uint32x4_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_u64 (uint64x2_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_p8 (poly8x16_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_p16 (poly16x8_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_p64 (poly64x2_t __a)
{
  return (int16x8_t) __a;
}

__extension__ extern __inline int16x8_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s16_p128 (poly128_t __a)
{
  return (int16x8_t)__a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_f16 (float16x4_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_f64 (float64x1_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_s8 (int8x8_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_s16 (int16x4_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_s64 (int64x1_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_f32 (float32x2_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_u8 (uint8x8_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_u16 (uint16x4_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_u32 (uint32x2_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_u64 (uint64x1_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_p8 (poly8x8_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_p16 (poly16x4_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_s32_p64 (poly64x1_t __a)
{
  return (int32x2_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_f64 (float64x2_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_s8 (int8x16_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_s16 (int16x8_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_s64 (int64x2_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_f16 (float16x8_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_f32 (float32x4_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_u8 (uint8x16_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_u16 (uint16x8_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_u32 (uint32x4_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_u64 (uint64x2_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_p8 (poly8x16_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_p16 (poly16x8_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_p64 (poly64x2_t __a)
{
  return (int32x4_t) __a;
}

__extension__ extern __inline int32x4_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_s32_p128 (poly128_t __a)
{
  return (int32x4_t)__a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_f16 (float16x4_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_f64 (float64x1_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_s8 (int8x8_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_s16 (int16x4_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_s32 (int32x2_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_s64 (int64x1_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_f32 (float32x2_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_u16 (uint16x4_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_u32 (uint32x2_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_u64 (uint64x1_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_p8 (poly8x8_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_p16 (poly16x4_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u8_p64 (poly64x1_t __a)
{
  return (uint8x8_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_f64 (float64x2_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_s8 (int8x16_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_s16 (int16x8_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_s32 (int32x4_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_s64 (int64x2_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_f16 (float16x8_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_f32 (float32x4_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_u16 (uint16x8_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_u32 (uint32x4_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_u64 (uint64x2_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_p8 (poly8x16_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_p16 (poly16x8_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_p64 (poly64x2_t __a)
{
  return (uint8x16_t) __a;
}

__extension__ extern __inline uint8x16_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u8_p128 (poly128_t __a)
{
  return (uint8x16_t)__a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_f16 (float16x4_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_f64 (float64x1_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_s8 (int8x8_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_s16 (int16x4_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_s32 (int32x2_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_s64 (int64x1_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_f32 (float32x2_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_u8 (uint8x8_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_u32 (uint32x2_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_u64 (uint64x1_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_p8 (poly8x8_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_p16 (poly16x4_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u16_p64 (poly64x1_t __a)
{
  return (uint16x4_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_f64 (float64x2_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_s8 (int8x16_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_s16 (int16x8_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_s32 (int32x4_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_s64 (int64x2_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_f16 (float16x8_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_f32 (float32x4_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_u8 (uint8x16_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_u32 (uint32x4_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_u64 (uint64x2_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_p8 (poly8x16_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_p16 (poly16x8_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_p64 (poly64x2_t __a)
{
  return (uint16x8_t) __a;
}

__extension__ extern __inline uint16x8_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u16_p128 (poly128_t __a)
{
  return (uint16x8_t)__a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_f16 (float16x4_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_f64 (float64x1_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_s8 (int8x8_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_s16 (int16x4_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_s32 (int32x2_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_s64 (int64x1_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_f32 (float32x2_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_u8 (uint8x8_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_u16 (uint16x4_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_u64 (uint64x1_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_p8 (poly8x8_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_p16 (poly16x4_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpret_u32_p64 (poly64x1_t __a)
{
  return (uint32x2_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_f64 (float64x2_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_s8 (int8x16_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_s16 (int16x8_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_s32 (int32x4_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_s64 (int64x2_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_f16 (float16x8_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_f32 (float32x4_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_u8 (uint8x16_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_u16 (uint16x8_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_u64 (uint64x2_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_p8 (poly8x16_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_p16 (poly16x8_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_p64 (poly64x2_t __a)
{
  return (uint32x4_t) __a;
}

__extension__ extern __inline uint32x4_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_u32_p128 (poly128_t __a)
{
  return (uint32x4_t)__a;
}

__extension__ extern __inline float64x2_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_f64_p128 (poly128_t __a)
{
  return (float64x2_t) __a;
}

__extension__ extern __inline poly128_t
__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
vreinterpretq_p128_f64 (float64x2_t __a)
{
  return (poly128_t) __a;
}

/* vset_lane  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_f16 (float16_t __elem, float16x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_f32 (float32_t __elem, float32x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_f64 (float64_t __elem, float64x1_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_p8 (poly8_t __elem, poly8x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_p16 (poly16_t __elem, poly16x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_p64 (poly64_t __elem, poly64x1_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_s8 (int8_t __elem, int8x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_s16 (int16_t __elem, int16x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_s32 (int32_t __elem, int32x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_s64 (int64_t __elem, int64x1_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_u8 (uint8_t __elem, uint8x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_u16 (uint16_t __elem, uint16x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_u32 (uint32_t __elem, uint32x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vset_lane_u64 (uint64_t __elem, uint64x1_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

/* vsetq_lane  */

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_f16 (float16_t __elem, float16x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_f32 (float32_t __elem, float32x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_f64 (float64_t __elem, float64x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_p8 (poly8_t __elem, poly8x16_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_p16 (poly16_t __elem, poly16x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_p64 (poly64_t __elem, poly64x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_s8 (int8_t __elem, int8x16_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_s16 (int16_t __elem, int16x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_s32 (int32_t __elem, int32x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_s64 (int64_t __elem, int64x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_u8 (uint8_t __elem, uint8x16_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_u16 (uint16_t __elem, uint16x8_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_u32 (uint32_t __elem, uint32x4_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsetq_lane_u64 (uint64_t __elem, uint64x2_t __vec, const int __index)
{
  return __aarch64_vset_lane_any (__elem, __vec, __index);
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_f16 (float16x8_t __a)
{
  return __builtin_aarch64_get_lowv8hf (__a);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_f32 (float32x4_t __a)
{
  return __builtin_aarch64_get_lowv4sf (__a);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_f64 (float64x2_t __a)
{
  return (float64x1_t) {__builtin_aarch64_get_lowv2df (__a)};
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_p8 (poly8x16_t __a)
{
  return (poly8x8_t) __builtin_aarch64_get_lowv16qi ((int8x16_t) __a);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_p16 (poly16x8_t __a)
{
  return (poly16x4_t) __builtin_aarch64_get_lowv8hi ((int16x8_t) __a);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_p64 (poly64x2_t __a)
{
  return (poly64x1_t) __builtin_aarch64_get_lowv2di ((int64x2_t) __a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_s8 (int8x16_t __a)
{
  return  __builtin_aarch64_get_lowv16qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_s16 (int16x8_t __a)
{
  return  __builtin_aarch64_get_lowv8hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_s32 (int32x4_t __a)
{
  return  __builtin_aarch64_get_lowv4si (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_s64 (int64x2_t __a)
{
  return  (int64x1_t) {__builtin_aarch64_get_lowv2di (__a)};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_u8 (uint8x16_t __a)
{
  return (uint8x8_t) __builtin_aarch64_get_lowv16qi ((int8x16_t) __a);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_u16 (uint16x8_t __a)
{
  return (uint16x4_t) __builtin_aarch64_get_lowv8hi ((int16x8_t) __a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_u32 (uint32x4_t __a)
{
  return (uint32x2_t) __builtin_aarch64_get_lowv4si ((int32x4_t) __a);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_low_u64 (uint64x2_t __a)
{
  return (uint64x1_t) {__builtin_aarch64_get_lowv2di ((int64x2_t) __a)};
}

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_f16 (float16x8_t __a)
{
  return __builtin_aarch64_get_highv8hf (__a);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_f32 (float32x4_t __a)
{
  return __builtin_aarch64_get_highv4sf (__a);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_f64 (float64x2_t __a)
{
  return (float64x1_t) {__builtin_aarch64_get_highv2df (__a)};
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_p8 (poly8x16_t __a)
{
  return (poly8x8_t) __builtin_aarch64_get_highv16qi ((int8x16_t) __a);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_p16 (poly16x8_t __a)
{
  return (poly16x4_t) __builtin_aarch64_get_highv8hi ((int16x8_t) __a);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_p64 (poly64x2_t __a)
{
  return (poly64x1_t) __builtin_aarch64_get_highv2di ((int64x2_t) __a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_s8 (int8x16_t __a)
{
  return  __builtin_aarch64_get_highv16qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_s16 (int16x8_t __a)
{
  return  __builtin_aarch64_get_highv8hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_s32 (int32x4_t __a)
{
  return  __builtin_aarch64_get_highv4si (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_s64 (int64x2_t __a)
{
  return  (int64x1_t) {__builtin_aarch64_get_highv2di (__a)};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_u8 (uint8x16_t __a)
{
  return (uint8x8_t) __builtin_aarch64_get_highv16qi ((int8x16_t) __a);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_u16 (uint16x8_t __a)
{
  return (uint16x4_t) __builtin_aarch64_get_highv8hi ((int16x8_t) __a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_u32 (uint32x4_t __a)
{
  return (uint32x2_t) __builtin_aarch64_get_highv4si ((int32x4_t) __a);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vget_high_u64 (uint64x2_t __a)
{
  return (uint64x1_t) {__builtin_aarch64_get_highv2di ((int64x2_t) __a)};
}


__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x16_t) __builtin_aarch64_combinev8qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x8_t) __builtin_aarch64_combinev4hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x4_t) __builtin_aarch64_combinev2si (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_s64 (int64x1_t __a, int64x1_t __b)
{
  return __builtin_aarch64_combinedi (__a[0], __b[0]);
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_f16 (float16x4_t __a, float16x4_t __b)
{
  return __builtin_aarch64_combinev4hf (__a, __b);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x4_t) __builtin_aarch64_combinev2sf (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x16_t) __builtin_aarch64_combinev8qi ((int8x8_t) __a,
						     (int8x8_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x8_t) __builtin_aarch64_combinev4hi ((int16x4_t) __a,
						     (int16x4_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x4_t) __builtin_aarch64_combinev2si ((int32x2_t) __a,
						     (int32x2_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x2_t) __builtin_aarch64_combinedi (__a[0], __b[0]);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_f64 (float64x1_t __a, float64x1_t __b)
{
  return __builtin_aarch64_combinedf (__a[0], __b[0]);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly8x16_t) __builtin_aarch64_combinev8qi ((int8x8_t) __a,
						     (int8x8_t) __b);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_p16 (poly16x4_t __a, poly16x4_t __b)
{
  return (poly16x8_t) __builtin_aarch64_combinev4hi ((int16x4_t) __a,
						     (int16x4_t) __b);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcombine_p64 (poly64x1_t __a, poly64x1_t __b)
{
  return (poly64x2_t) __builtin_aarch64_combinedi_ppp (__a[0], __b[0]);
}

/* Start of temporary inline asm implementations.  */

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_sabav8qi (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_sabav4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_sabav2si (__a, __b, __c);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return __builtin_aarch64_uabav8qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return __builtin_aarch64_uabav4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaba_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return __builtin_aarch64_uabav2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_s8 (int16x8_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_sabal2v16qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_sabal2v8hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_sabal2v4si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_u8 (uint16x8_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return __builtin_aarch64_uabal2v16qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_u16 (uint32x4_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return __builtin_aarch64_uabal2v8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_high_u32 (uint64x2_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return __builtin_aarch64_uabal2v4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_sabalv8qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_sabalv4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_sabalv2si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return __builtin_aarch64_uabalv8qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return __builtin_aarch64_uabalv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabal_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return __builtin_aarch64_uabalv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_sabav16qi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_sabav8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_sabav4si (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return __builtin_aarch64_uabav16qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return __builtin_aarch64_uabav8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabaq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return __builtin_aarch64_uabav4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_s8 (int8x8_t __a, int8x8_t __b)
{
  return  __builtin_aarch64_sabdv8qi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_s16 (int16x4_t __a, int16x4_t __b)
{
  return  __builtin_aarch64_sabdv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_s32 (int32x2_t __a, int32x2_t __b)
{
  return  __builtin_aarch64_sabdv2si (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return  __builtin_aarch64_uabdv8qi_uuu (__a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return  __builtin_aarch64_uabdv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return  __builtin_aarch64_uabdv2si_uuu (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_s8 (int8x16_t __a, int8x16_t __b)
{
  return __builtin_aarch64_sabdl2v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_s16 (int16x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_sabdl2v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_s32 (int32x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_sabdl2v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __builtin_aarch64_uabdl2v16qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_uabdl2v8hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_high_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_uabdl2v4si_uuu (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_s8 (int8x8_t __a, int8x8_t __b)
{
  return __builtin_aarch64_sabdlv8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_s16 (int16x4_t __a, int16x4_t __b)
{
  return __builtin_aarch64_sabdlv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_s32 (int32x2_t __a, int32x2_t __b)
{
  return __builtin_aarch64_sabdlv2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __builtin_aarch64_uabdlv8qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __builtin_aarch64_uabdlv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __builtin_aarch64_uabdlv2si_uuu (__a, __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_s8 (int8x16_t __a, int8x16_t __b)
{
  return  __builtin_aarch64_sabdv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_s16 (int16x8_t __a, int16x8_t __b)
{
  return  __builtin_aarch64_sabdv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_s32 (int32x4_t __a, int32x4_t __b)
{
  return  __builtin_aarch64_sabdv4si (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return  __builtin_aarch64_uabdv16qi_uuu (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return  __builtin_aarch64_uabdv8hi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return  __builtin_aarch64_uabdv4si_uuu (__a, __b);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_s8 (int8x8_t __a)
{
  return __builtin_aarch64_saddlvv8qi (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_s16 (int16x4_t __a)
{
  return __builtin_aarch64_saddlvv4hi (__a);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_u8 (uint8x8_t __a)
{
  return __builtin_aarch64_uaddlvv8qi_uu (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_u16 (uint16x4_t __a)
{
  return __builtin_aarch64_uaddlvv4hi_uu (__a);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_saddlvv16qi (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_saddlvv8hi (__a);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_saddlvv4si (__a);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_u8 (uint8x16_t __a)
{
  return __builtin_aarch64_uaddlvv16qi_uu (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_u16 (uint16x8_t __a)
{
  return __builtin_aarch64_uaddlvv8hi_uu (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlvq_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_uaddlvv4si_uu (__a);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtx_f32_f64 (float64x2_t __a)
{
  return __builtin_aarch64_float_trunc_rodd_lo_v2sf (__a);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtx_high_f32_f64 (float32x2_t __a, float64x2_t __b)
{
  return __builtin_aarch64_float_trunc_rodd_hi_v4sf (__a, __b);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtxd_f32_f64 (float64_t __a)
{
  return __builtin_aarch64_float_trunc_rodd_df (__a);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
{
  return __builtin_aarch64_float_mla_nv2sf (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_n_s16 (int16x4_t __a, int16x4_t __b, int16_t __c)
{
  return __builtin_aarch64_mla_nv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_n_s32 (int32x2_t __a, int32x2_t __b, int32_t __c)
{
  return __builtin_aarch64_mla_nv2si (__a, __b, __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_n_u16 (uint16x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint16x4_t) __builtin_aarch64_mla_nv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b,
                                                   (int16_t) __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_n_u32 (uint32x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint32x2_t) __builtin_aarch64_mla_nv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b,
                                                   (int32_t) __c);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_mlav8qi (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_mlav4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_mlav2si (__a, __b, __c);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t) __builtin_aarch64_mlav8qi ((int8x8_t) __a,
                                                (int8x8_t) __b,
                                                (int8x8_t) __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t) __builtin_aarch64_mlav4hi ((int16x4_t) __a,
                                                 (int16x4_t) __b,
                                                 (int16x4_t) __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmla_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t) __builtin_aarch64_mlav2si ((int32x2_t) __a,
                                                 (int32x2_t) __b,
                                                 (int32x2_t) __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_lane_s16(int32x4_t __a, int16x8_t __b, int16x4_t __v,
		    const int __lane)
{
  return __builtin_aarch64_smlal_hi_lanev8hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_lane_s32(int64x2_t __a, int32x4_t __b, int32x2_t __v,
		    const int __lane)
{
  return __builtin_aarch64_smlal_hi_lanev4si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_lane_u16(uint32x4_t __a, uint16x8_t __b, uint16x4_t __v,
		    const int __lane)
{
  return __builtin_aarch64_umlal_hi_lanev8hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_lane_u32(uint64x2_t __a, uint32x4_t __b, uint32x2_t __v,
		    const int __lane)
{
  return __builtin_aarch64_umlal_hi_lanev4si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_laneq_s16(int32x4_t __a, int16x8_t __b, int16x8_t __v,
		     const int __lane)
{
  return __builtin_aarch64_smlal_hi_laneqv8hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_laneq_s32(int64x2_t __a, int32x4_t __b, int32x4_t __v,
		     const int __lane)
{
  return __builtin_aarch64_smlal_hi_laneqv4si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_laneq_u16(uint32x4_t __a, uint16x8_t __b, uint16x8_t __v,
		     const int __lane)
{
  return __builtin_aarch64_umlal_hi_laneqv8hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_laneq_u32(uint64x2_t __a, uint32x4_t __b, uint32x4_t __v,
		     const int __lane)
{
  return __builtin_aarch64_umlal_hi_laneqv4si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_n_s16 (int32x4_t __a, int16x8_t __b, int16_t __c)
{
  return __builtin_aarch64_smlal_hi_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_n_s32 (int64x2_t __a, int32x4_t __b, int32_t __c)
{
  return __builtin_aarch64_smlal_hi_nv4si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_n_u16 (uint32x4_t __a, uint16x8_t __b, uint16_t __c)
{
  return __builtin_aarch64_umlal_hi_nv8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_n_u32 (uint64x2_t __a, uint32x4_t __b, uint32_t __c)
{
  return __builtin_aarch64_umlal_hi_nv4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_s8 (int16x8_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_smlal_hiv16qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_smlal_hiv8hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_smlal_hiv4si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_u8 (uint16x8_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return __builtin_aarch64_umlal_hiv16qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_u16 (uint32x4_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return __builtin_aarch64_umlal_hiv8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_high_u32 (uint64x2_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return __builtin_aarch64_umlal_hiv4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_lane_s16 (int32x4_t __acc, int16x4_t __a, int16x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_smlal_lane_v4hi (__acc, __a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_lane_s32 (int64x2_t __acc, int32x2_t __a, int32x2_t __b, const int __c)
{
  return __builtin_aarch64_vec_smlal_lane_v2si (__acc, __a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_lane_u16 (uint32x4_t __acc, uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_umlal_lane_v4hi_uuuus (__acc, __a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_lane_u32 (uint64x2_t __acc, uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return __builtin_aarch64_vec_umlal_lane_v2si_uuuus (__acc, __a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_laneq_s16 (int32x4_t __acc, int16x4_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_vec_smlal_laneq_v4hi (__acc, __a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_laneq_s32 (int64x2_t __acc, int32x2_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_smlal_laneq_v2si (__acc, __a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_laneq_u16 (uint32x4_t __acc, uint16x4_t __a, uint16x8_t __b, const int __c)
{
  return __builtin_aarch64_vec_umlal_laneq_v4hi_uuuus (__acc, __a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_laneq_u32 (uint64x2_t __acc, uint32x2_t __a, uint32x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_umlal_laneq_v2si_uuuus (__acc, __a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return __builtin_aarch64_smlal_nv4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return __builtin_aarch64_smlal_nv2si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_n_u16 (uint32x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return __builtin_aarch64_umlal_nv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_n_u32 (uint64x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return __builtin_aarch64_umlal_nv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_smlalv8qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_smlalv4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_smlalv2si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return __builtin_aarch64_umlalv8qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return __builtin_aarch64_umlalv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlal_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return __builtin_aarch64_umlalv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
{
  return __builtin_aarch64_float_mla_nv4sf (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_n_s16 (int16x8_t __a, int16x8_t __b, int16_t __c)
{
  return __builtin_aarch64_mla_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_n_s32 (int32x4_t __a, int32x4_t __b, int32_t __c)
{
  return __builtin_aarch64_mla_nv4si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_n_u16 (uint16x8_t __a, uint16x8_t __b, uint16_t __c)
{
  return (uint16x8_t) __builtin_aarch64_mla_nv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b,
                                                   (int16_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_n_u32 (uint32x4_t __a, uint32x4_t __b, uint32_t __c)
{
  return (uint32x4_t) __builtin_aarch64_mla_nv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b,
                                                   (int32_t) __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_mlav16qi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_mlav8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_mlav4si (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t) __builtin_aarch64_mlav16qi ((int8x16_t) __a,
                                                  (int8x16_t) __b,
                                                  (int8x16_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t) __builtin_aarch64_mlav8hi ((int16x8_t) __a,
                                                 (int16x8_t) __b,
                                                 (int16x8_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlaq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t) __builtin_aarch64_mlav4si ((int32x4_t) __a,
                                                 (int32x4_t) __b,
                                                 (int32x4_t) __c);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
{
  return __builtin_aarch64_float_mls_nv2sf (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_n_s16 (int16x4_t __a, int16x4_t __b, int16_t __c)
{
  return __builtin_aarch64_mls_nv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_n_s32 (int32x2_t __a, int32x2_t __b, int32_t __c)
{
  return __builtin_aarch64_mls_nv2si (__a, __b, __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_n_u16 (uint16x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint16x4_t) __builtin_aarch64_mls_nv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b,
                                                   (int16_t) __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_n_u32 (uint32x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint32x2_t) __builtin_aarch64_mls_nv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b,
                                                   (int32_t) __c);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_mlsv8qi (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_mlsv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_mlsv2si (__a, __b, __c);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t) __builtin_aarch64_mlsv8qi ((int8x8_t) __a,
                                                (int8x8_t) __b,
                                                (int8x8_t) __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t) __builtin_aarch64_mlsv4hi ((int16x4_t) __a,
                                                 (int16x4_t) __b,
                                                 (int16x4_t) __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmls_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t) __builtin_aarch64_mlsv2si ((int32x2_t) __a,
                                                 (int32x2_t) __b,
                                                 (int32x2_t) __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_lane_s16(int32x4_t __a, int16x8_t __b, int16x4_t __v,
		    const int __lane)
{
  return __builtin_aarch64_smlsl_hi_lanev8hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_lane_s32(int64x2_t __a, int32x4_t __b, int32x2_t __v,
		    const int __lane)
{
  return __builtin_aarch64_smlsl_hi_lanev4si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_lane_u16(uint32x4_t __a, uint16x8_t __b, uint16x4_t __v,
		    const int __lane)
{
  return __builtin_aarch64_umlsl_hi_lanev8hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_lane_u32(uint64x2_t __a, uint32x4_t __b, uint32x2_t __v,
		    const int __lane)
{
  return __builtin_aarch64_umlsl_hi_lanev4si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_laneq_s16(int32x4_t __a, int16x8_t __b, int16x8_t __v,
		     const int __lane)
{
  return __builtin_aarch64_smlsl_hi_laneqv8hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_laneq_s32(int64x2_t __a, int32x4_t __b, int32x4_t __v,
		     const int __lane)
{
  return __builtin_aarch64_smlsl_hi_laneqv4si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_laneq_u16(uint32x4_t __a, uint16x8_t __b, uint16x8_t __v,
		     const int __lane)
{
  return __builtin_aarch64_umlsl_hi_laneqv8hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_laneq_u32(uint64x2_t __a, uint32x4_t __b, uint32x4_t __v,
		     const int __lane)
{
  return __builtin_aarch64_umlsl_hi_laneqv4si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_n_s16 (int32x4_t __a, int16x8_t __b, int16_t __c)
{
  return __builtin_aarch64_smlsl_hi_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_n_s32 (int64x2_t __a, int32x4_t __b, int32_t __c)
{
  return __builtin_aarch64_smlsl_hi_nv4si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_n_u16 (uint32x4_t __a, uint16x8_t __b, uint16_t __c)
{
  return __builtin_aarch64_umlsl_hi_nv8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_n_u32 (uint64x2_t __a, uint32x4_t __b, uint32_t __c)
{
  return __builtin_aarch64_umlsl_hi_nv4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_s8 (int16x8_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_smlsl_hiv16qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_smlsl_hiv8hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_smlsl_hiv4si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_u8 (uint16x8_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return __builtin_aarch64_umlsl_hiv16qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_u16 (uint32x4_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return __builtin_aarch64_umlsl_hiv8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_high_u32 (uint64x2_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return __builtin_aarch64_umlsl_hiv4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __v, const int __lane)
{
  return __builtin_aarch64_vec_smlsl_lane_v4hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __v, const int __lane)
{
  return __builtin_aarch64_vec_smlsl_lane_v2si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_lane_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __v,
                const int __lane)
{
  return __builtin_aarch64_vec_umlsl_lane_v4hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_lane_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __v,
                const int __lane)
{
  return __builtin_aarch64_vec_umlsl_lane_v2si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_laneq_s16 (int32x4_t __a, int16x4_t __b, int16x8_t __v, const int __lane)
{
  return __builtin_aarch64_vec_smlsl_laneq_v4hi (__a, __b, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_laneq_s32 (int64x2_t __a, int32x2_t __b, int32x4_t __v, const int __lane)
{
  return __builtin_aarch64_vec_smlsl_laneq_v2si (__a, __b, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_laneq_u16 (uint32x4_t __a, uint16x4_t __b, uint16x8_t __v,
                 const int __lane)
{
  return __builtin_aarch64_vec_umlsl_laneq_v4hi_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_laneq_u32 (uint64x2_t __a, uint32x2_t __b, uint32x4_t __v,
                 const int __lane)
{
  return __builtin_aarch64_vec_umlsl_laneq_v2si_uuuus (__a, __b, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return __builtin_aarch64_smlsl_nv4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return __builtin_aarch64_smlsl_nv2si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_n_u16 (uint32x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return __builtin_aarch64_umlsl_nv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_n_u32 (uint64x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return __builtin_aarch64_umlsl_nv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_smlslv8qi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_smlslv4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_smlslv2si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return __builtin_aarch64_umlslv8qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return __builtin_aarch64_umlslv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsl_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return __builtin_aarch64_umlslv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
{
  return __builtin_aarch64_float_mls_nv4sf (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_n_s16 (int16x8_t __a, int16x8_t __b, int16_t __c)
{
  return __builtin_aarch64_mls_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_n_s32 (int32x4_t __a, int32x4_t __b, int32_t __c)
{
  return __builtin_aarch64_mls_nv4si (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_n_u16 (uint16x8_t __a, uint16x8_t __b, uint16_t __c)
{
  return (uint16x8_t) __builtin_aarch64_mls_nv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b,
                                                   (int16_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_n_u32 (uint32x4_t __a, uint32x4_t __b, uint32_t __c)
{
  return (uint32x4_t) __builtin_aarch64_mls_nv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b,
                                                   (int32_t) __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_mlsv16qi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_mlsv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_mlsv4si (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t) __builtin_aarch64_mlsv16qi ((int8x16_t) __a,
                                                  (int8x16_t) __b,
                                                  (int8x16_t) __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t) __builtin_aarch64_mlsv8hi ((int16x8_t) __a,
                                                 (int16x8_t) __b,
                                                 (int16x8_t) __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmlsq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t) __builtin_aarch64_mlsv4si ((int32x4_t) __a,
                                                 (int32x4_t) __b,
                                                 (int32x4_t) __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_s8 (int8x16_t __a)
{
  return __builtin_aarch64_vec_unpacks_hi_v16qi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_s16 (int16x8_t __a)
{
  return __builtin_aarch64_vec_unpacks_hi_v8hi (__a);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_s32 (int32x4_t __a)
{
  return __builtin_aarch64_vec_unpacks_hi_v4si (__a);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_u8 (uint8x16_t __a)
{
  return __builtin_aarch64_vec_unpacku_hi_v16qi_uu (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_u16 (uint16x8_t __a)
{
  return __builtin_aarch64_vec_unpacku_hi_v8hi_uu (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_high_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_vec_unpacku_hi_v4si_uu (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_s8 (int8x8_t __a)
{
  return __builtin_aarch64_sxtlv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_s16 (int16x4_t __a)
{
  return __builtin_aarch64_sxtlv4si (__a);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_s32 (int32x2_t __a)
{
  return __builtin_aarch64_sxtlv2di (__a);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_u8 (uint8x8_t __a)
{
  return __builtin_aarch64_uxtlv8hi_uu (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_u16 (uint16x4_t __a)
{
  return __builtin_aarch64_uxtlv4si_uu (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovl_u32 (uint32x2_t __a)
{
  return __builtin_aarch64_uxtlv2di_uu (__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_s16 (int8x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_xtn2v8hi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_s32 (int16x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_xtn2v4si (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_s64 (int32x2_t __a, int64x2_t __b)
{
  return __builtin_aarch64_xtn2v2di (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_u16 (uint8x8_t __a, uint16x8_t __b)
{
  return (uint8x16_t)
	   __builtin_aarch64_xtn2v8hi ((int8x8_t) __a, (int16x8_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_u32 (uint16x4_t __a, uint32x4_t __b)
{
  return (uint16x8_t)
	   __builtin_aarch64_xtn2v4si ((int16x4_t) __a, (int32x4_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_high_u64 (uint32x2_t __a, uint64x2_t __b)
{
  return (uint32x4_t)
	   __builtin_aarch64_xtn2v2di ((int32x2_t) __a, (int64x2_t) __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_s16 (int16x8_t __a)
{
  return __builtin_aarch64_xtnv8hi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_s32 (int32x4_t __a)
{
  return __builtin_aarch64_xtnv4si (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_s64 (int64x2_t __a)
{
  return __builtin_aarch64_xtnv2di (__a);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_u16 (uint16x8_t __a)
{
  return (uint8x8_t)__builtin_aarch64_xtnv8hi ((int16x8_t) __a);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_u32 (uint32x4_t __a)
{
  return (uint16x4_t) __builtin_aarch64_xtnv4si ((int32x4_t )__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmovn_u64 (uint64x2_t __a)
{
  return (uint32x2_t) __builtin_aarch64_xtnv2di ((int64x2_t) __a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_s16 (int16x8_t __a, const int __b)
{
  return __builtin_aarch64_shrnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_s32 (int32x4_t __a, const int __b)
{
  return __builtin_aarch64_shrnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_s64 (int64x2_t __a, const int __b)
{
  return __builtin_aarch64_shrnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_aarch64_shrnv8hi ((int16x8_t)__a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_aarch64_shrnv4si ((int32x4_t)__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_aarch64_shrnv2di ((int64x2_t)__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_lane_s16 (int16x8_t __a, int16x4_t __v, const int __lane)
{
  return __builtin_aarch64_smull_hi_lanev8hi (__a, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_lane_s32 (int32x4_t __a, int32x2_t __v, const int __lane)
{
  return __builtin_aarch64_smull_hi_lanev4si (__a, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_lane_u16 (uint16x8_t __a, uint16x4_t __v, const int __lane)
{
  return __builtin_aarch64_umull_hi_lanev8hi_uuus (__a, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_lane_u32 (uint32x4_t __a, uint32x2_t __v, const int __lane)
{
  return __builtin_aarch64_umull_hi_lanev4si_uuus (__a, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_laneq_s16 (int16x8_t __a, int16x8_t __v, const int __lane)
{
  return __builtin_aarch64_smull_hi_laneqv8hi (__a, __v, __lane);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_laneq_s32 (int32x4_t __a, int32x4_t __v, const int __lane)
{
  return __builtin_aarch64_smull_hi_laneqv4si (__a, __v, __lane);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_laneq_u16 (uint16x8_t __a, uint16x8_t __v, const int __lane)
{
  return __builtin_aarch64_umull_hi_laneqv8hi_uuus (__a, __v, __lane);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_laneq_u32 (uint32x4_t __a, uint32x4_t __v, const int __lane)
{
  return __builtin_aarch64_umull_hi_laneqv4si_uuus (__a, __v, __lane);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_n_s16 (int16x8_t __a, int16_t __b)
{
  return __builtin_aarch64_smull_hi_nv8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_n_s32 (int32x4_t __a, int32_t __b)
{
  return __builtin_aarch64_smull_hi_nv4si (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_n_u16 (uint16x8_t __a, uint16_t __b)
{
  return __builtin_aarch64_umull_hi_nv8hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_n_u32 (uint32x4_t __a, uint32_t __b)
{
 return __builtin_aarch64_umull_hi_nv4si_uuu (__a, __b);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (poly16x8_t) __builtin_aarch64_pmull_hiv16qi ((int8x16_t) __a,
						       (int8x16_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_s8 (int8x16_t __a, int8x16_t __b)
{
  return __builtin_aarch64_vec_widen_smult_hi_v16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_s16 (int16x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_vec_widen_smult_hi_v8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_s32 (int32x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_vec_widen_smult_hi_v4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return __builtin_aarch64_vec_widen_umult_hi_v16qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_vec_widen_umult_hi_v8hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_high_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_vec_widen_umult_hi_v4si_uuu (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_smult_lane_v4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return __builtin_aarch64_vec_smult_lane_v2si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_lane_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_umult_lane_v4hi_uuus (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_lane_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return __builtin_aarch64_vec_umult_lane_v2si_uuus (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_vec_smult_laneq_v4hi (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_smult_laneq_v2si (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_laneq_u16 (uint16x4_t __a, uint16x8_t __b, const int __c)
{
  return __builtin_aarch64_vec_umult_laneq_v4hi_uuus (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_laneq_u32 (uint32x2_t __a, uint32x4_t __b, const int __c)
{
  return __builtin_aarch64_vec_umult_laneq_v2si_uuus (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_n_s16 (int16x4_t __a, int16_t __b)
{
  return __builtin_aarch64_smull_nv4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_n_s32 (int32x2_t __a, int32_t __b)
{
  return __builtin_aarch64_smull_nv2si (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_n_u16 (uint16x4_t __a, uint16_t __b)
{
  return __builtin_aarch64_umull_nv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_n_u32 (uint32x2_t __a, uint32_t __b)
{
  return __builtin_aarch64_umull_nv2si_uuu (__a, __b);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly16x8_t) __builtin_aarch64_pmullv8qi ((int8x8_t) __a,
						   (int8x8_t) __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_s8 (int8x8_t __a, int8x8_t __b)
{
  return __builtin_aarch64_intrinsic_vec_smult_lo_v8qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_s16 (int16x4_t __a, int16x4_t __b)
{
  return __builtin_aarch64_intrinsic_vec_smult_lo_v4hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_s32 (int32x2_t __a, int32x2_t __b)
{
  return __builtin_aarch64_intrinsic_vec_smult_lo_v2si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return __builtin_aarch64_intrinsic_vec_umult_lo_v8qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return __builtin_aarch64_intrinsic_vec_umult_lo_v4hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vmull_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return __builtin_aarch64_intrinsic_vec_umult_lo_v2si_uuu (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_s8 (int16x4_t __a, int8x8_t __b)
{
  return __builtin_aarch64_sadalpv8qi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_s16 (int32x2_t __a, int16x4_t __b)
{
  return __builtin_aarch64_sadalpv4hi (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_s32 (int64x1_t __a, int32x2_t __b)
{
  return (int64x1_t) __builtin_aarch64_sadalpv2si (__a[0], __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_u8 (uint16x4_t __a, uint8x8_t __b)
{
  return __builtin_aarch64_uadalpv8qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_u16 (uint32x2_t __a, uint16x4_t __b)
{
  return __builtin_aarch64_uadalpv4hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadal_u32 (uint64x1_t __a, uint32x2_t __b)
{
  return (uint64x1_t) __builtin_aarch64_uadalpv2si_uuu (__a[0], __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_s8 (int16x8_t __a, int8x16_t __b)
{
  return __builtin_aarch64_sadalpv16qi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_s16 (int32x4_t __a, int16x8_t __b)
{
  return __builtin_aarch64_sadalpv8hi (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_s32 (int64x2_t __a, int32x4_t __b)
{
  return __builtin_aarch64_sadalpv4si (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_u8 (uint16x8_t __a, uint8x16_t __b)
{
  return __builtin_aarch64_uadalpv16qi_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_u16 (uint32x4_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_uadalpv8hi_uuu (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpadalq_u32 (uint64x2_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_uadalpv4si_uuu (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_s8 (int8x8_t __a)
{
  return __builtin_aarch64_saddlpv8qi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_s16 (int16x4_t __a)
{
  return __builtin_aarch64_saddlpv4hi (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_s32 (int32x2_t __a)
{
  return (int64x1_t) __builtin_aarch64_saddlpv2si (__a);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_u8 (uint8x8_t __a)
{
  return __builtin_aarch64_uaddlpv8qi_uu (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_u16 (uint16x4_t __a)
{
  return __builtin_aarch64_uaddlpv4hi_uu (__a);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddl_u32 (uint32x2_t __a)
{
  return (uint64x1_t) __builtin_aarch64_uaddlpv2si_uu (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_saddlpv16qi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_saddlpv8hi (__a);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_saddlpv4si (__a);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_u8 (uint8x16_t __a)
{
  return __builtin_aarch64_uaddlpv16qi_uu (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_u16 (uint16x8_t __a)
{
  return __builtin_aarch64_uaddlpv8hi_uu (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddlq_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_uaddlpv4si_uu (__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return __builtin_aarch64_addpv16qi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_addpv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_addpv4si (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_s64 (int64x2_t __a, int64x2_t __b)
{
  return __builtin_aarch64_addpv2di (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t) __builtin_aarch64_addpv16qi ((int8x16_t) __a,
						   (int8x16_t) __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t) __builtin_aarch64_addpv8hi ((int16x8_t) __a,
						  (int16x8_t) __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t) __builtin_aarch64_addpv4si ((int32x4_t) __a,
						  (int32x4_t) __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vpaddq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t) __builtin_aarch64_addpv2di ((int64x2_t) __a,
						  (int64x2_t) __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_n_s16 (int16x4_t __a, int16_t __b)
{
  return __builtin_aarch64_sqdmulh_nv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_n_s32 (int32x2_t __a, int32_t __b)
{
  return __builtin_aarch64_sqdmulh_nv2si (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_n_s16 (int16x8_t __a, int16_t __b)
{
  return __builtin_aarch64_sqdmulh_nv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_n_s32 (int32x4_t __a, int32_t __b)
{
  return __builtin_aarch64_sqdmulh_nv4si (__a, __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_s16 (int8x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_sqxtn2v8hi (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_s32 (int16x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_sqxtn2v4si (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_s64 (int32x2_t __a, int64x2_t __b)
{
  return __builtin_aarch64_sqxtn2v2di (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_u16 (uint8x8_t __a, uint16x8_t __b)
{
  return __builtin_aarch64_uqxtn2v8hi_uuu (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_u32 (uint16x4_t __a, uint32x4_t __b)
{
  return __builtin_aarch64_uqxtn2v4si_uuu (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovn_high_u64 (uint32x2_t __a, uint64x2_t __b)
{
  return __builtin_aarch64_uqxtn2v2di_uuu (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovun_high_s16 (uint8x8_t __a, int16x8_t __b)
{
  return __builtin_aarch64_sqxtun2v8hi_uus (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovun_high_s32 (uint16x4_t __a, int32x4_t __b)
{
  return __builtin_aarch64_sqxtun2v4si_uus (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqmovun_high_s64 (uint32x2_t __a, int64x2_t __b)
{
  return __builtin_aarch64_sqxtun2v2di_uus (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_n_s16 (int16x4_t __a, int16_t __b)
{
  return __builtin_aarch64_sqrdmulh_nv4hi (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_n_s32 (int32x2_t __a, int32_t __b)
{
  return __builtin_aarch64_sqrdmulh_nv2si (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_n_s16 (int16x8_t __a, int16_t __b)
{
  return __builtin_aarch64_sqrdmulh_nv8hi (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_n_s32 (int32x4_t __a, int32_t __b)
{
  return __builtin_aarch64_sqrdmulh_nv4si (__a, __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_s16 (int8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrn2_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_s32 (int16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrn2_nv4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_s64 (int32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrn2_nv2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_u16 (uint8x8_t __a, uint16x8_t __b, const int __c)
{
  return __builtin_aarch64_uqrshrn2_nv8hi_uuus (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_u32 (uint16x4_t __a, uint32x4_t __b, const int __c)
{
  return __builtin_aarch64_uqrshrn2_nv4si_uuus (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrn_high_n_u64 (uint32x2_t __a, uint64x2_t __b, const int __c)
{
  return __builtin_aarch64_uqrshrn2_nv2di_uuus (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrun_high_n_s16 (uint8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrun2_nv8hi_uuss (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrun_high_n_s32 (uint16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrun2_nv4si_uuss (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrshrun_high_n_s64 (uint32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_sqrshrun2_nv2di_uuss (__a, __b, __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_s16 (int8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqshrn2_nv8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_s32 (int16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqshrn2_nv4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_s64 (int32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_sqshrn2_nv2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_u16 (uint8x8_t __a, uint16x8_t __b, const int __c)
{
  return __builtin_aarch64_uqshrn2_nv8hi_uuus (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_u32 (uint16x4_t __a, uint32x4_t __b, const int __c)
{
  return __builtin_aarch64_uqshrn2_nv4si_uuus (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrn_high_n_u64 (uint32x2_t __a, uint64x2_t __b, const int __c)
{
  return __builtin_aarch64_uqshrn2_nv2di_uuus (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrun_high_n_s16 (uint8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqshrun2_nv8hi_uuss (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrun_high_n_s32 (uint16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqshrun2_nv4si_uuss (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqshrun_high_n_s64 (uint32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_sqshrun2_nv2di_uuss (__a, __b, __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_s16 (int8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_rshrn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_s32 (int16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_rshrn2v4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_s64 (int32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_rshrn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_u16 (uint8x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint8x16_t) __builtin_aarch64_rshrn2v8hi ((int8x8_t) __a,
						    (int16x8_t) __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_u32 (uint16x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint16x8_t) __builtin_aarch64_rshrn2v4si ((int16x4_t) __a,
						    (int32x4_t) __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_high_n_u64 (uint32x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint32x4_t) __builtin_aarch64_rshrn2v2di ((int32x2_t)__a,
						    (int64x2_t)__b, __c);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_s16 (int16x8_t __a, const int __b)
{
  return __builtin_aarch64_rshrnv8hi (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_s32 (int32x4_t __a, const int __b)
{
  return __builtin_aarch64_rshrnv4si (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_s64 (int64x2_t __a, const int __b)
{
  return __builtin_aarch64_rshrnv2di (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t) __builtin_aarch64_rshrnv8hi ((int16x8_t) __a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t) __builtin_aarch64_rshrnv4si ((int32x4_t) __a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t) __builtin_aarch64_rshrnv2di ((int64x2_t) __a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsqrte_u32 (uint32x2_t __a)
{
  return __builtin_aarch64_ursqrtev2si_uu (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsqrteq_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_ursqrtev4si_uu (__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_s16 (int8x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_shrn2v8hi (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_s32 (int16x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_shrn2v4si (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_s64 (int32x2_t __a, int64x2_t __b, const int __c)
{
  return __builtin_aarch64_shrn2v2di (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_u16 (uint8x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint8x16_t)
    __builtin_aarch64_shrn2v8hi ((int8x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_u32 (uint16x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint16x8_t)
    __builtin_aarch64_shrn2v4si ((int16x4_t) __a, (int32x4_t) __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vshrn_high_n_u64 (uint32x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint32x4_t)
    __builtin_aarch64_shrn2v2di ((int32x2_t) __a, (int64x2_t) __b, __c);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsli_n_p8 (poly8x8_t __a, poly8x8_t __b, const int __c)
{
  return __builtin_aarch64_ssli_nv8qi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsli_n_p16 (poly16x4_t __a, poly16x4_t __b, const int __c)
{
  return __builtin_aarch64_ssli_nv4hi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsliq_n_p8 (poly8x16_t __a, poly8x16_t __b, const int __c)
{
  return __builtin_aarch64_ssli_nv16qi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsliq_n_p16 (poly16x8_t __a, poly16x8_t __b, const int __c)
{
  return __builtin_aarch64_ssli_nv8hi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsri_n_p8 (poly8x8_t __a, poly8x8_t __b, const int __c)
{
  return __builtin_aarch64_ssri_nv8qi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsri_n_p16 (poly16x4_t __a, poly16x4_t __b, const int __c)
{
  return __builtin_aarch64_ssri_nv4hi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsri_n_p64 (poly64x1_t __a, poly64x1_t __b, const int __c)
{
  return (poly64x1_t) __builtin_aarch64_ssri_ndi_ppps (__a[0], __b[0], __c);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsriq_n_p8 (poly8x16_t __a, poly8x16_t __b, const int __c)
{
  return __builtin_aarch64_ssri_nv16qi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsriq_n_p16 (poly16x8_t __a, poly16x8_t __b, const int __c)
{
  return __builtin_aarch64_ssri_nv8hi_ppps (__a, __b, __c);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vsriq_n_p64 (poly64x2_t __a, poly64x2_t __b, const int __c)
{
  return __builtin_aarch64_ssri_nv2di_ppps (__a, __b, __c);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtst_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (uint8x8_t) ((((uint8x8_t) __a) & ((uint8x8_t) __b))
		       != 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtst_p16 (poly16x4_t __a, poly16x4_t __b)
{
  return (uint16x4_t) ((((uint16x4_t) __a) & ((uint16x4_t) __b))
		       != 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtst_p64 (poly64x1_t __a, poly64x1_t __b)
{
  return (uint64x1_t) ((__a & __b) != __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtstq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (uint8x16_t) ((((uint8x16_t) __a) & ((uint8x16_t) __b))
		       != 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtstq_p16 (poly16x8_t __a, poly16x8_t __b)
{
  return (uint16x8_t) ((((uint16x8_t) __a) & ((uint16x8_t) __b))
		       != 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtstq_p64 (poly64x2_t __a, poly64x2_t __b)
{
  return (uint64x2_t) ((((uint64x2_t) __a) & ((uint64x2_t) __b))
		       != __AARCH64_INT64_C (0));
}

/* End of temporary inline asm implementations.  */

/* Start of temporary inline asm for vldn, vstn and friends.  */

/* Create struct element types for duplicating loads.

   Create 2 element structures of:

   +------+----+----+----+----+
   |      | 8  | 16 | 32 | 64 |
   +------+----+----+----+----+
   |int   | Y  | Y  | N  | N  |
   +------+----+----+----+----+
   |uint  | Y  | Y  | N  | N  |
   +------+----+----+----+----+
   |float | -  | Y  | N  | N  |
   +------+----+----+----+----+
   |poly  | Y  | Y  | -  | -  |
   +------+----+----+----+----+

   Create 3 element structures of:

   +------+----+----+----+----+
   |      | 8  | 16 | 32 | 64 |
   +------+----+----+----+----+
   |int   | Y  | Y  | Y  | Y  |
   +------+----+----+----+----+
   |uint  | Y  | Y  | Y  | Y  |
   +------+----+----+----+----+
   |float | -  | Y  | Y  | Y  |
   +------+----+----+----+----+
   |poly  | Y  | Y  | -  | -  |
   +------+----+----+----+----+

   Create 4 element structures of:

   +------+----+----+----+----+
   |      | 8  | 16 | 32 | 64 |
   +------+----+----+----+----+
   |int   | Y  | N  | N  | Y  |
   +------+----+----+----+----+
   |uint  | Y  | N  | N  | Y  |
   +------+----+----+----+----+
   |float | -  | N  | N  | Y  |
   +------+----+----+----+----+
   |poly  | Y  | N  | -  | -  |
   +------+----+----+----+----+

  This is required for casting memory reference.  */
#define __STRUCTN(t, sz, nelem)			\
  typedef struct t ## sz ## x ## nelem ## _t {	\
    t ## sz ## _t val[nelem];			\
  }  t ## sz ## x ## nelem ## _t;

/* 2-element structs.  */
__STRUCTN (int, 8, 2)
__STRUCTN (int, 16, 2)
__STRUCTN (uint, 8, 2)
__STRUCTN (uint, 16, 2)
__STRUCTN (float, 16, 2)
__STRUCTN (poly, 8, 2)
__STRUCTN (poly, 16, 2)
/* 3-element structs.  */
__STRUCTN (int, 8, 3)
__STRUCTN (int, 16, 3)
__STRUCTN (int, 32, 3)
__STRUCTN (int, 64, 3)
__STRUCTN (uint, 8, 3)
__STRUCTN (uint, 16, 3)
__STRUCTN (uint, 32, 3)
__STRUCTN (uint, 64, 3)
__STRUCTN (float, 16, 3)
__STRUCTN (float, 32, 3)
__STRUCTN (float, 64, 3)
__STRUCTN (poly, 8, 3)
__STRUCTN (poly, 16, 3)
/* 4-element structs.  */
__STRUCTN (int, 8, 4)
__STRUCTN (int, 64, 4)
__STRUCTN (uint, 8, 4)
__STRUCTN (uint, 64, 4)
__STRUCTN (poly, 8, 4)
__STRUCTN (float, 64, 4)
#undef __STRUCTN


__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_f16 (float16_t *__ptr, float16x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  float16x8x2_t __temp;
  __temp.val[0] = vcombine_f16 (__val.val[0],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f16 (__val.val[1],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev4hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_f32 (float32_t *__ptr, float32x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  float32x4x2_t __temp;
  __temp.val[0] = vcombine_f32 (__val.val[0],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f32 (__val.val[1],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev2sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_f64 (float64_t *__ptr, float64x1x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  float64x2x2_t __temp;
  __temp.val[0] = vcombine_f64 (__val.val[0],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f64 (__val.val[1],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanedf ((__builtin_aarch64_simd_df *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_p8 (poly8_t *__ptr, poly8x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  poly8x16x2_t __temp;
  __temp.val[0] = vcombine_p8 (__val.val[0],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p8 (__val.val[1],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_p16 (poly16_t *__ptr, poly16x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  poly16x8x2_t __temp;
  __temp.val[0] = vcombine_p16 (__val.val[0],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p16 (__val.val[1],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_p64 (poly64_t *__ptr, poly64x1x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  poly64x2x2_t __temp;
  __temp.val[0] = vcombine_p64 (__val.val[0],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p64 (__val.val[1],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_s8 (int8_t *__ptr, int8x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  int8x16x2_t __temp;
  __temp.val[0] = vcombine_s8 (__val.val[0],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s8 (__val.val[1],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_s16 (int16_t *__ptr, int16x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  int16x8x2_t __temp;
  __temp.val[0] = vcombine_s16 (__val.val[0],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s16 (__val.val[1],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_s32 (int32_t *__ptr, int32x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  int32x4x2_t __temp;
  __temp.val[0] = vcombine_s32 (__val.val[0],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s32 (__val.val[1],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_s64 (int64_t *__ptr, int64x1x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  int64x2x2_t __temp;
  __temp.val[0] = vcombine_s64 (__val.val[0],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s64 (__val.val[1],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_u8 (uint8_t *__ptr, uint8x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  uint8x16x2_t __temp;
  __temp.val[0] = vcombine_u8 (__val.val[0],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u8 (__val.val[1],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_u16 (uint16_t *__ptr, uint16x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  uint16x8x2_t __temp;
  __temp.val[0] = vcombine_u16 (__val.val[0],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u16 (__val.val[1],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_u32 (uint32_t *__ptr, uint32x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  uint32x4x2_t __temp;
  __temp.val[0] = vcombine_u32 (__val.val[0],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u32 (__val.val[1],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2_lane_u64 (uint64_t *__ptr, uint64x1x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  uint64x2x2_t __temp;
  __temp.val[0] = vcombine_u64 (__val.val[0],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u64 (__val.val[1],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st2_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_f16 (float16_t *__ptr, float16x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev8hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_f32 (float32_t *__ptr, float32x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev4sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_f64 (float64_t *__ptr, float64x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev2df ((__builtin_aarch64_simd_df *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_p8 (poly8_t *__ptr, poly8x16x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_p16 (poly16_t *__ptr, poly16x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_p64 (poly64_t *__ptr, poly64x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_s8 (int8_t *__ptr, int8x16x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_s16 (int16_t *__ptr, int16x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_s32 (int32_t *__ptr, int32x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_s64 (int64_t *__ptr, int64x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_u8 (uint8_t *__ptr, uint8x16x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_u16 (uint16_t *__ptr, uint16x8x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_u32 (uint32_t *__ptr, uint32x4x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst2q_lane_u64 (uint64_t *__ptr, uint64x2x2_t __val, const int __lane)
{
  __builtin_aarch64_simd_oi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st2_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_f16 (float16_t *__ptr, float16x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  float16x8x3_t __temp;
  __temp.val[0] = vcombine_f16 (__val.val[0],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f16 (__val.val[1],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f16 (__val.val[2],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev4hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_f32 (float32_t *__ptr, float32x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  float32x4x3_t __temp;
  __temp.val[0] = vcombine_f32 (__val.val[0],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f32 (__val.val[1],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f32 (__val.val[2],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev2sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_f64 (float64_t *__ptr, float64x1x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  float64x2x3_t __temp;
  __temp.val[0] = vcombine_f64 (__val.val[0],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f64 (__val.val[1],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f64 (__val.val[2],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanedf ((__builtin_aarch64_simd_df *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_p8 (poly8_t *__ptr, poly8x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  poly8x16x3_t __temp;
  __temp.val[0] = vcombine_p8 (__val.val[0],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p8 (__val.val[1],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p8 (__val.val[2],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_p16 (poly16_t *__ptr, poly16x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  poly16x8x3_t __temp;
  __temp.val[0] = vcombine_p16 (__val.val[0],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p16 (__val.val[1],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p16 (__val.val[2],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_p64 (poly64_t *__ptr, poly64x1x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  poly64x2x3_t __temp;
  __temp.val[0] = vcombine_p64 (__val.val[0],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p64 (__val.val[1],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p64 (__val.val[2],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_s8 (int8_t *__ptr, int8x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  int8x16x3_t __temp;
  __temp.val[0] = vcombine_s8 (__val.val[0],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s8 (__val.val[1],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s8 (__val.val[2],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_s16 (int16_t *__ptr, int16x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  int16x8x3_t __temp;
  __temp.val[0] = vcombine_s16 (__val.val[0],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s16 (__val.val[1],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s16 (__val.val[2],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_s32 (int32_t *__ptr, int32x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  int32x4x3_t __temp;
  __temp.val[0] = vcombine_s32 (__val.val[0],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s32 (__val.val[1],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s32 (__val.val[2],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_s64 (int64_t *__ptr, int64x1x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  int64x2x3_t __temp;
  __temp.val[0] = vcombine_s64 (__val.val[0],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s64 (__val.val[1],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s64 (__val.val[2],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_u8 (uint8_t *__ptr, uint8x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  uint8x16x3_t __temp;
  __temp.val[0] = vcombine_u8 (__val.val[0],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u8 (__val.val[1],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u8 (__val.val[2],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_u16 (uint16_t *__ptr, uint16x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  uint16x8x3_t __temp;
  __temp.val[0] = vcombine_u16 (__val.val[0],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u16 (__val.val[1],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u16 (__val.val[2],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_u32 (uint32_t *__ptr, uint32x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  uint32x4x3_t __temp;
  __temp.val[0] = vcombine_u32 (__val.val[0],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u32 (__val.val[1],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u32 (__val.val[2],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3_lane_u64 (uint64_t *__ptr, uint64x1x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  uint64x2x3_t __temp;
  __temp.val[0] = vcombine_u64 (__val.val[0],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u64 (__val.val[1],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u64 (__val.val[2],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st3_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_f16 (float16_t *__ptr, float16x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev8hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_f32 (float32_t *__ptr, float32x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev4sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_f64 (float64_t *__ptr, float64x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev2df ((__builtin_aarch64_simd_df *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_p8 (poly8_t *__ptr, poly8x16x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_p16 (poly16_t *__ptr, poly16x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_p64 (poly64_t *__ptr, poly64x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_s8 (int8_t *__ptr, int8x16x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_s16 (int16_t *__ptr, int16x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_s32 (int32_t *__ptr, int32x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_s64 (int64_t *__ptr, int64x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_u8 (uint8_t *__ptr, uint8x16x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_u16 (uint16_t *__ptr, uint16x8x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_u32 (uint32_t *__ptr, uint32x4x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst3q_lane_u64 (uint64_t *__ptr, uint64x2x3_t __val, const int __lane)
{
  __builtin_aarch64_simd_ci __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st3_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_f16 (float16_t *__ptr, float16x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  float16x8x4_t __temp;
  __temp.val[0] = vcombine_f16 (__val.val[0],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f16 (__val.val[1],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f16 (__val.val[2],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_f16 (__val.val[3],
				vcreate_f16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev4hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_f32 (float32_t *__ptr, float32x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  float32x4x4_t __temp;
  __temp.val[0] = vcombine_f32 (__val.val[0],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f32 (__val.val[1],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f32 (__val.val[2],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_f32 (__val.val[3],
				vcreate_f32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev2sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_f64 (float64_t *__ptr, float64x1x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  float64x2x4_t __temp;
  __temp.val[0] = vcombine_f64 (__val.val[0],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_f64 (__val.val[1],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_f64 (__val.val[2],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_f64 (__val.val[3],
				vcreate_f64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanedf ((__builtin_aarch64_simd_df *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_p8 (poly8_t *__ptr, poly8x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  poly8x16x4_t __temp;
  __temp.val[0] = vcombine_p8 (__val.val[0],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p8 (__val.val[1],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p8 (__val.val[2],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_p8 (__val.val[3],
			       vcreate_p8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_p16 (poly16_t *__ptr, poly16x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  poly16x8x4_t __temp;
  __temp.val[0] = vcombine_p16 (__val.val[0],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p16 (__val.val[1],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p16 (__val.val[2],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_p16 (__val.val[3],
				vcreate_p16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_p64 (poly64_t *__ptr, poly64x1x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  poly64x2x4_t __temp;
  __temp.val[0] = vcombine_p64 (__val.val[0],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_p64 (__val.val[1],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_p64 (__val.val[2],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_p64 (__val.val[3],
				vcreate_p64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_s8 (int8_t *__ptr, int8x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  int8x16x4_t __temp;
  __temp.val[0] = vcombine_s8 (__val.val[0],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s8 (__val.val[1],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s8 (__val.val[2],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_s8 (__val.val[3],
			       vcreate_s8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_s16 (int16_t *__ptr, int16x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  int16x8x4_t __temp;
  __temp.val[0] = vcombine_s16 (__val.val[0],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s16 (__val.val[1],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s16 (__val.val[2],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_s16 (__val.val[3],
				vcreate_s16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_s32 (int32_t *__ptr, int32x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  int32x4x4_t __temp;
  __temp.val[0] = vcombine_s32 (__val.val[0],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s32 (__val.val[1],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s32 (__val.val[2],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_s32 (__val.val[3],
				vcreate_s32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_s64 (int64_t *__ptr, int64x1x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  int64x2x4_t __temp;
  __temp.val[0] = vcombine_s64 (__val.val[0],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_s64 (__val.val[1],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_s64 (__val.val[2],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_s64 (__val.val[3],
				vcreate_s64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_u8 (uint8_t *__ptr, uint8x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  uint8x16x4_t __temp;
  __temp.val[0] = vcombine_u8 (__val.val[0],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u8 (__val.val[1],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u8 (__val.val[2],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_u8 (__val.val[3],
			       vcreate_u8 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev8qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_u16 (uint16_t *__ptr, uint16x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  uint16x8x4_t __temp;
  __temp.val[0] = vcombine_u16 (__val.val[0],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u16 (__val.val[1],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u16 (__val.val[2],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_u16 (__val.val[3],
				vcreate_u16 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev4hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_u32 (uint32_t *__ptr, uint32x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  uint32x4x4_t __temp;
  __temp.val[0] = vcombine_u32 (__val.val[0],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u32 (__val.val[1],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u32 (__val.val[2],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_u32 (__val.val[3],
				vcreate_u32 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanev2si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4_lane_u64 (uint64_t *__ptr, uint64x1x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  uint64x2x4_t __temp;
  __temp.val[0] = vcombine_u64 (__val.val[0],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[1] = vcombine_u64 (__val.val[1],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[2] = vcombine_u64 (__val.val[2],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __temp.val[3] = vcombine_u64 (__val.val[3],
				vcreate_u64 (__AARCH64_UINT64_C (0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  __builtin_aarch64_st4_lanedi ((__builtin_aarch64_simd_di *) __ptr, __o,
				__lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_f16 (float16_t *__ptr, float16x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev8hf ((__builtin_aarch64_simd_hf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_f32 (float32_t *__ptr, float32x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev4sf ((__builtin_aarch64_simd_sf *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_f64 (float64_t *__ptr, float64x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev2df ((__builtin_aarch64_simd_df *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_p8 (poly8_t *__ptr, poly8x16x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_p16 (poly16_t *__ptr, poly16x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_p64 (poly64_t *__ptr, poly64x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_s8 (int8_t *__ptr, int8x16x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_s16 (int16_t *__ptr, int16x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_s32 (int32_t *__ptr, int32x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_s64 (int64_t *__ptr, int64x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_u8 (uint8_t *__ptr, uint8x16x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev16qi ((__builtin_aarch64_simd_qi *) __ptr, __o,
				   __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_u16 (uint16_t *__ptr, uint16x8x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev8hi ((__builtin_aarch64_simd_hi *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_u32 (uint32_t *__ptr, uint32x4x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev4si ((__builtin_aarch64_simd_si *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst4q_lane_u64 (uint64_t *__ptr, uint64x2x4_t __val, const int __lane)
{
  __builtin_aarch64_simd_xi __o;
  __builtin_memcpy (&__o, &__val, sizeof (__val));
  __builtin_aarch64_st4_lanev2di ((__builtin_aarch64_simd_di *) __ptr, __o,
				  __lane);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_s32 (int32x2_t __a)
{
  return __builtin_aarch64_saddlvv2si (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddlv_u32 (uint32x2_t __a)
{
  return __builtin_aarch64_uaddlvv2si_uu (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqdmulh_laneqv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulh_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqdmulh_laneqv2si (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_laneq_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqdmulh_laneqv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqdmulh_laneqv4si (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __c)
{
  return  __builtin_aarch64_sqrdmulh_laneqv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulh_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqrdmulh_laneqv2si (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_laneq_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return __builtin_aarch64_sqrdmulh_laneqv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return __builtin_aarch64_sqrdmulh_laneqv4si (__a, __b, __c);
}

/* Table intrinsics.  */

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1_p8 (poly8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbl1v8qi_ppu (__tab, __idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1_s8 (int8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbl1v8qi_ssu (__tab, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1_u8 (uint8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbl1v8qi_uuu (__tab, __idx);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1q_p8 (poly8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbl1v16qi_ppu (__tab, __idx);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1q_s8 (int8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbl1v16qi_ssu (__tab, __idx);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbl1q_u8 (uint8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbl1v16qi_uuu (__tab, __idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1_s8 (int8x8_t __r, int8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbx1v8qi_sssu (__r, __tab, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1_u8 (uint8x8_t __r, uint8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbx1v8qi_uuuu (__r, __tab, __idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1_p8 (poly8x8_t __r, poly8x16_t __tab, uint8x8_t __idx)
{
  return __builtin_aarch64_qtbx1v8qi_pppu (__r, __tab, __idx);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1q_s8 (int8x16_t __r, int8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbx1v16qi_sssu (__r, __tab, __idx);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1q_u8 (uint8x16_t __r, uint8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbx1v16qi_uuuu (__r, __tab, __idx);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqtbx1q_p8 (poly8x16_t __r, poly8x16_t __tab, uint8x16_t __idx)
{
  return __builtin_aarch64_qtbx1v16qi_pppu (__r, __tab, __idx);
}

/* V7 legacy table intrinsics.  */

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl1_s8 (int8x8_t __tab, int8x8_t __idx)
{
  int8x16_t __temp = vcombine_s8 (__tab,
				  vcreate_s8 (__AARCH64_UINT64_C (0x0)));
  return __builtin_aarch64_qtbl1v8qi (__temp, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl1_u8 (uint8x8_t __tab, uint8x8_t __idx)
{
  uint8x16_t __temp = vcombine_u8 (__tab,
				   vcreate_u8 (__AARCH64_UINT64_C (0x0)));
  return __builtin_aarch64_qtbl1v8qi_uuu (__temp, __idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl1_p8 (poly8x8_t __tab, uint8x8_t __idx)
{
  poly8x16_t __temp = vcombine_p8 (__tab,
				   vcreate_p8 (__AARCH64_UINT64_C (0x0)));
  return __builtin_aarch64_qtbl1v8qi_ppu (__temp, __idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl2_s8 (int8x8x2_t __tab, int8x8_t __idx)
{
  int8x16_t __temp = vcombine_s8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbl1v8qi (__temp, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl2_u8 (uint8x8x2_t __tab, uint8x8_t __idx)
{
  uint8x16_t __temp = vcombine_u8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbl1v8qi_uuu (__temp, __idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl2_p8 (poly8x8x2_t __tab, uint8x8_t __idx)
{
  poly8x16_t __temp = vcombine_p8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbl1v8qi_ppu (__temp, __idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl3_s8 (int8x8x3_t __tab, int8x8_t __idx)
{
  int8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_s8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_s8 (__tab.val[2],
			       vcreate_s8 (__AARCH64_UINT64_C (0x0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return __builtin_aarch64_qtbl2v8qi (__o, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl3_u8 (uint8x8x3_t __tab, uint8x8_t __idx)
{
  uint8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_u8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_u8 (__tab.val[2],
			       vcreate_u8 (__AARCH64_UINT64_C (0x0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return (uint8x8_t)__builtin_aarch64_qtbl2v8qi (__o, (int8x8_t)__idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl3_p8 (poly8x8x3_t __tab, uint8x8_t __idx)
{
  poly8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_p8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_p8 (__tab.val[2],
			       vcreate_p8 (__AARCH64_UINT64_C (0x0)));
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return (poly8x8_t)__builtin_aarch64_qtbl2v8qi (__o, (int8x8_t)__idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl4_s8 (int8x8x4_t __tab, int8x8_t __idx)
{
  int8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_s8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_s8 (__tab.val[2], __tab.val[3]);
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return __builtin_aarch64_qtbl2v8qi (__o, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl4_u8 (uint8x8x4_t __tab, uint8x8_t __idx)
{
  uint8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_u8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_u8 (__tab.val[2], __tab.val[3]);
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return (uint8x8_t)__builtin_aarch64_qtbl2v8qi (__o, (int8x8_t)__idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbl4_p8 (poly8x8x4_t __tab, uint8x8_t __idx)
{
  poly8x16x2_t __temp;
  __builtin_aarch64_simd_oi __o;
  __temp.val[0] = vcombine_p8 (__tab.val[0], __tab.val[1]);
  __temp.val[1] = vcombine_p8 (__tab.val[2], __tab.val[3]);
  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
  return(poly8x8_t)__builtin_aarch64_qtbl2v8qi (__o, (int8x8_t)__idx);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbx2_s8 (int8x8_t __r, int8x8x2_t __tab, int8x8_t __idx)
{
  int8x16_t __temp = vcombine_s8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbx1v8qi (__r, __temp, __idx);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbx2_u8 (uint8x8_t __r, uint8x8x2_t __tab, uint8x8_t __idx)
{
  uint8x16_t __temp = vcombine_u8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbx1v8qi_uuuu (__r, __temp, __idx);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vtbx2_p8 (poly8x8_t __r, poly8x8x2_t __tab, uint8x8_t __idx)
{
  poly8x16_t __temp = vcombine_p8 (__tab.val[0], __tab.val[1]);
  return __builtin_aarch64_qtbx1v8qi_pppu (__r, __temp, __idx);
}

/* End of temporary inline asm.  */

/* Start of optimal implementations in approved order.  */

/* vabd.  */

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabds_f32 (float32_t __a, float32_t __b)
{
  return __builtin_aarch64_fabdsf (__a, __b);
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdd_f64 (float64_t __a, float64_t __b)
{
  return __builtin_aarch64_fabddf (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_f32 (float32x2_t __a, float32x2_t __b)
{
  return __builtin_aarch64_fabdv2sf (__a, __b);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabd_f64 (float64x1_t __a, float64x1_t __b)
{
  return (float64x1_t) {vabdd_f64 (vget_lane_f64 (__a, 0),
				   vget_lane_f64 (__b, 0))};
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_f32 (float32x4_t __a, float32x4_t __b)
{
  return __builtin_aarch64_fabdv4sf (__a, __b);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabdq_f64 (float64x2_t __a, float64x2_t __b)
{
  return __builtin_aarch64_fabdv2df (__a, __b);
}

/* vabs  */

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_f32 (float32x2_t __a)
{
  return __builtin_aarch64_absv2sf (__a);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_f64 (float64x1_t __a)
{
  return (float64x1_t) {__builtin_fabs (__a[0])};
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_s8 (int8x8_t __a)
{
  return __builtin_aarch64_absv8qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_s16 (int16x4_t __a)
{
  return __builtin_aarch64_absv4hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_s32 (int32x2_t __a)
{
  return __builtin_aarch64_absv2si (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabs_s64 (int64x1_t __a)
{
  return (int64x1_t) {__builtin_aarch64_absdi (__a[0])};
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_f32 (float32x4_t __a)
{
  return __builtin_aarch64_absv4sf (__a);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_f64 (float64x2_t __a)
{
  return __builtin_aarch64_absv2df (__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_absv16qi (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_absv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_absv4si (__a);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsq_s64 (int64x2_t __a)
{
  return __builtin_aarch64_absv2di (__a);
}

/* Try to avoid moving between integer and vector registers.
   For why the cast to unsigned is needed check the vnegd_s64 intrinsic.
   There is a testcase related to this issue:
   gcc.target/aarch64/vabsd_s64.c.  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vabsd_s64 (int64_t __a)
{
  return __a < 0 ? - (uint64_t) __a : __a;
}

/* vadd */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddd_s64 (int64_t __a, int64_t __b)
{
  return __a + __b;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddd_u64 (uint64_t __a, uint64_t __b)
{
  return __a + __b;
}

/* vaddv */

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_s8 (int8x8_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v8qi (__a);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_s16 (int16x4_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v4hi (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_s32 (int32x2_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v2si (__a);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_u8 (uint8x8_t __a)
{
  return (uint8_t) __builtin_aarch64_reduc_plus_scal_v8qi ((int8x8_t) __a);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_u16 (uint16x4_t __a)
{
  return (uint16_t) __builtin_aarch64_reduc_plus_scal_v4hi ((int16x4_t) __a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_u32 (uint32x2_t __a)
{
  return (int32_t) __builtin_aarch64_reduc_plus_scal_v2si ((int32x2_t) __a);
}

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v16qi (__a);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v8hi (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v4si (__a);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_s64 (int64x2_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v2di (__a);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_u8 (uint8x16_t __a)
{
  return (uint8_t) __builtin_aarch64_reduc_plus_scal_v16qi ((int8x16_t) __a);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_u16 (uint16x8_t __a)
{
  return (uint16_t) __builtin_aarch64_reduc_plus_scal_v8hi ((int16x8_t) __a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_u32 (uint32x4_t __a)
{
  return (uint32_t) __builtin_aarch64_reduc_plus_scal_v4si ((int32x4_t) __a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_u64 (uint64x2_t __a)
{
  return (uint64_t) __builtin_aarch64_reduc_plus_scal_v2di ((int64x2_t) __a);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddv_f32 (float32x2_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v2sf (__a);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_f32 (float32x4_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v4sf (__a);
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaddvq_f64 (float64x2_t __a)
{
  return __builtin_aarch64_reduc_plus_scal_v2df (__a);
}

/* vbsl  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_f16 (uint16x4_t __a, float16x4_t __b, float16x4_t __c)
{
  return __builtin_aarch64_simd_bslv4hf_suss (__a, __b, __c);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_f32 (uint32x2_t __a, float32x2_t __b, float32x2_t __c)
{
  return __builtin_aarch64_simd_bslv2sf_suss (__a, __b, __c);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_f64 (uint64x1_t __a, float64x1_t __b, float64x1_t __c)
{
  return (float64x1_t)
    { __builtin_aarch64_simd_bsldf_suss (__a[0], __b[0], __c[0]) };
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_p8 (uint8x8_t __a, poly8x8_t __b, poly8x8_t __c)
{
  return __builtin_aarch64_simd_bslv8qi_pupp (__a, __b, __c);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_p16 (uint16x4_t __a, poly16x4_t __b, poly16x4_t __c)
{
  return __builtin_aarch64_simd_bslv4hi_pupp (__a, __b, __c);
}
__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_p64 (uint64x1_t __a, poly64x1_t __b, poly64x1_t __c)
{
  return (poly64x1_t)
      {__builtin_aarch64_simd_bsldi_pupp (__a[0], __b[0], __c[0])};
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_s8 (uint8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return __builtin_aarch64_simd_bslv8qi_suss (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_s16 (uint16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_simd_bslv4hi_suss (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_s32 (uint32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_simd_bslv2si_suss (__a, __b, __c);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_s64 (uint64x1_t __a, int64x1_t __b, int64x1_t __c)
{
  return (int64x1_t)
      {__builtin_aarch64_simd_bsldi_suss (__a[0], __b[0], __c[0])};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return __builtin_aarch64_simd_bslv8qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return __builtin_aarch64_simd_bslv4hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return __builtin_aarch64_simd_bslv2si_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbsl_u64 (uint64x1_t __a, uint64x1_t __b, uint64x1_t __c)
{
  return (uint64x1_t)
      {__builtin_aarch64_simd_bsldi_uuuu (__a[0], __b[0], __c[0])};
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_f16 (uint16x8_t __a, float16x8_t __b, float16x8_t __c)
{
  return __builtin_aarch64_simd_bslv8hf_suss (__a, __b, __c);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_f32 (uint32x4_t __a, float32x4_t __b, float32x4_t __c)
{
  return __builtin_aarch64_simd_bslv4sf_suss (__a, __b, __c);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_f64 (uint64x2_t __a, float64x2_t __b, float64x2_t __c)
{
  return __builtin_aarch64_simd_bslv2df_suss (__a, __b, __c);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_p8 (uint8x16_t __a, poly8x16_t __b, poly8x16_t __c)
{
  return __builtin_aarch64_simd_bslv16qi_pupp (__a, __b, __c);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_p16 (uint16x8_t __a, poly16x8_t __b, poly16x8_t __c)
{
  return __builtin_aarch64_simd_bslv8hi_pupp (__a, __b, __c);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_s8 (uint8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return __builtin_aarch64_simd_bslv16qi_suss (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_s16 (uint16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_simd_bslv8hi_suss (__a, __b, __c);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_p64 (uint64x2_t __a, poly64x2_t __b, poly64x2_t __c)
{
  return __builtin_aarch64_simd_bslv2di_pupp (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_s32 (uint32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_simd_bslv4si_suss (__a, __b, __c);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_s64 (uint64x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return __builtin_aarch64_simd_bslv2di_suss (__a, __b, __c);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return __builtin_aarch64_simd_bslv16qi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return __builtin_aarch64_simd_bslv8hi_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return __builtin_aarch64_simd_bslv4si_uuuu (__a, __b, __c);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vbslq_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return __builtin_aarch64_simd_bslv2di_uuuu (__a, __b, __c);
}

/* ARMv8.1-A instrinsics.  */
#pragma GCC push_options
#pragma GCC target ("+nothing+rdma")

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_sqrdmlahv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_sqrdmlahv2si (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_sqrdmlahv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_sqrdmlahv4si (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return __builtin_aarch64_sqrdmlshv4hi (__a, __b, __c);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return __builtin_aarch64_sqrdmlshv2si (__a, __b, __c);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return __builtin_aarch64_sqrdmlshv8hi (__a, __b, __c);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return __builtin_aarch64_sqrdmlshv4si (__a, __b, __c);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_laneq_s16 (int16x4_t __a, int16x4_t __b, int16x8_t __c, const int __d)
{
  return  __builtin_aarch64_sqrdmlah_laneqv4hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_laneq_s32 (int32x2_t __a, int32x2_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_laneqv2si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_laneq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_laneqv8hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_laneq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_laneqv4si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_laneq_s16 (int16x4_t __a, int16x4_t __b, int16x8_t __c, const int __d)
{
  return  __builtin_aarch64_sqrdmlsh_laneqv4hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_laneq_s32 (int32x2_t __a, int32x2_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_laneqv2si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_laneq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_laneqv8hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_laneq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_laneqv4si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return  __builtin_aarch64_sqrdmlah_lanev4hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlah_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_lanev2si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_lanev8hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_lanev4si (__a, __b, __c, __d);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahh_s16 (int16_t __a, int16_t __b, int16_t __c)
{
  return (int16_t) __builtin_aarch64_sqrdmlahhi (__a, __b, __c);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahh_lane_s16 (int16_t __a, int16_t __b, int16x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_lanehi (__a, __b, __c, __d);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahh_laneq_s16 (int16_t __a, int16_t __b, int16x8_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_laneqhi (__a, __b, __c, __d);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahs_s32 (int32_t __a, int32_t __b, int32_t __c)
{
  return (int32_t) __builtin_aarch64_sqrdmlahsi (__a, __b, __c);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahs_lane_s32 (int32_t __a, int32_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_lanesi (__a, __b, __c, __d);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlahs_laneq_s32 (int32_t __a, int32_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlah_laneqsi (__a, __b, __c, __d);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return  __builtin_aarch64_sqrdmlsh_lanev4hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlsh_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_lanev2si (__a, __b, __c, __d);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_lanev8hi (__a, __b, __c, __d);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_lanev4si (__a, __b, __c, __d);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshh_s16 (int16_t __a, int16_t __b, int16_t __c)
{
  return (int16_t) __builtin_aarch64_sqrdmlshhi (__a, __b, __c);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshh_lane_s16 (int16_t __a, int16_t __b, int16x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_lanehi (__a, __b, __c, __d);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshh_laneq_s16 (int16_t __a, int16_t __b, int16x8_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_laneqhi (__a, __b, __c, __d);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshs_s32 (int32_t __a, int32_t __b, int32_t __c)
{
  return (int32_t) __builtin_aarch64_sqrdmlshsi (__a, __b, __c);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshs_lane_s32 (int32_t __a, int32_t __b, int32x2_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_lanesi (__a, __b, __c, __d);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vqrdmlshs_laneq_s32 (int32_t __a, int32_t __b, int32x4_t __c, const int __d)
{
  return __builtin_aarch64_sqrdmlsh_laneqsi (__a, __b, __c, __d);
}
#pragma GCC pop_options

#pragma GCC push_options
#pragma GCC target ("+nothing+crypto")
/* vaes  */

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaeseq_u8 (uint8x16_t data, uint8x16_t key)
{
  return __builtin_aarch64_crypto_aesev16qi_uuu (data, key);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaesdq_u8 (uint8x16_t data, uint8x16_t key)
{
  return __builtin_aarch64_crypto_aesdv16qi_uuu (data, key);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaesmcq_u8 (uint8x16_t data)
{
  return __builtin_aarch64_crypto_aesmcv16qi_uu (data);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vaesimcq_u8 (uint8x16_t data)
{
  return __builtin_aarch64_crypto_aesimcv16qi_uu (data);
}
#pragma GCC pop_options

/* vcage  */

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcage_f64 (float64x1_t __a, float64x1_t __b)
{
  return vabs_f64 (__a) >= vabs_f64 (__b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcages_f32 (float32_t __a, float32_t __b)
{
  return __builtin_fabsf (__a) >= __builtin_fabsf (__b) ? -1 : 0;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcage_f32 (float32x2_t __a, float32x2_t __b)
{
  return vabs_f32 (__a) >= vabs_f32 (__b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcageq_f32 (float32x4_t __a, float32x4_t __b)
{
  return vabsq_f32 (__a) >= vabsq_f32 (__b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaged_f64 (float64_t __a, float64_t __b)
{
  return __builtin_fabs (__a) >= __builtin_fabs (__b) ? -1 : 0;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcageq_f64 (float64x2_t __a, float64x2_t __b)
{
  return vabsq_f64 (__a) >= vabsq_f64 (__b);
}

/* vcagt  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagts_f32 (float32_t __a, float32_t __b)
{
  return __builtin_fabsf (__a) > __builtin_fabsf (__b) ? -1 : 0;
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagt_f32 (float32x2_t __a, float32x2_t __b)
{
  return vabs_f32 (__a) > vabs_f32 (__b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagt_f64 (float64x1_t __a, float64x1_t __b)
{
  return vabs_f64 (__a) > vabs_f64 (__b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagtq_f32 (float32x4_t __a, float32x4_t __b)
{
  return vabsq_f32 (__a) > vabsq_f32 (__b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagtd_f64 (float64_t __a, float64_t __b)
{
  return __builtin_fabs (__a) > __builtin_fabs (__b) ? -1 : 0;
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcagtq_f64 (float64x2_t __a, float64x2_t __b)
{
  return vabsq_f64 (__a) > vabsq_f64 (__b);
}

/* vcale  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcale_f32 (float32x2_t __a, float32x2_t __b)
{
  return vabs_f32 (__a) <= vabs_f32 (__b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcale_f64 (float64x1_t __a, float64x1_t __b)
{
  return vabs_f64 (__a) <= vabs_f64 (__b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaled_f64 (float64_t __a, float64_t __b)
{
  return __builtin_fabs (__a) <= __builtin_fabs (__b) ? -1 : 0;
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcales_f32 (float32_t __a, float32_t __b)
{
  return __builtin_fabsf (__a) <= __builtin_fabsf (__b) ? -1 : 0;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaleq_f32 (float32x4_t __a, float32x4_t __b)
{
  return vabsq_f32 (__a) <= vabsq_f32 (__b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaleq_f64 (float64x2_t __a, float64x2_t __b)
{
  return vabsq_f64 (__a) <= vabsq_f64 (__b);
}

/* vcalt  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcalt_f32 (float32x2_t __a, float32x2_t __b)
{
  return vabs_f32 (__a) < vabs_f32 (__b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcalt_f64 (float64x1_t __a, float64x1_t __b)
{
  return vabs_f64 (__a) < vabs_f64 (__b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaltd_f64 (float64_t __a, float64_t __b)
{
  return __builtin_fabs (__a) < __builtin_fabs (__b) ? -1 : 0;
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaltq_f32 (float32x4_t __a, float32x4_t __b)
{
  return vabsq_f32 (__a) < vabsq_f32 (__b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcaltq_f64 (float64x2_t __a, float64x2_t __b)
{
  return vabsq_f64 (__a) < vabsq_f64 (__b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcalts_f32 (float32_t __a, float32_t __b)
{
  return __builtin_fabsf (__a) < __builtin_fabsf (__b) ? -1 : 0;
}

/* vceq - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t) (__a == __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_f64 (float64x1_t __a, float64x1_t __b)
{
  return (uint64x1_t) (__a == __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (uint8x8_t) (__a == __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_p64 (poly64x1_t __a, poly64x1_t __b)
{
  return (uint64x1_t) (__a == __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t) (__a == __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t) (__a == __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t) (__a == __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_s64 (int64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t) (__a == __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceq_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t) (__a == __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_f64 (float64x2_t __a, float64x2_t __b)
{
  return (uint64x2_t) (__a == __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (uint8x16_t) (__a == __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t) (__a == __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t) (__a == __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t) (__a == __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t) (__a == __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (__a == __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqq_p64 (poly64x2_t __a, poly64x2_t __b)
{
  return (__a == __b);
}

/* vceq - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqs_f32 (float32_t __a, float32_t __b)
{
  return __a == __b ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqd_s64 (int64_t __a, int64_t __b)
{
  return __a == __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqd_u64 (uint64_t __a, uint64_t __b)
{
  return __a == __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqd_f64 (float64_t __a, float64_t __b)
{
  return __a == __b ? -1ll : 0ll;
}

/* vceqz - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_f32 (float32x2_t __a)
{
  return (uint32x2_t) (__a == 0.0f);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_f64 (float64x1_t __a)
{
  return (uint64x1_t) (__a == (float64x1_t) {0.0});
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_p8 (poly8x8_t __a)
{
  return (uint8x8_t) (__a == 0);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_s8 (int8x8_t __a)
{
  return (uint8x8_t) (__a == 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_s16 (int16x4_t __a)
{
  return (uint16x4_t) (__a == 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_s32 (int32x2_t __a)
{
  return (uint32x2_t) (__a == 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_s64 (int64x1_t __a)
{
  return (uint64x1_t) (__a == __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_u8 (uint8x8_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_u16 (uint16x4_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_u32 (uint32x2_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_u64 (uint64x1_t __a)
{
  return (__a == __AARCH64_UINT64_C (0));
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqz_p64 (poly64x1_t __a)
{
  return (__a == __AARCH64_UINT64_C (0));
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_f32 (float32x4_t __a)
{
  return (uint32x4_t) (__a == 0.0f);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_f64 (float64x2_t __a)
{
  return (uint64x2_t) (__a == 0.0f);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_p8 (poly8x16_t __a)
{
  return (uint8x16_t) (__a == 0);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_s8 (int8x16_t __a)
{
  return (uint8x16_t) (__a == 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_s16 (int16x8_t __a)
{
  return (uint16x8_t) (__a == 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_s32 (int32x4_t __a)
{
  return (uint32x4_t) (__a == 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_s64 (int64x2_t __a)
{
  return (uint64x2_t) (__a == __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_u8 (uint8x16_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_u16 (uint16x8_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_u32 (uint32x4_t __a)
{
  return (__a == 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_u64 (uint64x2_t __a)
{
  return (__a == __AARCH64_UINT64_C (0));
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzq_p64 (poly64x2_t __a)
{
  return (__a == __AARCH64_UINT64_C (0));
}

/* vceqz - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzs_f32 (float32_t __a)
{
  return __a == 0.0f ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzd_s64 (int64_t __a)
{
  return __a == 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzd_u64 (uint64_t __a)
{
  return __a == 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vceqzd_f64 (float64_t __a)
{
  return __a == 0.0 ? -1ll : 0ll;
}

/* vcge - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t) (__a >= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_f64 (float64x1_t __a, float64x1_t __b)
{
  return (uint64x1_t) (__a >= __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t) (__a >= __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t) (__a >= __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t) (__a >= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_s64 (int64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t) (__a >= __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcge_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t) (__a >= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_f64 (float64x2_t __a, float64x2_t __b)
{
  return (uint64x2_t) (__a >= __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t) (__a >= __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t) (__a >= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t) (__a >= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t) (__a >= __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (__a >= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgeq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (__a >= __b);
}

/* vcge - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcges_f32 (float32_t __a, float32_t __b)
{
  return __a >= __b ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcged_s64 (int64_t __a, int64_t __b)
{
  return __a >= __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcged_u64 (uint64_t __a, uint64_t __b)
{
  return __a >= __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcged_f64 (float64_t __a, float64_t __b)
{
  return __a >= __b ? -1ll : 0ll;
}

/* vcgez - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_f32 (float32x2_t __a)
{
  return (uint32x2_t) (__a >= 0.0f);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_f64 (float64x1_t __a)
{
  return (uint64x1_t) (__a[0] >= (float64x1_t) {0.0});
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_s8 (int8x8_t __a)
{
  return (uint8x8_t) (__a >= 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_s16 (int16x4_t __a)
{
  return (uint16x4_t) (__a >= 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_s32 (int32x2_t __a)
{
  return (uint32x2_t) (__a >= 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgez_s64 (int64x1_t __a)
{
  return (uint64x1_t) (__a >= __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_f32 (float32x4_t __a)
{
  return (uint32x4_t) (__a >= 0.0f);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_f64 (float64x2_t __a)
{
  return (uint64x2_t) (__a >= 0.0);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_s8 (int8x16_t __a)
{
  return (uint8x16_t) (__a >= 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_s16 (int16x8_t __a)
{
  return (uint16x8_t) (__a >= 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_s32 (int32x4_t __a)
{
  return (uint32x4_t) (__a >= 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezq_s64 (int64x2_t __a)
{
  return (uint64x2_t) (__a >= __AARCH64_INT64_C (0));
}

/* vcgez - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezs_f32 (float32_t __a)
{
  return __a >= 0.0f ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezd_s64 (int64_t __a)
{
  return __a >= 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgezd_f64 (float64_t __a)
{
  return __a >= 0.0 ? -1ll : 0ll;
}

/* vcgt - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t) (__a > __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_f64 (float64x1_t __a, float64x1_t __b)
{
  return (uint64x1_t) (__a > __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t) (__a > __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t) (__a > __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t) (__a > __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_s64 (int64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t) (__a > __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgt_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t) (__a > __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_f64 (float64x2_t __a, float64x2_t __b)
{
  return (uint64x2_t) (__a > __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t) (__a > __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t) (__a > __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t) (__a > __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t) (__a > __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (__a > __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (__a > __b);
}

/* vcgt - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgts_f32 (float32_t __a, float32_t __b)
{
  return __a > __b ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtd_s64 (int64_t __a, int64_t __b)
{
  return __a > __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtd_u64 (uint64_t __a, uint64_t __b)
{
  return __a > __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtd_f64 (float64_t __a, float64_t __b)
{
  return __a > __b ? -1ll : 0ll;
}

/* vcgtz - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_f32 (float32x2_t __a)
{
  return (uint32x2_t) (__a > 0.0f);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_f64 (float64x1_t __a)
{
  return (uint64x1_t) (__a > (float64x1_t) {0.0});
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_s8 (int8x8_t __a)
{
  return (uint8x8_t) (__a > 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_s16 (int16x4_t __a)
{
  return (uint16x4_t) (__a > 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_s32 (int32x2_t __a)
{
  return (uint32x2_t) (__a > 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtz_s64 (int64x1_t __a)
{
  return (uint64x1_t) (__a > __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_f32 (float32x4_t __a)
{
  return (uint32x4_t) (__a > 0.0f);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_f64 (float64x2_t __a)
{
    return (uint64x2_t) (__a > 0.0);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_s8 (int8x16_t __a)
{
  return (uint8x16_t) (__a > 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_s16 (int16x8_t __a)
{
  return (uint16x8_t) (__a > 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_s32 (int32x4_t __a)
{
  return (uint32x4_t) (__a > 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzq_s64 (int64x2_t __a)
{
  return (uint64x2_t) (__a > __AARCH64_INT64_C (0));
}

/* vcgtz - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzs_f32 (float32_t __a)
{
  return __a > 0.0f ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzd_s64 (int64_t __a)
{
  return __a > 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcgtzd_f64 (float64_t __a)
{
  return __a > 0.0 ? -1ll : 0ll;
}

/* vcle - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t) (__a <= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_f64 (float64x1_t __a, float64x1_t __b)
{
  return (uint64x1_t) (__a <= __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t) (__a <= __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t) (__a <= __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t) (__a <= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_s64 (int64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t) (__a <= __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcle_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t) (__a <= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_f64 (float64x2_t __a, float64x2_t __b)
{
  return (uint64x2_t) (__a <= __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t) (__a <= __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t) (__a <= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t) (__a <= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t) (__a <= __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (__a <= __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcleq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (__a <= __b);
}

/* vcle - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcles_f32 (float32_t __a, float32_t __b)
{
  return __a <= __b ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcled_s64 (int64_t __a, int64_t __b)
{
  return __a <= __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcled_u64 (uint64_t __a, uint64_t __b)
{
  return __a <= __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcled_f64 (float64_t __a, float64_t __b)
{
  return __a <= __b ? -1ll : 0ll;
}

/* vclez - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_f32 (float32x2_t __a)
{
  return (uint32x2_t) (__a <= 0.0f);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_f64 (float64x1_t __a)
{
  return (uint64x1_t) (__a <= (float64x1_t) {0.0});
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_s8 (int8x8_t __a)
{
  return (uint8x8_t) (__a <= 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_s16 (int16x4_t __a)
{
  return (uint16x4_t) (__a <= 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_s32 (int32x2_t __a)
{
  return (uint32x2_t) (__a <= 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclez_s64 (int64x1_t __a)
{
  return (uint64x1_t) (__a <= __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_f32 (float32x4_t __a)
{
  return (uint32x4_t) (__a <= 0.0f);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_f64 (float64x2_t __a)
{
  return (uint64x2_t) (__a <= 0.0);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_s8 (int8x16_t __a)
{
  return (uint8x16_t) (__a <= 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_s16 (int16x8_t __a)
{
  return (uint16x8_t) (__a <= 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_s32 (int32x4_t __a)
{
  return (uint32x4_t) (__a <= 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezq_s64 (int64x2_t __a)
{
  return (uint64x2_t) (__a <= __AARCH64_INT64_C (0));
}

/* vclez - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezs_f32 (float32_t __a)
{
  return __a <= 0.0f ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezd_s64 (int64_t __a)
{
  return __a <= 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclezd_f64 (float64_t __a)
{
  return __a <= 0.0 ? -1ll : 0ll;
}

/* vclt - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t) (__a < __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_f64 (float64x1_t __a, float64x1_t __b)
{
  return (uint64x1_t) (__a < __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t) (__a < __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t) (__a < __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t) (__a < __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_s64 (int64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t) (__a < __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclt_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t) (__a < __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_f64 (float64x2_t __a, float64x2_t __b)
{
  return (uint64x2_t) (__a < __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t) (__a < __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t) (__a < __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t) (__a < __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t) (__a < __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (__a < __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (__a < __b);
}

/* vclt - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclts_f32 (float32_t __a, float32_t __b)
{
  return __a < __b ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltd_s64 (int64_t __a, int64_t __b)
{
  return __a < __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltd_u64 (uint64_t __a, uint64_t __b)
{
  return __a < __b ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltd_f64 (float64_t __a, float64_t __b)
{
  return __a < __b ? -1ll : 0ll;
}

/* vcltz - vector.  */

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_f32 (float32x2_t __a)
{
  return (uint32x2_t) (__a < 0.0f);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_f64 (float64x1_t __a)
{
  return (uint64x1_t) (__a < (float64x1_t) {0.0});
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_s8 (int8x8_t __a)
{
  return (uint8x8_t) (__a < 0);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_s16 (int16x4_t __a)
{
  return (uint16x4_t) (__a < 0);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_s32 (int32x2_t __a)
{
  return (uint32x2_t) (__a < 0);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltz_s64 (int64x1_t __a)
{
  return (uint64x1_t) (__a < __AARCH64_INT64_C (0));
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_f32 (float32x4_t __a)
{
  return (uint32x4_t) (__a < 0.0f);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_f64 (float64x2_t __a)
{
  return (uint64x2_t) (__a < 0.0);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_s8 (int8x16_t __a)
{
  return (uint8x16_t) (__a < 0);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_s16 (int16x8_t __a)
{
  return (uint16x8_t) (__a < 0);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_s32 (int32x4_t __a)
{
  return (uint32x4_t) (__a < 0);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzq_s64 (int64x2_t __a)
{
  return (uint64x2_t) (__a < __AARCH64_INT64_C (0));
}

/* vcltz - scalar.  */

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzs_f32 (float32_t __a)
{
  return __a < 0.0f ? -1 : 0;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzd_s64 (int64_t __a)
{
  return __a < 0 ? -1ll : 0ll;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcltzd_f64 (float64_t __a)
{
  return __a < 0.0 ? -1ll : 0ll;
}

/* vcls.  */

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_s8 (int8x8_t __a)
{
  return __builtin_aarch64_clrsbv8qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_s16 (int16x4_t __a)
{
  return __builtin_aarch64_clrsbv4hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_s32 (int32x2_t __a)
{
  return __builtin_aarch64_clrsbv2si (__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_clrsbv16qi (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_clrsbv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_clrsbv4si (__a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_u8 (uint8x8_t __a)
{
  return __builtin_aarch64_clrsbv8qi ((int8x8_t) __a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_u16 (uint16x4_t __a)
{
  return __builtin_aarch64_clrsbv4hi ((int16x4_t) __a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcls_u32 (uint32x2_t __a)
{
  return __builtin_aarch64_clrsbv2si ((int32x2_t) __a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_u8 (uint8x16_t __a)
{
  return __builtin_aarch64_clrsbv16qi ((int8x16_t) __a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_u16 (uint16x8_t __a)
{
  return __builtin_aarch64_clrsbv8hi ((int16x8_t) __a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclsq_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_clrsbv4si ((int32x4_t) __a);
}

/* vclz.  */

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_s8 (int8x8_t __a)
{
  return __builtin_aarch64_clzv8qi (__a);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_s16 (int16x4_t __a)
{
  return __builtin_aarch64_clzv4hi (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_s32 (int32x2_t __a)
{
  return __builtin_aarch64_clzv2si (__a);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_aarch64_clzv8qi ((int8x8_t)__a);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_u16 (uint16x4_t __a)
{
  return (uint16x4_t)__builtin_aarch64_clzv4hi ((int16x4_t)__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclz_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_aarch64_clzv2si ((int32x2_t)__a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_clzv16qi (__a);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_s16 (int16x8_t __a)
{
  return __builtin_aarch64_clzv8hi (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_s32 (int32x4_t __a)
{
  return __builtin_aarch64_clzv4si (__a);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_aarch64_clzv16qi ((int8x16_t)__a);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_u16 (uint16x8_t __a)
{
  return (uint16x8_t)__builtin_aarch64_clzv8hi ((int16x8_t)__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vclzq_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_aarch64_clzv4si ((int32x4_t)__a);
}

/* vcnt.  */

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcnt_p8 (poly8x8_t __a)
{
  return (poly8x8_t) __builtin_aarch64_popcountv8qi ((int8x8_t) __a);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcnt_s8 (int8x8_t __a)
{
  return __builtin_aarch64_popcountv8qi (__a);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcnt_u8 (uint8x8_t __a)
{
  return (uint8x8_t) __builtin_aarch64_popcountv8qi ((int8x8_t) __a);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcntq_p8 (poly8x16_t __a)
{
  return (poly8x16_t) __builtin_aarch64_popcountv16qi ((int8x16_t) __a);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcntq_s8 (int8x16_t __a)
{
  return __builtin_aarch64_popcountv16qi (__a);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcntq_u8 (uint8x16_t __a)
{
  return (uint8x16_t) __builtin_aarch64_popcountv16qi ((int8x16_t) __a);
}

/* vcopy_lane.  */

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_f32 (float32x2_t __a, const int __lane1,
		float32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_f64 (float64x1_t __a, const int __lane1,
		float64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_p8 (poly8x8_t __a, const int __lane1,
	       poly8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_p16 (poly16x4_t __a, const int __lane1,
		poly16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_p64 (poly64x1_t __a, const int __lane1,
		poly64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_s8 (int8x8_t __a, const int __lane1,
	       int8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_s16 (int16x4_t __a, const int __lane1,
		int16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_s32 (int32x2_t __a, const int __lane1,
		int32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_s64 (int64x1_t __a, const int __lane1,
		int64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_u8 (uint8x8_t __a, const int __lane1,
	       uint8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_u16 (uint16x4_t __a, const int __lane1,
		uint16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_u32 (uint32x2_t __a, const int __lane1,
		uint32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_lane_u64 (uint64x1_t __a, const int __lane1,
		uint64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

/* vcopy_laneq.  */

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_f32 (float32x2_t __a, const int __lane1,
		 float32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_f64 (float64x1_t __a, const int __lane1,
		 float64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_p8 (poly8x8_t __a, const int __lane1,
		poly8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_p16 (poly16x4_t __a, const int __lane1,
		 poly16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_p64 (poly64x1_t __a, const int __lane1,
		 poly64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_s8 (int8x8_t __a, const int __lane1,
		int8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_s16 (int16x4_t __a, const int __lane1,
		 int16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_s32 (int32x2_t __a, const int __lane1,
		 int32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_s64 (int64x1_t __a, const int __lane1,
		 int64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_u8 (uint8x8_t __a, const int __lane1,
		uint8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				 __a, __lane1);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_u16 (uint16x4_t __a, const int __lane1,
		 uint16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_u32 (uint32x2_t __a, const int __lane1,
		 uint32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopy_laneq_u64 (uint64x1_t __a, const int __lane1,
		 uint64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

/* vcopyq_lane.  */

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_f32 (float32x4_t __a, const int __lane1,
		 float32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_f64 (float64x2_t __a, const int __lane1,
		 float64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_p8 (poly8x16_t __a, const int __lane1,
		poly8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_p16 (poly16x8_t __a, const int __lane1,
		 poly16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_p64 (poly64x2_t __a, const int __lane1,
		 poly64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_s8 (int8x16_t __a, const int __lane1,
		int8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_s16 (int16x8_t __a, const int __lane1,
		 int16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_s32 (int32x4_t __a, const int __lane1,
		 int32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_s64 (int64x2_t __a, const int __lane1,
		 int64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_u8 (uint8x16_t __a, const int __lane1,
		uint8x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_u16 (uint16x8_t __a, const int __lane1,
		 uint16x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_u32 (uint32x4_t __a, const int __lane1,
		 uint32x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_lane_u64 (uint64x2_t __a, const int __lane1,
		 uint64x1_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

/* vcopyq_laneq.  */

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_f32 (float32x4_t __a, const int __lane1,
		  float32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_f64 (float64x2_t __a, const int __lane1,
		  float64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_p8 (poly8x16_t __a, const int __lane1,
		 poly8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_p16 (poly16x8_t __a, const int __lane1,
		  poly16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_p64 (poly64x2_t __a, const int __lane1,
		  poly64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_s8 (int8x16_t __a, const int __lane1,
		 int8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_s16 (int16x8_t __a, const int __lane1,
		  int16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_s32 (int32x4_t __a, const int __lane1,
		  int32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_s64 (int64x2_t __a, const int __lane1,
		  int64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_u8 (uint8x16_t __a, const int __lane1,
		 uint8x16_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				  __a, __lane1);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_u16 (uint16x8_t __a, const int __lane1,
		  uint16x8_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_u32 (uint32x4_t __a, const int __lane1,
		  uint32x4_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcopyq_laneq_u64 (uint64x2_t __a, const int __lane1,
		  uint64x2_t __b, const int __lane2)
{
  return __aarch64_vset_lane_any (__aarch64_vget_lane_any (__b, __lane2),
				   __a, __lane1);
}

/* vcvt (double -> float).  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f16_f32 (float32x4_t __a)
{
  return __builtin_aarch64_float_truncate_lo_v4hf (__a);
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_high_f16_f32 (float16x4_t __a, float32x4_t __b)
{
  return __builtin_aarch64_float_truncate_hi_v8hf (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f32_f64 (float64x2_t __a)
{
  return __builtin_aarch64_float_truncate_lo_v2sf (__a);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_high_f32_f64 (float32x2_t __a, float64x2_t __b)
{
  return __builtin_aarch64_float_truncate_hi_v4sf (__a, __b);
}

/* vcvt (float -> double).  */

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f32_f16 (float16x4_t __a)
{
  return __builtin_aarch64_float_extend_lo_v4sf (__a);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f64_f32 (float32x2_t __a)
{

  return __builtin_aarch64_float_extend_lo_v2df (__a);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_high_f32_f16 (float16x8_t __a)
{
  return __builtin_aarch64_vec_unpacks_hi_v8hf (__a);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_high_f64_f32 (float32x4_t __a)
{
  return __builtin_aarch64_vec_unpacks_hi_v4sf (__a);
}

/* vcvt (<u>fixed-point -> float).  */

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_n_f64_s64 (int64_t __a, const int __b)
{
  return __builtin_aarch64_scvtfdi (__a, __b);
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_n_f64_u64 (uint64_t __a, const int __b)
{
  return __builtin_aarch64_ucvtfdi_sus (__a, __b);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_n_f32_s32 (int32_t __a, const int __b)
{
  return __builtin_aarch64_scvtfsi (__a, __b);
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_n_f32_u32 (uint32_t __a, const int __b)
{
  return __builtin_aarch64_ucvtfsi_sus (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_f32_s32 (int32x2_t __a, const int __b)
{
  return __builtin_aarch64_scvtfv2si (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_f32_u32 (uint32x2_t __a, const int __b)
{
  return __builtin_aarch64_ucvtfv2si_sus (__a, __b);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_f64_s64 (int64x1_t __a, const int __b)
{
  return (float64x1_t)
    { __builtin_aarch64_scvtfdi (vget_lane_s64 (__a, 0), __b) };
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_f64_u64 (uint64x1_t __a, const int __b)
{
  return (float64x1_t)
    { __builtin_aarch64_ucvtfdi_sus (vget_lane_u64 (__a, 0), __b) };
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_f32_s32 (int32x4_t __a, const int __b)
{
  return __builtin_aarch64_scvtfv4si (__a, __b);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_f32_u32 (uint32x4_t __a, const int __b)
{
  return __builtin_aarch64_ucvtfv4si_sus (__a, __b);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_f64_s64 (int64x2_t __a, const int __b)
{
  return __builtin_aarch64_scvtfv2di (__a, __b);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_f64_u64 (uint64x2_t __a, const int __b)
{
  return __builtin_aarch64_ucvtfv2di_sus (__a, __b);
}

/* vcvt (float -> <u>fixed-point).  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_n_s64_f64 (float64_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzsdf (__a, __b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_n_u64_f64 (float64_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzudf_uss (__a, __b);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_n_s32_f32 (float32_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzssf (__a, __b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_n_u32_f32 (float32_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzusf_uss (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_s32_f32 (float32x2_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzsv2sf (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_u32_f32 (float32x2_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzuv2sf_uss (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_s64_f64 (float64x1_t __a, const int __b)
{
  return (int64x1_t)
    { __builtin_aarch64_fcvtzsdf (vget_lane_f64 (__a, 0), __b) };
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_n_u64_f64 (float64x1_t __a, const int __b)
{
  return (uint64x1_t)
    { __builtin_aarch64_fcvtzudf_uss (vget_lane_f64 (__a, 0), __b) };
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_s32_f32 (float32x4_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzsv4sf (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_u32_f32 (float32x4_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzuv4sf_uss (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_s64_f64 (float64x2_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzsv2df (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_n_u64_f64 (float64x2_t __a, const int __b)
{
  return __builtin_aarch64_fcvtzuv2df_uss (__a, __b);
}

/* vcvt  (<u>int -> float)  */

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_f64_s64 (int64_t __a)
{
  return (float64_t) __a;
}

__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_f64_u64 (uint64_t __a)
{
  return (float64_t) __a;
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_f32_s32 (int32_t __a)
{
  return (float32_t) __a;
}

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_f32_u32 (uint32_t __a)
{
  return (float32_t) __a;
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f32_s32 (int32x2_t __a)
{
  return __builtin_aarch64_floatv2siv2sf (__a);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f32_u32 (uint32x2_t __a)
{
  return __builtin_aarch64_floatunsv2siv2sf ((int32x2_t) __a);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f64_s64 (int64x1_t __a)
{
  return (float64x1_t) { vget_lane_s64 (__a, 0) };
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_f64_u64 (uint64x1_t __a)
{
  return (float64x1_t) { vget_lane_u64 (__a, 0) };
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_f32_s32 (int32x4_t __a)
{
  return __builtin_aarch64_floatv4siv4sf (__a);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_f32_u32 (uint32x4_t __a)
{
  return __builtin_aarch64_floatunsv4siv4sf ((int32x4_t) __a);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_f64_s64 (int64x2_t __a)
{
  return __builtin_aarch64_floatv2div2df (__a);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_f64_u64 (uint64x2_t __a)
{
  return __builtin_aarch64_floatunsv2div2df ((int64x2_t) __a);
}

/* vcvt (float -> <u>int)  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_s64_f64 (float64_t __a)
{
  return (int64_t) __a;
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtd_u64_f64 (float64_t __a)
{
  return (uint64_t) __a;
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_s32_f32 (float32_t __a)
{
  return (int32_t) __a;
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvts_u32_f32 (float32_t __a)
{
  return (uint32_t) __a;
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_s32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lbtruncv2sfv2si (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_u32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lbtruncuv2sfv2si_us (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_s32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lbtruncv4sfv4si (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_u32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lbtruncuv4sfv4si_us (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) {vcvtd_s64_f64 (__a[0])};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvt_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) {vcvtd_u64_f64 (__a[0])};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_s64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lbtruncv2dfv2di (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtq_u64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lbtruncuv2dfv2di_us (__a);
}

/* vcvta  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtad_s64_f64 (float64_t __a)
{
  return __builtin_aarch64_lrounddfdi (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtad_u64_f64 (float64_t __a)
{
  return __builtin_aarch64_lroundudfdi_us (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtas_s32_f32 (float32_t __a)
{
  return __builtin_aarch64_lroundsfsi (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtas_u32_f32 (float32_t __a)
{
  return __builtin_aarch64_lroundusfsi_us (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvta_s32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lroundv2sfv2si (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvta_u32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lrounduv2sfv2si_us (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtaq_s32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lroundv4sfv4si (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtaq_u32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lrounduv4sfv4si_us (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvta_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) {vcvtad_s64_f64 (__a[0])};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvta_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) {vcvtad_u64_f64 (__a[0])};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtaq_s64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lroundv2dfv2di (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtaq_u64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lrounduv2dfv2di_us (__a);
}

/* vcvtm  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmd_s64_f64 (float64_t __a)
{
  return __builtin_llfloor (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmd_u64_f64 (float64_t __a)
{
  return __builtin_aarch64_lfloorudfdi_us (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtms_s32_f32 (float32_t __a)
{
  return __builtin_ifloorf (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtms_u32_f32 (float32_t __a)
{
  return __builtin_aarch64_lfloorusfsi_us (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtm_s32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lfloorv2sfv2si (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtm_u32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lflooruv2sfv2si_us (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmq_s32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lfloorv4sfv4si (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmq_u32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lflooruv4sfv4si_us (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtm_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) {vcvtmd_s64_f64 (__a[0])};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtm_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) {vcvtmd_u64_f64 (__a[0])};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmq_s64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lfloorv2dfv2di (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtmq_u64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lflooruv2dfv2di_us (__a);
}

/* vcvtn  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnd_s64_f64 (float64_t __a)
{
  return __builtin_aarch64_lfrintndfdi (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnd_u64_f64 (float64_t __a)
{
  return __builtin_aarch64_lfrintnudfdi_us (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtns_s32_f32 (float32_t __a)
{
  return __builtin_aarch64_lfrintnsfsi (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtns_u32_f32 (float32_t __a)
{
  return __builtin_aarch64_lfrintnusfsi_us (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtn_s32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lfrintnv2sfv2si (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtn_u32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lfrintnuv2sfv2si_us (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnq_s32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lfrintnv4sfv4si (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnq_u32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lfrintnuv4sfv4si_us (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtn_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) {vcvtnd_s64_f64 (__a[0])};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtn_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) {vcvtnd_u64_f64 (__a[0])};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnq_s64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lfrintnv2dfv2di (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtnq_u64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lfrintnuv2dfv2di_us (__a);
}

/* vcvtp  */

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpd_s64_f64 (float64_t __a)
{
  return __builtin_llceil (__a);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpd_u64_f64 (float64_t __a)
{
  return __builtin_aarch64_lceiludfdi_us (__a);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtps_s32_f32 (float32_t __a)
{
  return __builtin_iceilf (__a);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtps_u32_f32 (float32_t __a)
{
  return __builtin_aarch64_lceilusfsi_us (__a);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtp_s32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lceilv2sfv2si (__a);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtp_u32_f32 (float32x2_t __a)
{
  return __builtin_aarch64_lceiluv2sfv2si_us (__a);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpq_s32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lceilv4sfv4si (__a);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpq_u32_f32 (float32x4_t __a)
{
  return __builtin_aarch64_lceiluv4sfv4si_us (__a);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtp_s64_f64 (float64x1_t __a)
{
  return (int64x1_t) {vcvtpd_s64_f64 (__a[0])};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtp_u64_f64 (float64x1_t __a)
{
  return (uint64x1_t) {vcvtpd_u64_f64 (__a[0])};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpq_s64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lceilv2dfv2di (__a);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vcvtpq_u64_f64 (float64x2_t __a)
{
  return __builtin_aarch64_lceiluv2dfv2di_us (__a);
}

/* vdup_n  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_f16 (float16_t __a)
{
  return (float16x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_f32 (float32_t __a)
{
  return (float32x2_t) {__a, __a};
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_f64 (float64_t __a)
{
  return (float64x1_t) {__a};
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_p8 (poly8_t __a)
{
  return (poly8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_p16 (poly16_t __a)
{
  return (poly16x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_p64 (poly64_t __a)
{
  return (poly64x1_t) {__a};
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_s8 (int8_t __a)
{
  return (int8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_s16 (int16_t __a)
{
  return (int16x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_s32 (int32_t __a)
{
  return (int32x2_t) {__a, __a};
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_s64 (int64_t __a)
{
  return (int64x1_t) {__a};
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_u8 (uint8_t __a)
{
  return (uint8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_u16 (uint16_t __a)
{
  return (uint16x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_u32 (uint32_t __a)
{
  return (uint32x2_t) {__a, __a};
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_n_u64 (uint64_t __a)
{
  return (uint64x1_t) {__a};
}

/* vdupq_n  */

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_f16 (float16_t __a)
{
  return (float16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_f32 (float32_t __a)
{
  return (float32x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_f64 (float64_t __a)
{
  return (float64x2_t) {__a, __a};
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_p8 (poly8_t __a)
{
  return (poly8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
		       __a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_p16 (poly16_t __a)
{
  return (poly16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_p64 (poly64_t __a)
{
  return (poly64x2_t) {__a, __a};
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_s8 (int8_t __a)
{
  return (int8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
		      __a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_s16 (int16_t __a)
{
  return (int16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_s32 (int32_t __a)
{
  return (int32x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_s64 (int64_t __a)
{
  return (int64x2_t) {__a, __a};
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_u8 (uint8_t __a)
{
  return (uint8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
		       __a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_u16 (uint16_t __a)
{
  return (uint16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_u32 (uint32_t __a)
{
  return (uint32x4_t) {__a, __a, __a, __a};
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_n_u64 (uint64_t __a)
{
  return (uint64x2_t) {__a, __a};
}

/* vdup_lane  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_f16 (float16x4_t __a, const int __b)
{
  return __aarch64_vdup_lane_f16 (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_f32 (float32x2_t __a, const int __b)
{
  return __aarch64_vdup_lane_f32 (__a, __b);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_f64 (float64x1_t __a, const int __b)
{
  return __aarch64_vdup_lane_f64 (__a, __b);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_p8 (poly8x8_t __a, const int __b)
{
  return __aarch64_vdup_lane_p8 (__a, __b);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_p16 (poly16x4_t __a, const int __b)
{
  return __aarch64_vdup_lane_p16 (__a, __b);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_p64 (poly64x1_t __a, const int __b)
{
  return __aarch64_vdup_lane_p64 (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_s8 (int8x8_t __a, const int __b)
{
  return __aarch64_vdup_lane_s8 (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_s16 (int16x4_t __a, const int __b)
{
  return __aarch64_vdup_lane_s16 (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_s32 (int32x2_t __a, const int __b)
{
  return __aarch64_vdup_lane_s32 (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_s64 (int64x1_t __a, const int __b)
{
  return __aarch64_vdup_lane_s64 (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_u8 (uint8x8_t __a, const int __b)
{
  return __aarch64_vdup_lane_u8 (__a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_u16 (uint16x4_t __a, const int __b)
{
  return __aarch64_vdup_lane_u16 (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_u32 (uint32x2_t __a, const int __b)
{
  return __aarch64_vdup_lane_u32 (__a, __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_lane_u64 (uint64x1_t __a, const int __b)
{
  return __aarch64_vdup_lane_u64 (__a, __b);
}

/* vdup_laneq  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_f16 (float16x8_t __a, const int __b)
{
  return __aarch64_vdup_laneq_f16 (__a, __b);
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_f32 (float32x4_t __a, const int __b)
{
  return __aarch64_vdup_laneq_f32 (__a, __b);
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_f64 (float64x2_t __a, const int __b)
{
  return __aarch64_vdup_laneq_f64 (__a, __b);
}

__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_p8 (poly8x16_t __a, const int __b)
{
  return __aarch64_vdup_laneq_p8 (__a, __b);
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_p16 (poly16x8_t __a, const int __b)
{
  return __aarch64_vdup_laneq_p16 (__a, __b);
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_p64 (poly64x2_t __a, const int __b)
{
  return __aarch64_vdup_laneq_p64 (__a, __b);
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_s8 (int8x16_t __a, const int __b)
{
  return __aarch64_vdup_laneq_s8 (__a, __b);
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_s16 (int16x8_t __a, const int __b)
{
  return __aarch64_vdup_laneq_s16 (__a, __b);
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_s32 (int32x4_t __a, const int __b)
{
  return __aarch64_vdup_laneq_s32 (__a, __b);
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_s64 (int64x2_t __a, const int __b)
{
  return __aarch64_vdup_laneq_s64 (__a, __b);
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_u8 (uint8x16_t __a, const int __b)
{
  return __aarch64_vdup_laneq_u8 (__a, __b);
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_u16 (uint16x8_t __a, const int __b)
{
  return __aarch64_vdup_laneq_u16 (__a, __b);
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_u32 (uint32x4_t __a, const int __b)
{
  return __aarch64_vdup_laneq_u32 (__a, __b);
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdup_laneq_u64 (uint64x2_t __a, const int __b)
{
  return __aarch64_vdup_laneq_u64 (__a, __b);
}

/* vdupq_lane  */

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_f16 (float16x4_t __a, const int __b)
{
  return __aarch64_vdupq_lane_f16 (__a, __b);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_f32 (float32x2_t __a, const int __b)
{
  return __aarch64_vdupq_lane_f32 (__a, __b);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_f64 (float64x1_t __a, const int __b)
{
  return __aarch64_vdupq_lane_f64 (__a, __b);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_p8 (poly8x8_t __a, const int __b)
{
  return __aarch64_vdupq_lane_p8 (__a, __b);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_p16 (poly16x4_t __a, const int __b)
{
  return __aarch64_vdupq_lane_p16 (__a, __b);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_p64 (poly64x1_t __a, const int __b)
{
  return __aarch64_vdupq_lane_p64 (__a, __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_s8 (int8x8_t __a, const int __b)
{
  return __aarch64_vdupq_lane_s8 (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_s16 (int16x4_t __a, const int __b)
{
  return __aarch64_vdupq_lane_s16 (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_s32 (int32x2_t __a, const int __b)
{
  return __aarch64_vdupq_lane_s32 (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_s64 (int64x1_t __a, const int __b)
{
  return __aarch64_vdupq_lane_s64 (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_u8 (uint8x8_t __a, const int __b)
{
  return __aarch64_vdupq_lane_u8 (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_u16 (uint16x4_t __a, const int __b)
{
  return __aarch64_vdupq_lane_u16 (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_u32 (uint32x2_t __a, const int __b)
{
  return __aarch64_vdupq_lane_u32 (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_lane_u64 (uint64x1_t __a, const int __b)
{
  return __aarch64_vdupq_lane_u64 (__a, __b);
}

/* vdupq_laneq  */

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_f16 (float16x8_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_f16 (__a, __b);
}

__extension__ extern __inline float32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_f32 (float32x4_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_f32 (__a, __b);
}

__extension__ extern __inline float64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_f64 (float64x2_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_f64 (__a, __b);
}

__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_p8 (poly8x16_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_p8 (__a, __b);
}

__extension__ extern __inline poly16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_p16 (poly16x8_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_p16 (__a, __b);
}

__extension__ extern __inline poly64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_p64 (poly64x2_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_p64 (__a, __b);
}

__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_s8 (int8x16_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_s8 (__a, __b);
}

__extension__ extern __inline int16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_s16 (int16x8_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_s16 (__a, __b);
}

__extension__ extern __inline int32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_s32 (int32x4_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_s32 (__a, __b);
}

__extension__ extern __inline int64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_s64 (int64x2_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_s64 (__a, __b);
}

__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_u8 (uint8x16_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_u8 (__a, __b);
}

__extension__ extern __inline uint16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_u16 (uint16x8_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_u16 (__a, __b);
}

__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_u32 (uint32x4_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_u32 (__a, __b);
}

__extension__ extern __inline uint64x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupq_laneq_u64 (uint64x2_t __a, const int __b)
{
  return __aarch64_vdupq_laneq_u64 (__a, __b);
}

/* vdupb_lane  */
__extension__ extern __inline poly8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_lane_p8 (poly8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_lane_s8 (int8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_lane_u8 (uint8x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vduph_lane  */

__extension__ extern __inline float16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_lane_f16 (float16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_lane_p16 (poly16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_lane_s16 (int16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_lane_u16 (uint16x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vdups_lane  */

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_lane_f32 (float32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_lane_s32 (int32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_lane_u32 (uint32x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vdupd_lane  */
__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_lane_f64 (float64x1_t __a, const int __b)
{
  __AARCH64_LANE_CHECK (__a, __b);
  return __a[0];
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_lane_s64 (int64x1_t __a, const int __b)
{
  __AARCH64_LANE_CHECK (__a, __b);
  return __a[0];
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_lane_u64 (uint64x1_t __a, const int __b)
{
  __AARCH64_LANE_CHECK (__a, __b);
  return __a[0];
}

/* vdupb_laneq  */
__extension__ extern __inline poly8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_laneq_p8 (poly8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_laneq_s8 (int8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupb_laneq_u8 (uint8x16_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vduph_laneq  */

__extension__ extern __inline float16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_laneq_f16 (float16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline poly16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_laneq_p16 (poly16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_laneq_s16 (int16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vduph_laneq_u16 (uint16x8_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vdups_laneq  */

__extension__ extern __inline float32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_laneq_f32 (float32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_laneq_s32 (int32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint32_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdups_laneq_u32 (uint32x4_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vdupd_laneq  */
__extension__ extern __inline float64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_laneq_f64 (float64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline int64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_laneq_s64 (int64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

__extension__ extern __inline uint64_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vdupd_laneq_u64 (uint64x2_t __a, const int __b)
{
  return __aarch64_vget_lane_any (__a, __b);
}

/* vext  */

__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_f16 (float16x4_t __a, float16x4_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a,
			    (uint16x4_t) {4 - __c, 5 - __c, 6 - __c, 7 - __c});
#else
  return __builtin_shuffle (__a, __b,
			    (uint16x4_t) {__c, __c + 1, __c + 2, __c + 3});
#endif
}

__extension__ extern __inline float32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_f32 (float32x2_t __a, float32x2_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint32x2_t) {2-__c, 3-__c});
#else
  return __builtin_shuffle (__a, __b, (uint32x2_t) {__c, __c+1});
#endif
}

__extension__ extern __inline float64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_f64 (float64x1_t __a, float64x1_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
  /* The only possible index to the assembler instruction returns element 0.  */
  return __a;
}
__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_p8 (poly8x8_t __a, poly8x8_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint8x8_t)
      {8-__c, 9-__c, 10-__c, 11-__c, 12-__c, 13-__c, 14-__c, 15-__c});
#else
  return __builtin_shuffle (__a, __b,
      (uint8x8_t) {__c, __c+1, __c+2, __c+3, __c+4, __c+5, __c+6, __c+7});
#endif
}

__extension__ extern __inline poly16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_p16 (poly16x4_t __a, poly16x4_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a,
      (uint16x4_t) {4-__c, 5-__c, 6-__c, 7-__c});
#else
  return __builtin_shuffle (__a, __b, (uint16x4_t) {__c, __c+1, __c+2, __c+3});
#endif
}

__extension__ extern __inline poly64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_p64 (poly64x1_t __a, poly64x1_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
  /* The only possible index to the assembler instruction returns element 0.  */
  return __a;
}

__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_s8 (int8x8_t __a, int8x8_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint8x8_t)
      {8-__c, 9-__c, 10-__c, 11-__c, 12-__c, 13-__c, 14-__c, 15-__c});
#else
  return __builtin_shuffle (__a, __b,
      (uint8x8_t) {__c, __c+1, __c+2, __c+3, __c+4, __c+5, __c+6, __c+7});
#endif
}

__extension__ extern __inline int16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_s16 (int16x4_t __a, int16x4_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a,
      (uint16x4_t) {4-__c, 5-__c, 6-__c, 7-__c});
#else
  return __builtin_shuffle (__a, __b, (uint16x4_t) {__c, __c+1, __c+2, __c+3});
#endif
}

__extension__ extern __inline int32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_s32 (int32x2_t __a, int32x2_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint32x2_t) {2-__c, 3-__c});
#else
  return __builtin_shuffle (__a, __b, (uint32x2_t) {__c, __c+1});
#endif
}

__extension__ extern __inline int64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_s64 (int64x1_t __a, int64x1_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
  /* The only possible index to the assembler instruction returns element 0.  */
  return __a;
}

__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_u8 (uint8x8_t __a, uint8x8_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint8x8_t)
      {8-__c, 9-__c, 10-__c, 11-__c, 12-__c, 13-__c, 14-__c, 15-__c});
#else
  return __builtin_shuffle (__a, __b,
      (uint8x8_t) {__c, __c+1, __c+2, __c+3, __c+4, __c+5, __c+6, __c+7});
#endif
}

__extension__ extern __inline uint16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_u16 (uint16x4_t __a, uint16x4_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a,
      (uint16x4_t) {4-__c, 5-__c, 6-__c, 7-__c});
#else
  return __builtin_shuffle (__a, __b, (uint16x4_t) {__c, __c+1, __c+2, __c+3});
#endif
}

__extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_u32 (uint32x2_t __a, uint32x2_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
#ifdef __AARCH64EB__
  return __builtin_shuffle (__b, __a, (uint32x2_t) {2-__c, 3-__c});
#else
  return __builtin_shuffle (__a, __b, (uint32x2_t) {__c, __c+1});
#endif
}

__extension__ extern __inline uint64x1_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vext_u64 (uint64x1_t __a, uint64x1_t __b, __const int __c)
{
  __AARCH64_LANE_CHECK (__a, __c);
  /* The only possible index to the assembler instruction returns element 0.  */
  return __a;
}

__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))