4.6.x-google/include/arm_neon.h

/* ARM NEON intrinsics include file. This file is generated automatically
   using neon-gen.ml.  Please do not edit manually.

   Copyright (C) 2006, 2007, 2009 Free Software Foundation, Inc.
   Contributed by CodeSourcery.

   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 _GCC_ARM_NEON_H
#define _GCC_ARM_NEON_H 1

#ifndef __ARM_NEON__
#error You must enable NEON instructions (e.g. -mfloat-abi=softfp -mfpu=neon) to use arm_neon.h
#else

#ifdef __cplusplus
extern "C" {
#endif

#include <stdint.h>

typedef __builtin_neon_qi int8x8_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_hi int16x4_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_si int32x2_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_di int64x1_t;
typedef __builtin_neon_sf float32x2_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_poly8 poly8x8_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_poly16 poly16x4_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_uqi uint8x8_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_uhi uint16x4_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_usi uint32x2_t	__attribute__ ((__vector_size__ (8)));
typedef __builtin_neon_udi uint64x1_t;
typedef __builtin_neon_qi int8x16_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_hi int16x8_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_si int32x4_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_di int64x2_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_sf float32x4_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_poly8 poly8x16_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_poly16 poly16x8_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_uqi uint8x16_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_uhi uint16x8_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_usi uint32x4_t	__attribute__ ((__vector_size__ (16)));
typedef __builtin_neon_udi uint64x2_t	__attribute__ ((__vector_size__ (16)));

typedef float float32_t;
typedef __builtin_neon_poly8 poly8_t;
typedef __builtin_neon_poly16 poly16_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 float32x2x2_t
{
  float32x2_t val[2];
} float32x2x2_t;

typedef struct float32x4x2_t
{
  float32x4_t val[2];
} float32x4x2_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 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 float32x2x3_t
{
  float32x2_t val[3];
} float32x2x3_t;

typedef struct float32x4x3_t
{
  float32x4_t val[3];
} float32x4x3_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 float32x2x4_t
{
  float32x2_t val[4];
} float32x2x4_t;

typedef struct float32x4x4_t
{
  float32x4_t val[4];
} float32x4x4_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;


__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vaddv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vaddv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vaddv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vadd_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vaddv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vaddv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vaddv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vaddv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vadd_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vadddi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vadd_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vadddi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vaddv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vaddv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vaddv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vaddq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vaddv2di (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vaddq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vaddv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vaddv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vaddv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vaddv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vaddq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vaddv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vaddl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vaddlv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vaddl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vaddlv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vaddl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vaddlv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vaddl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vaddlv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vaddl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vaddlv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vaddl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vaddlv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vaddw_s8 (int16x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vaddwv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vaddw_s16 (int32x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vaddwv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vaddw_s32 (int64x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vaddwv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vaddw_u8 (uint16x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vaddwv8qi ((int16x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vaddw_u16 (uint32x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vaddwv4hi ((int32x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vaddw_u32 (uint64x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vaddwv2si ((int64x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vhadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vhaddv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vhadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vhaddv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vhadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vhaddv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vhadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vhaddv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vhadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vhaddv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vhadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vhaddv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vhaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vhaddv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vhaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vhaddv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vhaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vhaddv4si (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vhaddv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vhaddv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vhaddv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrhadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vhaddv8qi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrhadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vhaddv4hi (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrhadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vhaddv2si (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrhadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vhaddv8qi ((int8x8_t) __a, (int8x8_t) __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrhadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vhaddv4hi ((int16x4_t) __a, (int16x4_t) __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrhadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vhaddv2si ((int32x2_t) __a, (int32x2_t) __b, 4);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrhaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vhaddv16qi (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrhaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vhaddv8hi (__a, __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vrhaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vhaddv4si (__a, __b, 5);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vhaddv16qi ((int8x16_t) __a, (int8x16_t) __b, 4);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vhaddv8hi ((int16x8_t) __a, (int16x8_t) __b, 4);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vhaddv4si ((int32x4_t) __a, (int32x4_t) __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vqaddv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqaddv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqaddv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vqadd_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vqadddi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vqaddv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vqaddv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vqaddv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqadd_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vqadddi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqaddq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vqaddv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqaddq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqaddv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqaddq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqaddv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqaddq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vqaddv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqaddq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vqaddv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqaddq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vqaddv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqaddq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vqaddv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqaddq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vqaddv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vaddhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t)__builtin_neon_vaddhnv8hi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vaddhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t)__builtin_neon_vaddhnv4si (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vaddhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t)__builtin_neon_vaddhnv2di (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vaddhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vaddhnv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vaddhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vaddhnv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vaddhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vaddhnv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vraddhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t)__builtin_neon_vaddhnv8hi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vraddhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t)__builtin_neon_vaddhnv4si (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vraddhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t)__builtin_neon_vaddhnv2di (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vraddhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vaddhnv8hi ((int16x8_t) __a, (int16x8_t) __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vraddhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vaddhnv4si ((int32x4_t) __a, (int32x4_t) __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vraddhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vaddhnv2di ((int64x2_t) __a, (int64x2_t) __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmul_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vmulv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmul_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vmulv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmul_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vmulv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmul_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vmulv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmul_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vmulv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmul_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vmulv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmul_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vmulv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vmul_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly8x8_t)__builtin_neon_vmulv8qi ((int8x8_t) __a, (int8x8_t) __b, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmulq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vmulv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmulq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vmulv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmulq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vmulv4si (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmulq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vmulv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmulq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vmulv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmulq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vmulv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmulq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vmulv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vmulq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (poly8x16_t)__builtin_neon_vmulv16qi ((int8x16_t) __a, (int8x16_t) __b, 2);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqdmulh_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqdmulhv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqdmulh_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqdmulhv2si (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqdmulhq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqdmulhv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmulhq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmulhv4si (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqrdmulh_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqdmulhv4hi (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqrdmulh_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqdmulhv2si (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqrdmulhq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqdmulhv8hi (__a, __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqrdmulhq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmulhv4si (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmull_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vmullv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmull_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vmullv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmull_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vmullv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmull_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vmullv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmull_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vmullv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmull_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vmullv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vmull_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly16x8_t)__builtin_neon_vmullv8qi ((int8x8_t) __a, (int8x8_t) __b, 2);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmull_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmullv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmull_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vqdmullv2si (__a, __b, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmla_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int8x8_t)__builtin_neon_vmlav8qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmla_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int16x4_t)__builtin_neon_vmlav4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmla_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int32x2_t)__builtin_neon_vmlav2si (__a, __b, __c, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmla_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c)
{
  return (float32x2_t)__builtin_neon_vmlav2sf (__a, __b, __c, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmla_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t)__builtin_neon_vmlav8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmla_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t)__builtin_neon_vmlav4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmla_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t)__builtin_neon_vmlav2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmlaq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return (int8x16_t)__builtin_neon_vmlav16qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlaq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int16x8_t)__builtin_neon_vmlav8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlaq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int32x4_t)__builtin_neon_vmlav4si (__a, __b, __c, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlaq_f32 (float32x4_t __a, float32x4_t __b, float32x4_t __c)
{
  return (float32x4_t)__builtin_neon_vmlav4sf (__a, __b, __c, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmlaq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t)__builtin_neon_vmlav16qi ((int8x16_t) __a, (int8x16_t) __b, (int8x16_t) __c, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlaq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vmlav8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlaq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlav4si ((int32x4_t) __a, (int32x4_t) __b, (int32x4_t) __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlal_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int16x8_t)__builtin_neon_vmlalv8qi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int32x4_t)__builtin_neon_vmlalv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int64x2_t)__builtin_neon_vmlalv2si (__a, __b, __c, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlal_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vmlalv8qi ((int16x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlal_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlalv4hi ((int32x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlal_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint64x2_t)__builtin_neon_vmlalv2si ((int64x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int32x4_t)__builtin_neon_vqdmlalv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int64x2_t)__builtin_neon_vqdmlalv2si (__a, __b, __c, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmls_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int8x8_t)__builtin_neon_vmlsv8qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmls_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int16x4_t)__builtin_neon_vmlsv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmls_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int32x2_t)__builtin_neon_vmlsv2si (__a, __b, __c, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmls_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c)
{
  return (float32x2_t)__builtin_neon_vmlsv2sf (__a, __b, __c, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmls_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t)__builtin_neon_vmlsv8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmls_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t)__builtin_neon_vmlsv4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmls_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t)__builtin_neon_vmlsv2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmlsq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return (int8x16_t)__builtin_neon_vmlsv16qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlsq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int16x8_t)__builtin_neon_vmlsv8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int32x4_t)__builtin_neon_vmlsv4si (__a, __b, __c, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlsq_f32 (float32x4_t __a, float32x4_t __b, float32x4_t __c)
{
  return (float32x4_t)__builtin_neon_vmlsv4sf (__a, __b, __c, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmlsq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t)__builtin_neon_vmlsv16qi ((int8x16_t) __a, (int8x16_t) __b, (int8x16_t) __c, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlsq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vmlsv8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlsv4si ((int32x4_t) __a, (int32x4_t) __b, (int32x4_t) __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlsl_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int16x8_t)__builtin_neon_vmlslv8qi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsl_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int32x4_t)__builtin_neon_vmlslv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlsl_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int64x2_t)__builtin_neon_vmlslv2si (__a, __b, __c, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlsl_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vmlslv8qi ((int16x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsl_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlslv4hi ((int32x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlsl_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint64x2_t)__builtin_neon_vmlslv2si ((int64x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlsl_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int32x4_t)__builtin_neon_vqdmlslv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlsl_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int64x2_t)__builtin_neon_vqdmlslv2si (__a, __b, __c, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vsubv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vsubv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vsubv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vsub_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vsubv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vsubv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vsubv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vsubv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vsub_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vsubdi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vsub_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vsubdi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vsubv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vsubv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vsubv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsubq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vsubv2di (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vsubq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vsubv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vsubv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vsubv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vsubv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsubq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vsubv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsubl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vsublv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsubl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vsublv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsubl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vsublv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsubl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vsublv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsubl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vsublv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsubl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vsublv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsubw_s8 (int16x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vsubwv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsubw_s16 (int32x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vsubwv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsubw_s32 (int64x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vsubwv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsubw_u8 (uint16x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vsubwv8qi ((int16x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsubw_u16 (uint32x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vsubwv4hi ((int32x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsubw_u32 (uint64x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vsubwv2si ((int64x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vhsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vhsubv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vhsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vhsubv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vhsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vhsubv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vhsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vhsubv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vhsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vhsubv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vhsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vhsubv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vhsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vhsubv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vhsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vhsubv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vhsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vhsubv4si (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vhsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vhsubv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vhsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vhsubv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vhsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vhsubv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqsub_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vqsubv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqsub_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqsubv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqsub_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqsubv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vqsub_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vqsubdi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqsub_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vqsubv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqsub_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vqsubv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqsub_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vqsubv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqsub_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vqsubdi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqsubq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vqsubv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqsubq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqsubv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqsubq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqsubv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqsubq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vqsubv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqsubq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vqsubv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqsubq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vqsubv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqsubq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vqsubv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqsubq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vqsubv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vsubhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t)__builtin_neon_vsubhnv8hi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vsubhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t)__builtin_neon_vsubhnv4si (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vsubhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t)__builtin_neon_vsubhnv2di (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vsubhnv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vsubhnv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vsubhnv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrsubhn_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int8x8_t)__builtin_neon_vsubhnv8hi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrsubhn_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int16x4_t)__builtin_neon_vsubhnv4si (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrsubhn_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int32x2_t)__builtin_neon_vsubhnv2di (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vsubhnv8hi ((int16x8_t) __a, (int16x8_t) __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vsubhnv4si ((int32x4_t) __a, (int32x4_t) __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vsubhnv2di ((int64x2_t) __a, (int64x2_t) __b, 4);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vceq_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vceqv8qi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vceq_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vceqv4hi (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vceq_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vceqv2si (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vceq_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vceqv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vceq_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vceqv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vceq_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vceqv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vceq_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vceqv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vceq_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vceqv8qi ((int8x8_t) __a, (int8x8_t) __b, 2);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vceqq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vceqv16qi (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vceqq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vceqv8hi (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vceqq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vceqv4si (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vceqq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vceqv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vceqq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vceqv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vceqq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vceqv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vceqq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vceqv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vceqq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vceqv16qi ((int8x16_t) __a, (int8x16_t) __b, 2);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcge_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgev8qi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcge_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgev4hi (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcge_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2si (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcge_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcge_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgev8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcge_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgev4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcge_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcgeq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgev16qi (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcgeq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgev8hi (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgeq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4si (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgeq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcgeq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgev16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcgeq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgev8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgeq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcle_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgev8qi (__b, __a, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcle_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgev4hi (__b, __a, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcle_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2si (__b, __a, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcle_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2sf (__b, __a, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcle_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgev8qi ((int8x8_t) __b, (int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcle_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgev4hi ((int16x4_t) __b, (int16x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcle_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgev2si ((int32x2_t) __b, (int32x2_t) __a, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcleq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgev16qi (__b, __a, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcleq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgev8hi (__b, __a, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcleq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4si (__b, __a, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcleq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4sf (__b, __a, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcleq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgev16qi ((int8x16_t) __b, (int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcleq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgev8hi ((int16x8_t) __b, (int16x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcleq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgev4si ((int32x4_t) __b, (int32x4_t) __a, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcgt_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgtv8qi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcgt_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgtv4hi (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcgt_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2si (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcgt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcgt_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgtv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcgt_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgtv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcgt_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcgtq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgtv16qi (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcgtq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgtv8hi (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgtq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4si (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgtq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcgtq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgtv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcgtq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgtv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcgtq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vclt_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgtv8qi (__b, __a, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vclt_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgtv4hi (__b, __a, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vclt_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2si (__b, __a, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vclt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2sf (__b, __a, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vclt_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vcgtv8qi ((int8x8_t) __b, (int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vclt_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vcgtv4hi ((int16x4_t) __b, (int16x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vclt_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcgtv2si ((int32x2_t) __b, (int32x2_t) __a, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcltq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgtv16qi (__b, __a, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcltq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgtv8hi (__b, __a, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcltq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4si (__b, __a, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcltq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4sf (__b, __a, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcltq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vcgtv16qi ((int8x16_t) __b, (int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcltq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vcgtv8hi ((int16x8_t) __b, (int16x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcltq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcgtv4si ((int32x4_t) __b, (int32x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcage_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcagev2sf (__a, __b, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcageq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcagev4sf (__a, __b, 3);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcale_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcagev2sf (__b, __a, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcaleq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcagev4sf (__b, __a, 3);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcagt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcagtv2sf (__a, __b, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcagtq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcagtv4sf (__a, __b, 3);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcalt_f32 (float32x2_t __a, float32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vcagtv2sf (__b, __a, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcaltq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vcagtv4sf (__b, __a, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtst_s8 (int8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vtstv8qi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vtst_s16 (int16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vtstv4hi (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vtst_s32 (int32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vtstv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtst_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vtstv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vtst_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vtstv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vtst_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vtstv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtst_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vtstv8qi ((int8x8_t) __a, (int8x8_t) __b, 2);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vtstq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vtstv16qi (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vtstq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vtstv8hi (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vtstq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vtstv4si (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vtstq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vtstv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vtstq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vtstv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vtstq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vtstv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vtstq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vtstv16qi ((int8x16_t) __a, (int8x16_t) __b, 2);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vabd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vabdv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vabd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vabdv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vabd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vabdv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vabd_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vabdv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vabd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vabdv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vabd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vabdv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vabd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vabdv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vabdq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vabdv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vabdq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vabdv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vabdq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vabdv4si (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vabdq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vabdv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vabdq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vabdv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vabdq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vabdv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vabdq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vabdv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vabdl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int16x8_t)__builtin_neon_vabdlv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vabdl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int32x4_t)__builtin_neon_vabdlv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vabdl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int64x2_t)__builtin_neon_vabdlv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vabdl_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vabdlv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vabdl_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vabdlv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vabdl_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vabdlv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vaba_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int8x8_t)__builtin_neon_vabav8qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vaba_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int16x4_t)__builtin_neon_vabav4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vaba_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int32x2_t)__builtin_neon_vabav2si (__a, __b, __c, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vaba_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t)__builtin_neon_vabav8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vaba_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t)__builtin_neon_vabav4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vaba_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t)__builtin_neon_vabav2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vabaq_s8 (int8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return (int8x16_t)__builtin_neon_vabav16qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vabaq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int16x8_t)__builtin_neon_vabav8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vabaq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int32x4_t)__builtin_neon_vabav4si (__a, __b, __c, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vabaq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t)__builtin_neon_vabav16qi ((int8x16_t) __a, (int8x16_t) __b, (int8x16_t) __c, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vabaq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vabav8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vabaq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vabav4si ((int32x4_t) __a, (int32x4_t) __b, (int32x4_t) __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vabal_s8 (int16x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int16x8_t)__builtin_neon_vabalv8qi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vabal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int32x4_t)__builtin_neon_vabalv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vabal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int64x2_t)__builtin_neon_vabalv2si (__a, __b, __c, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vabal_u8 (uint16x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vabalv8qi ((int16x8_t) __a, (int8x8_t) __b, (int8x8_t) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vabal_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vabalv4hi ((int32x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vabal_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint64x2_t)__builtin_neon_vabalv2si ((int64x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmax_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vmaxv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmax_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vmaxv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmax_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vmaxv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmax_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vmaxv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmax_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vmaxv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmax_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vmaxv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmax_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vmaxv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmaxq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vmaxv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmaxq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vmaxv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmaxq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vmaxv4si (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmaxq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vmaxv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmaxq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vmaxv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmaxq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vmaxv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmaxq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vmaxv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmin_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vminv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmin_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vminv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmin_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vminv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmin_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vminv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmin_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vminv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmin_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vminv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmin_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vminv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vminq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vminv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vminq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vminv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vminq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vminv4si (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vminq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vminv4sf (__a, __b, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vminq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vminv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vminq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vminv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vminq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vminv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vpadd_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vpaddv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vpadd_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vpaddv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vpadd_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vpaddv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vpadd_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vpaddv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vpadd_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vpaddv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vpadd_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vpaddv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vpadd_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vpaddv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vpaddl_s8 (int8x8_t __a)
{
  return (int16x4_t)__builtin_neon_vpaddlv8qi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vpaddl_s16 (int16x4_t __a)
{
  return (int32x2_t)__builtin_neon_vpaddlv4hi (__a, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vpaddl_s32 (int32x2_t __a)
{
  return (int64x1_t)__builtin_neon_vpaddlv2si (__a, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vpaddl_u8 (uint8x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vpaddlv8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vpaddl_u16 (uint16x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vpaddlv4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vpaddl_u32 (uint32x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vpaddlv2si ((int32x2_t) __a, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vpaddlq_s8 (int8x16_t __a)
{
  return (int16x8_t)__builtin_neon_vpaddlv16qi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vpaddlq_s16 (int16x8_t __a)
{
  return (int32x4_t)__builtin_neon_vpaddlv8hi (__a, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vpaddlq_s32 (int32x4_t __a)
{
  return (int64x2_t)__builtin_neon_vpaddlv4si (__a, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vpaddlq_u8 (uint8x16_t __a)
{
  return (uint16x8_t)__builtin_neon_vpaddlv16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vpaddlq_u16 (uint16x8_t __a)
{
  return (uint32x4_t)__builtin_neon_vpaddlv8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vpaddlq_u32 (uint32x4_t __a)
{
  return (uint64x2_t)__builtin_neon_vpaddlv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vpadal_s8 (int16x4_t __a, int8x8_t __b)
{
  return (int16x4_t)__builtin_neon_vpadalv8qi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vpadal_s16 (int32x2_t __a, int16x4_t __b)
{
  return (int32x2_t)__builtin_neon_vpadalv4hi (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vpadal_s32 (int64x1_t __a, int32x2_t __b)
{
  return (int64x1_t)__builtin_neon_vpadalv2si (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vpadal_u8 (uint16x4_t __a, uint8x8_t __b)
{
  return (uint16x4_t)__builtin_neon_vpadalv8qi ((int16x4_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vpadal_u16 (uint32x2_t __a, uint16x4_t __b)
{
  return (uint32x2_t)__builtin_neon_vpadalv4hi ((int32x2_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vpadal_u32 (uint64x1_t __a, uint32x2_t __b)
{
  return (uint64x1_t)__builtin_neon_vpadalv2si ((int64x1_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vpadalq_s8 (int16x8_t __a, int8x16_t __b)
{
  return (int16x8_t)__builtin_neon_vpadalv16qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vpadalq_s16 (int32x4_t __a, int16x8_t __b)
{
  return (int32x4_t)__builtin_neon_vpadalv8hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vpadalq_s32 (int64x2_t __a, int32x4_t __b)
{
  return (int64x2_t)__builtin_neon_vpadalv4si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vpadalq_u8 (uint16x8_t __a, uint8x16_t __b)
{
  return (uint16x8_t)__builtin_neon_vpadalv16qi ((int16x8_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vpadalq_u16 (uint32x4_t __a, uint16x8_t __b)
{
  return (uint32x4_t)__builtin_neon_vpadalv8hi ((int32x4_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vpadalq_u32 (uint64x2_t __a, uint32x4_t __b)
{
  return (uint64x2_t)__builtin_neon_vpadalv4si ((int64x2_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vpmax_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vpmaxv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vpmax_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vpmaxv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vpmax_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vpmaxv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vpmax_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vpmaxv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vpmax_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vpmaxv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vpmax_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vpmaxv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vpmax_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vpmaxv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vpmin_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vpminv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vpmin_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vpminv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vpmin_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vpminv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vpmin_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vpminv2sf (__a, __b, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vpmin_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vpminv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vpmin_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vpminv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vpmin_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vpminv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vrecps_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vrecpsv2sf (__a, __b, 3);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vrecpsq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vrecpsv4sf (__a, __b, 3);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vrsqrts_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x2_t)__builtin_neon_vrsqrtsv2sf (__a, __b, 3);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vrsqrtsq_f32 (float32x4_t __a, float32x4_t __b)
{
  return (float32x4_t)__builtin_neon_vrsqrtsv4sf (__a, __b, 3);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vshl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vshlv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vshl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vshlv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vshl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vshlv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vshl_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vshldi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vshl_u8 (uint8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vshlv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vshl_u16 (uint16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vshlv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vshl_u32 (uint32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vshlv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vshl_u64 (uint64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vshldi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vshlq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vshlv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vshlq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vshlv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vshlq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vshlv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vshlq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vshlv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vshlq_u8 (uint8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vshlv16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vshlq_u16 (uint16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vshlv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vshlq_u32 (uint32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vshlv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vshlq_u64 (uint64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vshlv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrshl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vshlv8qi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrshl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vshlv4hi (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrshl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vshlv2si (__a, __b, 5);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vrshl_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vshldi (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrshl_u8 (uint8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vshlv8qi ((int8x8_t) __a, __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrshl_u16 (uint16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vshlv4hi ((int16x4_t) __a, __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrshl_u32 (uint32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vshlv2si ((int32x2_t) __a, __b, 4);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vrshl_u64 (uint64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vshldi ((int64x1_t) __a, __b, 4);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrshlq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vshlv16qi (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrshlq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vshlv8hi (__a, __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vrshlq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vshlv4si (__a, __b, 5);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vrshlq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vshlv2di (__a, __b, 5);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrshlq_u8 (uint8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vshlv16qi ((int8x16_t) __a, __b, 4);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrshlq_u16 (uint16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vshlv8hi ((int16x8_t) __a, __b, 4);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrshlq_u32 (uint32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vshlv4si ((int32x4_t) __a, __b, 4);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vrshlq_u64 (uint64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vshlv2di ((int64x2_t) __a, __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqshl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vqshlv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqshl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqshlv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqshl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqshlv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vqshl_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vqshldi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqshl_u8 (uint8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vqshlv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqshl_u16 (uint16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vqshlv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqshl_u32 (uint32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vqshlv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqshl_u64 (uint64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vqshldi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqshlq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vqshlv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqshlq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqshlv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqshlq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqshlv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqshlq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vqshlv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqshlq_u8 (uint8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vqshlv16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqshlq_u16 (uint16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vqshlv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqshlq_u32 (uint32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vqshlv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqshlq_u64 (uint64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vqshlv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqrshl_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vqshlv8qi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqrshl_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vqshlv4hi (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqrshl_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vqshlv2si (__a, __b, 5);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vqrshl_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vqshldi (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqrshl_u8 (uint8x8_t __a, int8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vqshlv8qi ((int8x8_t) __a, __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqrshl_u16 (uint16x4_t __a, int16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vqshlv4hi ((int16x4_t) __a, __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqrshl_u32 (uint32x2_t __a, int32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vqshlv2si ((int32x2_t) __a, __b, 4);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqrshl_u64 (uint64x1_t __a, int64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vqshldi ((int64x1_t) __a, __b, 4);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqrshlq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vqshlv16qi (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqrshlq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vqshlv8hi (__a, __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqrshlq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vqshlv4si (__a, __b, 5);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqrshlq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vqshlv2di (__a, __b, 5);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqrshlq_u8 (uint8x16_t __a, int8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vqshlv16qi ((int8x16_t) __a, __b, 4);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqrshlq_u16 (uint16x8_t __a, int16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vqshlv8hi ((int16x8_t) __a, __b, 4);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqrshlq_u32 (uint32x4_t __a, int32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vqshlv4si ((int32x4_t) __a, __b, 4);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqrshlq_u64 (uint64x2_t __a, int64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vqshlv2di ((int64x2_t) __a, __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vshr_n_s8 (int8x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vshr_nv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vshr_n_s16 (int16x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vshr_nv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vshr_n_s32 (int32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vshr_nv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vshr_n_s64 (int64x1_t __a, const int __b)
{
  return (int64x1_t)__builtin_neon_vshr_ndi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vshr_n_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vshr_nv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vshr_n_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vshr_nv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vshr_n_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vshr_nv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vshr_n_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vshr_ndi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vshrq_n_s8 (int8x16_t __a, const int __b)
{
  return (int8x16_t)__builtin_neon_vshr_nv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vshrq_n_s16 (int16x8_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vshr_nv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vshrq_n_s32 (int32x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vshr_nv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vshrq_n_s64 (int64x2_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vshr_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vshrq_n_u8 (uint8x16_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vshr_nv16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vshrq_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vshr_nv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vshrq_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vshr_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vshrq_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vshr_nv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrshr_n_s8 (int8x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vshr_nv8qi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrshr_n_s16 (int16x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vshr_nv4hi (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrshr_n_s32 (int32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vshr_nv2si (__a, __b, 5);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vrshr_n_s64 (int64x1_t __a, const int __b)
{
  return (int64x1_t)__builtin_neon_vshr_ndi (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrshr_n_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vshr_nv8qi ((int8x8_t) __a, __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrshr_n_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vshr_nv4hi ((int16x4_t) __a, __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrshr_n_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vshr_nv2si ((int32x2_t) __a, __b, 4);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vrshr_n_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vshr_ndi ((int64x1_t) __a, __b, 4);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrshrq_n_s8 (int8x16_t __a, const int __b)
{
  return (int8x16_t)__builtin_neon_vshr_nv16qi (__a, __b, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrshrq_n_s16 (int16x8_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vshr_nv8hi (__a, __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vrshrq_n_s32 (int32x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vshr_nv4si (__a, __b, 5);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vrshrq_n_s64 (int64x2_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vshr_nv2di (__a, __b, 5);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrshrq_n_u8 (uint8x16_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vshr_nv16qi ((int8x16_t) __a, __b, 4);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrshrq_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vshr_nv8hi ((int16x8_t) __a, __b, 4);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrshrq_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vshr_nv4si ((int32x4_t) __a, __b, 4);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vrshrq_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vshr_nv2di ((int64x2_t) __a, __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vshrn_n_s16 (int16x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vshrn_nv8hi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vshrn_n_s32 (int32x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vshrn_nv4si (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vshrn_n_s64 (int64x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vshrn_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vshrn_nv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vshrn_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vshrn_nv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrshrn_n_s16 (int16x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vshrn_nv8hi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrshrn_n_s32 (int32x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vshrn_nv4si (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrshrn_n_s64 (int64x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vshrn_nv2di (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vshrn_nv8hi ((int16x8_t) __a, __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vshrn_nv4si ((int32x4_t) __a, __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vshrn_nv2di ((int64x2_t) __a, __b, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqshrn_n_s16 (int16x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vqshrn_nv8hi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqshrn_n_s32 (int32x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vqshrn_nv4si (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqshrn_n_s64 (int64x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vqshrn_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshrn_nv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshrn_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshrn_nv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqrshrn_n_s16 (int16x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vqshrn_nv8hi (__a, __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqrshrn_n_s32 (int32x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vqshrn_nv4si (__a, __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqrshrn_n_s64 (int64x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vqshrn_nv2di (__a, __b, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqrshrn_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshrn_nv8hi ((int16x8_t) __a, __b, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqrshrn_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshrn_nv4si ((int32x4_t) __a, __b, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqrshrn_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshrn_nv2di ((int64x2_t) __a, __b, 4);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqshrun_n_s16 (int16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshrun_nv8hi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqshrun_n_s32 (int32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshrun_nv4si (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqshrun_n_s64 (int64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshrun_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqrshrun_n_s16 (int16x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshrun_nv8hi (__a, __b, 5);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqrshrun_n_s32 (int32x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshrun_nv4si (__a, __b, 5);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqrshrun_n_s64 (int64x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshrun_nv2di (__a, __b, 5);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vshl_n_s8 (int8x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vshl_nv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vshl_n_s16 (int16x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vshl_nv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vshl_n_s32 (int32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vshl_nv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vshl_n_s64 (int64x1_t __a, const int __b)
{
  return (int64x1_t)__builtin_neon_vshl_ndi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vshl_n_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vshl_nv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vshl_n_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vshl_nv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vshl_n_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vshl_nv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vshl_n_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vshl_ndi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vshlq_n_s8 (int8x16_t __a, const int __b)
{
  return (int8x16_t)__builtin_neon_vshl_nv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vshlq_n_s16 (int16x8_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vshl_nv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vshlq_n_s32 (int32x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vshl_nv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vshlq_n_s64 (int64x2_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vshl_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vshlq_n_u8 (uint8x16_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vshl_nv16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vshlq_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vshl_nv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vshlq_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vshl_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vshlq_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vshl_nv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqshl_n_s8 (int8x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vqshl_nv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqshl_n_s16 (int16x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vqshl_nv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqshl_n_s32 (int32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vqshl_nv2si (__a, __b, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vqshl_n_s64 (int64x1_t __a, const int __b)
{
  return (int64x1_t)__builtin_neon_vqshl_ndi (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqshl_n_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshl_nv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqshl_n_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshl_nv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqshl_n_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshl_nv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqshl_n_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vqshl_ndi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqshlq_n_s8 (int8x16_t __a, const int __b)
{
  return (int8x16_t)__builtin_neon_vqshl_nv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqshlq_n_s16 (int16x8_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vqshl_nv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqshlq_n_s32 (int32x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vqshl_nv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqshlq_n_s64 (int64x2_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vqshl_nv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqshlq_n_u8 (uint8x16_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vqshl_nv16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqshlq_n_u16 (uint16x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vqshl_nv8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqshlq_n_u32 (uint32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vqshl_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqshlq_n_u64 (uint64x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vqshl_nv2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqshlu_n_s8 (int8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vqshlu_nv8qi (__a, __b, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqshlu_n_s16 (int16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vqshlu_nv4hi (__a, __b, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqshlu_n_s32 (int32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vqshlu_nv2si (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vqshlu_n_s64 (int64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vqshlu_ndi (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vqshluq_n_s8 (int8x16_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vqshlu_nv16qi (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vqshluq_n_s16 (int16x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vqshlu_nv8hi (__a, __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vqshluq_n_s32 (int32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vqshlu_nv4si (__a, __b, 1);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vqshluq_n_s64 (int64x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vqshlu_nv2di (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vshll_n_s8 (int8x8_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vshll_nv8qi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vshll_n_s16 (int16x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vshll_nv4hi (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vshll_n_s32 (int32x2_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vshll_nv2si (__a, __b, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vshll_n_u8 (uint8x8_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vshll_nv8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vshll_n_u16 (uint16x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vshll_nv4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vshll_n_u32 (uint32x2_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vshll_nv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vsra_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vsra_nv8qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vsra_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vsra_nv4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vsra_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vsra_nv2si (__a, __b, __c, 1);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vsra_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vsra_ndi (__a, __b, __c, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vsra_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vsra_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vsra_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vsra_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vsra_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vsra_nv2si ((int32x2_t) __a, (int32x2_t) __b, __c, 0);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vsra_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vsra_ndi ((int64x1_t) __a, (int64x1_t) __b, __c, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vsraq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vsra_nv16qi (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsraq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vsra_nv8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsraq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vsra_nv4si (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsraq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vsra_nv2di (__a, __b, __c, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vsraq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vsra_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsraq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vsra_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsraq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vsra_nv4si ((int32x4_t) __a, (int32x4_t) __b, __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsraq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vsra_nv2di ((int64x2_t) __a, (int64x2_t) __b, __c, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrsra_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vsra_nv8qi (__a, __b, __c, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrsra_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vsra_nv4hi (__a, __b, __c, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrsra_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vsra_nv2si (__a, __b, __c, 5);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vrsra_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vsra_ndi (__a, __b, __c, 5);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrsra_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vsra_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c, 4);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrsra_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vsra_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c, 4);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrsra_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vsra_nv2si ((int32x2_t) __a, (int32x2_t) __b, __c, 4);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vrsra_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vsra_ndi ((int64x1_t) __a, (int64x1_t) __b, __c, 4);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrsraq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vsra_nv16qi (__a, __b, __c, 5);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrsraq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vsra_nv8hi (__a, __b, __c, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vrsraq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vsra_nv4si (__a, __b, __c, 5);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vrsraq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vsra_nv2di (__a, __b, __c, 5);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrsraq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vsra_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c, 4);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrsraq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vsra_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c, 4);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrsraq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vsra_nv4si ((int32x4_t) __a, (int32x4_t) __b, __c, 4);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vrsraq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vsra_nv2di ((int64x2_t) __a, (int64x2_t) __b, __c, 4);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vsri_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vsri_nv8qi (__a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vsri_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vsri_nv4hi (__a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vsri_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vsri_nv2si (__a, __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vsri_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vsri_ndi (__a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vsri_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vsri_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vsri_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vsri_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vsri_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vsri_nv2si ((int32x2_t) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vsri_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vsri_ndi ((int64x1_t) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vsri_n_p8 (poly8x8_t __a, poly8x8_t __b, const int __c)
{
  return (poly8x8_t)__builtin_neon_vsri_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))