# 1 "jsimd_arm.c"
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/libjpeg-turbo/simd//"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "jsimd_arm.c"
# 18 "jsimd_arm.c"
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/arm_neon.h" 1 3 4
# 39 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/arm_neon.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stdint.h" 1 3 4


# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 1 3 4
# 26 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 1 3 4
# 323 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/predefs.h" 1 3 4
# 324 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 2 3 4
# 356 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/cdefs.h" 1 3 4
# 353 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/cdefs.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/wordsize.h" 1 3 4
# 354 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/cdefs.h" 2 3 4
# 357 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 2 3 4
# 388 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/gnu/stubs.h" 1 3 4
# 389 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/features.h" 2 3 4
# 27 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 2 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/wchar.h" 1 3 4
# 28 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 2 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/wordsize.h" 1 3 4
# 29 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 2 3 4
# 37 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 3 4
typedef signed char int8_t;
typedef short int int16_t;
typedef int int32_t;



__extension__
typedef long long int int64_t;




typedef unsigned char uint8_t;
typedef unsigned short int uint16_t;

typedef unsigned int uint32_t;





__extension__
typedef unsigned long long int uint64_t;






typedef signed char int_least8_t;
typedef short int int_least16_t;
typedef int int_least32_t;



__extension__
typedef long long int int_least64_t;



typedef unsigned char uint_least8_t;
typedef unsigned short int uint_least16_t;
typedef unsigned int uint_least32_t;



__extension__
typedef unsigned long long int uint_least64_t;






typedef signed char int_fast8_t;





typedef int int_fast16_t;
typedef int int_fast32_t;
__extension__
typedef long long int int_fast64_t;



typedef unsigned char uint_fast8_t;





typedef unsigned int uint_fast16_t;
typedef unsigned int uint_fast32_t;
__extension__
typedef unsigned long long int uint_fast64_t;
# 126 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 3 4
typedef int intptr_t;


typedef unsigned int uintptr_t;
# 138 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdint.h" 3 4
__extension__
typedef long long int intmax_t;
__extension__
typedef unsigned long long int uintmax_t;
# 4 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stdint.h" 2 3 4
# 40 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/arm_neon.h" 2 3 4

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__))
vsri_n_p16 (poly16x4_t __a, poly16x4_t __b, const int __c)
{
  return (poly16x4_t)__builtin_neon_vsri_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vsriq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vsri_nv16qi (__a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsriq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vsri_nv8hi (__a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsriq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vsri_nv4si (__a, __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsriq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vsri_nv2di (__a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vsriq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vsri_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsriq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vsri_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsriq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vsri_nv4si ((int32x4_t) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsriq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vsri_nv2di ((int64x2_t) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vsriq_n_p8 (poly8x16_t __a, poly8x16_t __b, const int __c)
{
  return (poly8x16_t)__builtin_neon_vsri_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vsriq_n_p16 (poly16x8_t __a, poly16x8_t __b, const int __c)
{
  return (poly16x8_t)__builtin_neon_vsri_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vsli_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vsli_nv8qi (__a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vsli_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vsli_nv4hi (__a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vsli_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vsli_nv2si (__a, __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vsli_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vsli_ndi (__a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vsli_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vsli_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vsli_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vsli_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vsli_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vsli_nv2si ((int32x2_t) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vsli_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vsli_ndi ((int64x1_t) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vsli_n_p8 (poly8x8_t __a, poly8x8_t __b, const int __c)
{
  return (poly8x8_t)__builtin_neon_vsli_nv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vsli_n_p16 (poly16x4_t __a, poly16x4_t __b, const int __c)
{
  return (poly16x4_t)__builtin_neon_vsli_nv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vsliq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vsli_nv16qi (__a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsliq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vsli_nv8hi (__a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsliq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vsli_nv4si (__a, __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsliq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vsli_nv2di (__a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vsliq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vsli_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsliq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vsli_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsliq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vsli_nv4si ((int32x4_t) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsliq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vsli_nv2di ((int64x2_t) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vsliq_n_p8 (poly8x16_t __a, poly8x16_t __b, const int __c)
{
  return (poly8x16_t)__builtin_neon_vsli_nv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vsliq_n_p16 (poly16x8_t __a, poly16x8_t __b, const int __c)
{
  return (poly16x8_t)__builtin_neon_vsli_nv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vabs_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vabsv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vabs_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vabsv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vabs_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vabsv2si (__a, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vabs_f32 (float32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vabsv2sf (__a, 3);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vabsq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vabsv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vabsq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vabsv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vabsq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vabsv4si (__a, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vabsq_f32 (float32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vabsv4sf (__a, 3);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqabs_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vqabsv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqabs_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vqabsv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqabs_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vqabsv2si (__a, 1);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqabsq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vqabsv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqabsq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vqabsv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqabsq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vqabsv4si (__a, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vneg_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vnegv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vneg_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vnegv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vneg_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vnegv2si (__a, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vneg_f32 (float32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vnegv2sf (__a, 3);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vnegq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vnegv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vnegq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vnegv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vnegq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vnegv4si (__a, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vnegq_f32 (float32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vnegv4sf (__a, 3);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqneg_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vqnegv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqneg_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vqnegv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqneg_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vqnegv2si (__a, 1);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vqnegq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vqnegv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqnegq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vqnegv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqnegq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vqnegv4si (__a, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmvn_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vmvnv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmvn_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vmvnv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmvn_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vmvnv2si (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmvn_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vmvnv8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmvn_u16 (uint16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vmvnv4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmvn_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vmvnv2si ((int32x2_t) __a, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vmvn_p8 (poly8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vmvnv8qi ((int8x8_t) __a, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmvnq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vmvnv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmvnq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vmvnv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmvnq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vmvnv4si (__a, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmvnq_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vmvnv16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmvnq_u16 (uint16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vmvnv8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmvnq_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vmvnv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vmvnq_p8 (poly8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vmvnv16qi ((int8x16_t) __a, 2);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vcls_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vclsv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vcls_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vclsv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vcls_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vclsv2si (__a, 1);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vclsq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vclsv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vclsq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vclsv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vclsq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vclsv4si (__a, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vclz_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vclzv8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vclz_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vclzv4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vclz_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vclzv2si (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vclz_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vclzv8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vclz_u16 (uint16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vclzv4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vclz_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vclzv2si ((int32x2_t) __a, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vclzq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vclzv16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vclzq_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vclzv8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vclzq_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vclzv4si (__a, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vclzq_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vclzv16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vclzq_u16 (uint16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vclzv8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vclzq_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vclzv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vcnt_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vcntv8qi (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcnt_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vcntv8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vcnt_p8 (poly8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vcntv8qi ((int8x8_t) __a, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vcntq_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vcntv16qi (__a, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcntq_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vcntv16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vcntq_p8 (poly8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vcntv16qi ((int8x16_t) __a, 2);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vrecpe_f32 (float32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vrecpev2sf (__a, 3);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrecpe_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vrecpev2si ((int32x2_t) __a, 0);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vrecpeq_f32 (float32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vrecpev4sf (__a, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrecpeq_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vrecpev4si ((int32x4_t) __a, 0);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vrsqrte_f32 (float32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vrsqrtev2sf (__a, 3);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrsqrte_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vrsqrtev2si ((int32x2_t) __a, 0);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vrsqrteq_f32 (float32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vrsqrtev4sf (__a, 3);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrsqrteq_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vrsqrtev4si ((int32x4_t) __a, 0);
}

__extension__ static __inline int8_t __attribute__ ((__always_inline__))
vget_lane_s8 (int8x8_t __a, const int __b)
{
  return (int8_t)__builtin_neon_vget_lanev8qi (__a, __b, 1);
}

__extension__ static __inline int16_t __attribute__ ((__always_inline__))
vget_lane_s16 (int16x4_t __a, const int __b)
{
  return (int16_t)__builtin_neon_vget_lanev4hi (__a, __b, 1);
}

__extension__ static __inline int32_t __attribute__ ((__always_inline__))
vget_lane_s32 (int32x2_t __a, const int __b)
{
  return (int32_t)__builtin_neon_vget_lanev2si (__a, __b, 1);
}

__extension__ static __inline float32_t __attribute__ ((__always_inline__))
vget_lane_f32 (float32x2_t __a, const int __b)
{
  return (float32_t)__builtin_neon_vget_lanev2sf (__a, __b, 3);
}

__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
vget_lane_u8 (uint8x8_t __a, const int __b)
{
  return (uint8_t)__builtin_neon_vget_lanev8qi ((int8x8_t) __a, __b, 0);
}

__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
vget_lane_u16 (uint16x4_t __a, const int __b)
{
  return (uint16_t)__builtin_neon_vget_lanev4hi ((int16x4_t) __a, __b, 0);
}

__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
vget_lane_u32 (uint32x2_t __a, const int __b)
{
  return (uint32_t)__builtin_neon_vget_lanev2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
vget_lane_p8 (poly8x8_t __a, const int __b)
{
  return (poly8_t)__builtin_neon_vget_lanev8qi ((int8x8_t) __a, __b, 2);
}

__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
vget_lane_p16 (poly16x4_t __a, const int __b)
{
  return (poly16_t)__builtin_neon_vget_lanev4hi ((int16x4_t) __a, __b, 2);
}

__extension__ static __inline int64_t __attribute__ ((__always_inline__))
vget_lane_s64 (int64x1_t __a, const int __b)
{
  return (int64_t)__builtin_neon_vget_lanedi (__a, __b, 1);
}

__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
vget_lane_u64 (uint64x1_t __a, const int __b)
{
  return (uint64_t)__builtin_neon_vget_lanedi ((int64x1_t) __a, __b, 0);
}

__extension__ static __inline int8_t __attribute__ ((__always_inline__))
vgetq_lane_s8 (int8x16_t __a, const int __b)
{
  return (int8_t)__builtin_neon_vget_lanev16qi (__a, __b, 1);
}

__extension__ static __inline int16_t __attribute__ ((__always_inline__))
vgetq_lane_s16 (int16x8_t __a, const int __b)
{
  return (int16_t)__builtin_neon_vget_lanev8hi (__a, __b, 1);
}

__extension__ static __inline int32_t __attribute__ ((__always_inline__))
vgetq_lane_s32 (int32x4_t __a, const int __b)
{
  return (int32_t)__builtin_neon_vget_lanev4si (__a, __b, 1);
}

__extension__ static __inline float32_t __attribute__ ((__always_inline__))
vgetq_lane_f32 (float32x4_t __a, const int __b)
{
  return (float32_t)__builtin_neon_vget_lanev4sf (__a, __b, 3);
}

__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
vgetq_lane_u8 (uint8x16_t __a, const int __b)
{
  return (uint8_t)__builtin_neon_vget_lanev16qi ((int8x16_t) __a, __b, 0);
}

__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
vgetq_lane_u16 (uint16x8_t __a, const int __b)
{
  return (uint16_t)__builtin_neon_vget_lanev8hi ((int16x8_t) __a, __b, 0);
}

__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
vgetq_lane_u32 (uint32x4_t __a, const int __b)
{
  return (uint32_t)__builtin_neon_vget_lanev4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
vgetq_lane_p8 (poly8x16_t __a, const int __b)
{
  return (poly8_t)__builtin_neon_vget_lanev16qi ((int8x16_t) __a, __b, 2);
}

__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
vgetq_lane_p16 (poly16x8_t __a, const int __b)
{
  return (poly16_t)__builtin_neon_vget_lanev8hi ((int16x8_t) __a, __b, 2);
}

__extension__ static __inline int64_t __attribute__ ((__always_inline__))
vgetq_lane_s64 (int64x2_t __a, const int __b)
{
  return (int64_t)__builtin_neon_vget_lanev2di (__a, __b, 1);
}

__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
vgetq_lane_u64 (uint64x2_t __a, const int __b)
{
  return (uint64_t)__builtin_neon_vget_lanev2di ((int64x2_t) __a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vset_lane_s8 (int8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vset_lanev8qi ((__builtin_neon_qi) __a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vset_lane_s16 (int16_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vset_lanev4hi ((__builtin_neon_hi) __a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vset_lane_s32 (int32_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vset_lanev2si ((__builtin_neon_si) __a, __b, __c);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vset_lane_f32 (float32_t __a, float32x2_t __b, const int __c)
{
  return (float32x2_t)__builtin_neon_vset_lanev2sf ((__builtin_neon_sf) __a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vset_lane_u8 (uint8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vset_lanev8qi ((__builtin_neon_qi) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vset_lane_u16 (uint16_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vset_lanev4hi ((__builtin_neon_hi) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vset_lane_u32 (uint32_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vset_lanev2si ((__builtin_neon_si) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vset_lane_p8 (poly8_t __a, poly8x8_t __b, const int __c)
{
  return (poly8x8_t)__builtin_neon_vset_lanev8qi ((__builtin_neon_qi) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vset_lane_p16 (poly16_t __a, poly16x4_t __b, const int __c)
{
  return (poly16x4_t)__builtin_neon_vset_lanev4hi ((__builtin_neon_hi) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vset_lane_s64 (int64_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vset_lanedi ((__builtin_neon_di) __a, __b, __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vset_lane_u64 (uint64_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vset_lanedi ((__builtin_neon_di) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vsetq_lane_s8 (int8_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vset_lanev16qi ((__builtin_neon_qi) __a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vsetq_lane_s16 (int16_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vset_lanev8hi ((__builtin_neon_hi) __a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vsetq_lane_s32 (int32_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vset_lanev4si ((__builtin_neon_si) __a, __b, __c);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vsetq_lane_f32 (float32_t __a, float32x4_t __b, const int __c)
{
  return (float32x4_t)__builtin_neon_vset_lanev4sf ((__builtin_neon_sf) __a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vsetq_lane_u8 (uint8_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vset_lanev16qi ((__builtin_neon_qi) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vsetq_lane_u16 (uint16_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vset_lanev8hi ((__builtin_neon_hi) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vsetq_lane_u32 (uint32_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vset_lanev4si ((__builtin_neon_si) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vsetq_lane_p8 (poly8_t __a, poly8x16_t __b, const int __c)
{
  return (poly8x16_t)__builtin_neon_vset_lanev16qi ((__builtin_neon_qi) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vsetq_lane_p16 (poly16_t __a, poly16x8_t __b, const int __c)
{
  return (poly16x8_t)__builtin_neon_vset_lanev8hi ((__builtin_neon_hi) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vsetq_lane_s64 (int64_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vset_lanev2di ((__builtin_neon_di) __a, __b, __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vsetq_lane_u64 (uint64_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vset_lanev2di ((__builtin_neon_di) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vcreate_s8 (uint64_t __a)
{
  return (int8x8_t)__builtin_neon_vcreatev8qi ((__builtin_neon_di) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vcreate_s16 (uint64_t __a)
{
  return (int16x4_t)__builtin_neon_vcreatev4hi ((__builtin_neon_di) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vcreate_s32 (uint64_t __a)
{
  return (int32x2_t)__builtin_neon_vcreatev2si ((__builtin_neon_di) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vcreate_s64 (uint64_t __a)
{
  return (int64x1_t)__builtin_neon_vcreatedi ((__builtin_neon_di) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vcreate_f32 (uint64_t __a)
{
  return (float32x2_t)__builtin_neon_vcreatev2sf ((__builtin_neon_di) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vcreate_u8 (uint64_t __a)
{
  return (uint8x8_t)__builtin_neon_vcreatev8qi ((__builtin_neon_di) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vcreate_u16 (uint64_t __a)
{
  return (uint16x4_t)__builtin_neon_vcreatev4hi ((__builtin_neon_di) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcreate_u32 (uint64_t __a)
{
  return (uint32x2_t)__builtin_neon_vcreatev2si ((__builtin_neon_di) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vcreate_u64 (uint64_t __a)
{
  return (uint64x1_t)__builtin_neon_vcreatedi ((__builtin_neon_di) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vcreate_p8 (uint64_t __a)
{
  return (poly8x8_t)__builtin_neon_vcreatev8qi ((__builtin_neon_di) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vcreate_p16 (uint64_t __a)
{
  return (poly16x4_t)__builtin_neon_vcreatev4hi ((__builtin_neon_di) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vdup_n_s8 (int8_t __a)
{
  return (int8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vdup_n_s16 (int16_t __a)
{
  return (int16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vdup_n_s32 (int32_t __a)
{
  return (int32x2_t)__builtin_neon_vdup_nv2si ((__builtin_neon_si) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vdup_n_f32 (float32_t __a)
{
  return (float32x2_t)__builtin_neon_vdup_nv2sf ((__builtin_neon_sf) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vdup_n_u8 (uint8_t __a)
{
  return (uint8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vdup_n_u16 (uint16_t __a)
{
  return (uint16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vdup_n_u32 (uint32_t __a)
{
  return (uint32x2_t)__builtin_neon_vdup_nv2si ((__builtin_neon_si) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vdup_n_p8 (poly8_t __a)
{
  return (poly8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vdup_n_p16 (poly16_t __a)
{
  return (poly16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vdup_n_s64 (int64_t __a)
{
  return (int64x1_t)__builtin_neon_vdup_ndi ((__builtin_neon_di) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vdup_n_u64 (uint64_t __a)
{
  return (uint64x1_t)__builtin_neon_vdup_ndi ((__builtin_neon_di) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vdupq_n_s8 (int8_t __a)
{
  return (int8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vdupq_n_s16 (int16_t __a)
{
  return (int16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vdupq_n_s32 (int32_t __a)
{
  return (int32x4_t)__builtin_neon_vdup_nv4si ((__builtin_neon_si) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vdupq_n_f32 (float32_t __a)
{
  return (float32x4_t)__builtin_neon_vdup_nv4sf ((__builtin_neon_sf) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vdupq_n_u8 (uint8_t __a)
{
  return (uint8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vdupq_n_u16 (uint16_t __a)
{
  return (uint16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vdupq_n_u32 (uint32_t __a)
{
  return (uint32x4_t)__builtin_neon_vdup_nv4si ((__builtin_neon_si) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vdupq_n_p8 (poly8_t __a)
{
  return (poly8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vdupq_n_p16 (poly16_t __a)
{
  return (poly16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vdupq_n_s64 (int64_t __a)
{
  return (int64x2_t)__builtin_neon_vdup_nv2di ((__builtin_neon_di) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vdupq_n_u64 (uint64_t __a)
{
  return (uint64x2_t)__builtin_neon_vdup_nv2di ((__builtin_neon_di) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmov_n_s8 (int8_t __a)
{
  return (int8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmov_n_s16 (int16_t __a)
{
  return (int16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmov_n_s32 (int32_t __a)
{
  return (int32x2_t)__builtin_neon_vdup_nv2si ((__builtin_neon_si) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmov_n_f32 (float32_t __a)
{
  return (float32x2_t)__builtin_neon_vdup_nv2sf ((__builtin_neon_sf) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmov_n_u8 (uint8_t __a)
{
  return (uint8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmov_n_u16 (uint16_t __a)
{
  return (uint16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmov_n_u32 (uint32_t __a)
{
  return (uint32x2_t)__builtin_neon_vdup_nv2si ((__builtin_neon_si) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vmov_n_p8 (poly8_t __a)
{
  return (poly8x8_t)__builtin_neon_vdup_nv8qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vmov_n_p16 (poly16_t __a)
{
  return (poly16x4_t)__builtin_neon_vdup_nv4hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vmov_n_s64 (int64_t __a)
{
  return (int64x1_t)__builtin_neon_vdup_ndi ((__builtin_neon_di) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vmov_n_u64 (uint64_t __a)
{
  return (uint64x1_t)__builtin_neon_vdup_ndi ((__builtin_neon_di) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vmovq_n_s8 (int8_t __a)
{
  return (int8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmovq_n_s16 (int16_t __a)
{
  return (int16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmovq_n_s32 (int32_t __a)
{
  return (int32x4_t)__builtin_neon_vdup_nv4si ((__builtin_neon_si) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmovq_n_f32 (float32_t __a)
{
  return (float32x4_t)__builtin_neon_vdup_nv4sf ((__builtin_neon_sf) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vmovq_n_u8 (uint8_t __a)
{
  return (uint8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmovq_n_u16 (uint16_t __a)
{
  return (uint16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmovq_n_u32 (uint32_t __a)
{
  return (uint32x4_t)__builtin_neon_vdup_nv4si ((__builtin_neon_si) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vmovq_n_p8 (poly8_t __a)
{
  return (poly8x16_t)__builtin_neon_vdup_nv16qi ((__builtin_neon_qi) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vmovq_n_p16 (poly16_t __a)
{
  return (poly16x8_t)__builtin_neon_vdup_nv8hi ((__builtin_neon_hi) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmovq_n_s64 (int64_t __a)
{
  return (int64x2_t)__builtin_neon_vdup_nv2di ((__builtin_neon_di) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmovq_n_u64 (uint64_t __a)
{
  return (uint64x2_t)__builtin_neon_vdup_nv2di ((__builtin_neon_di) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vdup_lane_s8 (int8x8_t __a, const int __b)
{
  return (int8x8_t)__builtin_neon_vdup_lanev8qi (__a, __b);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vdup_lane_s16 (int16x4_t __a, const int __b)
{
  return (int16x4_t)__builtin_neon_vdup_lanev4hi (__a, __b);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vdup_lane_s32 (int32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vdup_lanev2si (__a, __b);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vdup_lane_f32 (float32x2_t __a, const int __b)
{
  return (float32x2_t)__builtin_neon_vdup_lanev2sf (__a, __b);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vdup_lane_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x8_t)__builtin_neon_vdup_lanev8qi ((int8x8_t) __a, __b);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vdup_lane_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x4_t)__builtin_neon_vdup_lanev4hi ((int16x4_t) __a, __b);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vdup_lane_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vdup_lanev2si ((int32x2_t) __a, __b);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vdup_lane_p8 (poly8x8_t __a, const int __b)
{
  return (poly8x8_t)__builtin_neon_vdup_lanev8qi ((int8x8_t) __a, __b);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vdup_lane_p16 (poly16x4_t __a, const int __b)
{
  return (poly16x4_t)__builtin_neon_vdup_lanev4hi ((int16x4_t) __a, __b);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vdup_lane_s64 (int64x1_t __a, const int __b)
{
  return (int64x1_t)__builtin_neon_vdup_lanedi (__a, __b);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vdup_lane_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x1_t)__builtin_neon_vdup_lanedi ((int64x1_t) __a, __b);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vdupq_lane_s8 (int8x8_t __a, const int __b)
{
  return (int8x16_t)__builtin_neon_vdup_lanev16qi (__a, __b);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vdupq_lane_s16 (int16x4_t __a, const int __b)
{
  return (int16x8_t)__builtin_neon_vdup_lanev8hi (__a, __b);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vdupq_lane_s32 (int32x2_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vdup_lanev4si (__a, __b);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vdupq_lane_f32 (float32x2_t __a, const int __b)
{
  return (float32x4_t)__builtin_neon_vdup_lanev4sf (__a, __b);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vdupq_lane_u8 (uint8x8_t __a, const int __b)
{
  return (uint8x16_t)__builtin_neon_vdup_lanev16qi ((int8x8_t) __a, __b);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vdupq_lane_u16 (uint16x4_t __a, const int __b)
{
  return (uint16x8_t)__builtin_neon_vdup_lanev8hi ((int16x4_t) __a, __b);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vdupq_lane_u32 (uint32x2_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vdup_lanev4si ((int32x2_t) __a, __b);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vdupq_lane_p8 (poly8x8_t __a, const int __b)
{
  return (poly8x16_t)__builtin_neon_vdup_lanev16qi ((int8x8_t) __a, __b);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vdupq_lane_p16 (poly16x4_t __a, const int __b)
{
  return (poly16x8_t)__builtin_neon_vdup_lanev8hi ((int16x4_t) __a, __b);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vdupq_lane_s64 (int64x1_t __a, const int __b)
{
  return (int64x2_t)__builtin_neon_vdup_lanev2di (__a, __b);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vdupq_lane_u64 (uint64x1_t __a, const int __b)
{
  return (uint64x2_t)__builtin_neon_vdup_lanev2di ((int64x1_t) __a, __b);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vcombine_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x16_t)__builtin_neon_vcombinev8qi (__a, __b);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vcombine_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x8_t)__builtin_neon_vcombinev4hi (__a, __b);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vcombine_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x4_t)__builtin_neon_vcombinev2si (__a, __b);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vcombine_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x2_t)__builtin_neon_vcombinedi (__a, __b);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vcombine_f32 (float32x2_t __a, float32x2_t __b)
{
  return (float32x4_t)__builtin_neon_vcombinev2sf (__a, __b);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vcombine_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x16_t)__builtin_neon_vcombinev8qi ((int8x8_t) __a, (int8x8_t) __b);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vcombine_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x8_t)__builtin_neon_vcombinev4hi ((int16x4_t) __a, (int16x4_t) __b);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcombine_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x4_t)__builtin_neon_vcombinev2si ((int32x2_t) __a, (int32x2_t) __b);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vcombine_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x2_t)__builtin_neon_vcombinedi ((int64x1_t) __a, (int64x1_t) __b);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vcombine_p8 (poly8x8_t __a, poly8x8_t __b)
{
  return (poly8x16_t)__builtin_neon_vcombinev8qi ((int8x8_t) __a, (int8x8_t) __b);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vcombine_p16 (poly16x4_t __a, poly16x4_t __b)
{
  return (poly16x8_t)__builtin_neon_vcombinev4hi ((int16x4_t) __a, (int16x4_t) __b);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vget_high_s8 (int8x16_t __a)
{
  return (int8x8_t)__builtin_neon_vget_highv16qi (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vget_high_s16 (int16x8_t __a)
{
  return (int16x4_t)__builtin_neon_vget_highv8hi (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vget_high_s32 (int32x4_t __a)
{
  return (int32x2_t)__builtin_neon_vget_highv4si (__a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vget_high_s64 (int64x2_t __a)
{
  return (int64x1_t)__builtin_neon_vget_highv2di (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vget_high_f32 (float32x4_t __a)
{
  return (float32x2_t)__builtin_neon_vget_highv4sf (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vget_high_u8 (uint8x16_t __a)
{
  return (uint8x8_t)__builtin_neon_vget_highv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vget_high_u16 (uint16x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vget_highv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vget_high_u32 (uint32x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vget_highv4si ((int32x4_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vget_high_u64 (uint64x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vget_highv2di ((int64x2_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vget_high_p8 (poly8x16_t __a)
{
  return (poly8x8_t)__builtin_neon_vget_highv16qi ((int8x16_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vget_high_p16 (poly16x8_t __a)
{
  return (poly16x4_t)__builtin_neon_vget_highv8hi ((int16x8_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vget_low_s8 (int8x16_t __a)
{
  return (int8x8_t)__builtin_neon_vget_lowv16qi (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vget_low_s16 (int16x8_t __a)
{
  return (int16x4_t)__builtin_neon_vget_lowv8hi (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vget_low_s32 (int32x4_t __a)
{
  return (int32x2_t)__builtin_neon_vget_lowv4si (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vget_low_f32 (float32x4_t __a)
{
  return (float32x2_t)__builtin_neon_vget_lowv4sf (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vget_low_u8 (uint8x16_t __a)
{
  return (uint8x8_t)__builtin_neon_vget_lowv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vget_low_u16 (uint16x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vget_lowv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vget_low_u32 (uint32x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vget_lowv4si ((int32x4_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vget_low_p8 (poly8x16_t __a)
{
  return (poly8x8_t)__builtin_neon_vget_lowv16qi ((int8x16_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vget_low_p16 (poly16x8_t __a)
{
  return (poly16x4_t)__builtin_neon_vget_lowv8hi ((int16x8_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vget_low_s64 (int64x2_t __a)
{
  return (int64x1_t)__builtin_neon_vget_lowv2di (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vget_low_u64 (uint64x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vget_lowv2di ((int64x2_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vcvt_s32_f32 (float32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vcvtv2sf (__a, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vcvt_f32_s32 (int32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vcvtv2si (__a, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vcvt_f32_u32 (uint32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vcvtv2si ((int32x2_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcvt_u32_f32 (float32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vcvtv2sf (__a, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vcvtq_s32_f32 (float32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vcvtv4sf (__a, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vcvtq_f32_s32 (int32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vcvtv4si (__a, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vcvtq_f32_u32 (uint32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vcvtv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcvtq_u32_f32 (float32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vcvtv4sf (__a, 0);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vcvt_n_s32_f32 (float32x2_t __a, const int __b)
{
  return (int32x2_t)__builtin_neon_vcvt_nv2sf (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vcvt_n_f32_s32 (int32x2_t __a, const int __b)
{
  return (float32x2_t)__builtin_neon_vcvt_nv2si (__a, __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vcvt_n_f32_u32 (uint32x2_t __a, const int __b)
{
  return (float32x2_t)__builtin_neon_vcvt_nv2si ((int32x2_t) __a, __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vcvt_n_u32_f32 (float32x2_t __a, const int __b)
{
  return (uint32x2_t)__builtin_neon_vcvt_nv2sf (__a, __b, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vcvtq_n_s32_f32 (float32x4_t __a, const int __b)
{
  return (int32x4_t)__builtin_neon_vcvt_nv4sf (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vcvtq_n_f32_s32 (int32x4_t __a, const int __b)
{
  return (float32x4_t)__builtin_neon_vcvt_nv4si (__a, __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vcvtq_n_f32_u32 (uint32x4_t __a, const int __b)
{
  return (float32x4_t)__builtin_neon_vcvt_nv4si ((int32x4_t) __a, __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vcvtq_n_u32_f32 (float32x4_t __a, const int __b)
{
  return (uint32x4_t)__builtin_neon_vcvt_nv4sf (__a, __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vmovn_s16 (int16x8_t __a)
{
  return (int8x8_t)__builtin_neon_vmovnv8hi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmovn_s32 (int32x4_t __a)
{
  return (int16x4_t)__builtin_neon_vmovnv4si (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmovn_s64 (int64x2_t __a)
{
  return (int32x2_t)__builtin_neon_vmovnv2di (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vmovn_u16 (uint16x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vmovnv8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmovn_u32 (uint32x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vmovnv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmovn_u64 (uint64x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vmovnv2di ((int64x2_t) __a, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vqmovn_s16 (int16x8_t __a)
{
  return (int8x8_t)__builtin_neon_vqmovnv8hi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqmovn_s32 (int32x4_t __a)
{
  return (int16x4_t)__builtin_neon_vqmovnv4si (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqmovn_s64 (int64x2_t __a)
{
  return (int32x2_t)__builtin_neon_vqmovnv2di (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqmovn_u16 (uint16x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vqmovnv8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqmovn_u32 (uint32x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vqmovnv4si ((int32x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqmovn_u64 (uint64x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vqmovnv2di ((int64x2_t) __a, 0);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vqmovun_s16 (int16x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vqmovunv8hi (__a, 1);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vqmovun_s32 (int32x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vqmovunv4si (__a, 1);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vqmovun_s64 (int64x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vqmovunv2di (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmovl_s8 (int8x8_t __a)
{
  return (int16x8_t)__builtin_neon_vmovlv8qi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmovl_s16 (int16x4_t __a)
{
  return (int32x4_t)__builtin_neon_vmovlv4hi (__a, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmovl_s32 (int32x2_t __a)
{
  return (int64x2_t)__builtin_neon_vmovlv2si (__a, 1);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmovl_u8 (uint8x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vmovlv8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmovl_u16 (uint16x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vmovlv4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmovl_u32 (uint32x2_t __a)
{
  return (uint64x2_t)__builtin_neon_vmovlv2si ((int32x2_t) __a, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbl1_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vtbl1v8qi (__a, __b);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbl1_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vtbl1v8qi ((int8x8_t) __a, (int8x8_t) __b);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbl1_p8 (poly8x8_t __a, uint8x8_t __b)
{
  return (poly8x8_t)__builtin_neon_vtbl1v8qi ((int8x8_t) __a, (int8x8_t) __b);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbl2_s8 (int8x8x2_t __a, int8x8_t __b)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __au = { __a };
  return (int8x8_t)__builtin_neon_vtbl2v8qi (__au.__o, __b);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbl2_u8 (uint8x8x2_t __a, uint8x8_t __b)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __au = { __a };
  return (uint8x8_t)__builtin_neon_vtbl2v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbl2_p8 (poly8x8x2_t __a, uint8x8_t __b)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __au = { __a };
  return (poly8x8_t)__builtin_neon_vtbl2v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbl3_s8 (int8x8x3_t __a, int8x8_t __b)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __au = { __a };
  return (int8x8_t)__builtin_neon_vtbl3v8qi (__au.__o, __b);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbl3_u8 (uint8x8x3_t __a, uint8x8_t __b)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __au = { __a };
  return (uint8x8_t)__builtin_neon_vtbl3v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbl3_p8 (poly8x8x3_t __a, uint8x8_t __b)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __au = { __a };
  return (poly8x8_t)__builtin_neon_vtbl3v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbl4_s8 (int8x8x4_t __a, int8x8_t __b)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __au = { __a };
  return (int8x8_t)__builtin_neon_vtbl4v8qi (__au.__o, __b);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbl4_u8 (uint8x8x4_t __a, uint8x8_t __b)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __au = { __a };
  return (uint8x8_t)__builtin_neon_vtbl4v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbl4_p8 (poly8x8x4_t __a, uint8x8_t __b)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __au = { __a };
  return (poly8x8_t)__builtin_neon_vtbl4v8qi (__au.__o, (int8x8_t) __b);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbx1_s8 (int8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int8x8_t)__builtin_neon_vtbx1v8qi (__a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbx1_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t)__builtin_neon_vtbx1v8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbx1_p8 (poly8x8_t __a, poly8x8_t __b, uint8x8_t __c)
{
  return (poly8x8_t)__builtin_neon_vtbx1v8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbx2_s8 (int8x8_t __a, int8x8x2_t __b, int8x8_t __c)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  return (int8x8_t)__builtin_neon_vtbx2v8qi (__a, __bu.__o, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbx2_u8 (uint8x8_t __a, uint8x8x2_t __b, uint8x8_t __c)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  return (uint8x8_t)__builtin_neon_vtbx2v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbx2_p8 (poly8x8_t __a, poly8x8x2_t __b, uint8x8_t __c)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  return (poly8x8_t)__builtin_neon_vtbx2v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbx3_s8 (int8x8_t __a, int8x8x3_t __b, int8x8_t __c)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  return (int8x8_t)__builtin_neon_vtbx3v8qi (__a, __bu.__o, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbx3_u8 (uint8x8_t __a, uint8x8x3_t __b, uint8x8_t __c)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  return (uint8x8_t)__builtin_neon_vtbx3v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbx3_p8 (poly8x8_t __a, poly8x8x3_t __b, uint8x8_t __c)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  return (poly8x8_t)__builtin_neon_vtbx3v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vtbx4_s8 (int8x8_t __a, int8x8x4_t __b, int8x8_t __c)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  return (int8x8_t)__builtin_neon_vtbx4v8qi (__a, __bu.__o, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vtbx4_u8 (uint8x8_t __a, uint8x8x4_t __b, uint8x8_t __c)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  return (uint8x8_t)__builtin_neon_vtbx4v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vtbx4_p8 (poly8x8_t __a, poly8x8x4_t __b, uint8x8_t __c)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  return (poly8x8_t)__builtin_neon_vtbx4v8qi ((int8x8_t) __a, __bu.__o, (int8x8_t) __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmul_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vmul_lanev4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmul_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vmul_lanev2si (__a, __b, __c, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmul_lane_f32 (float32x2_t __a, float32x2_t __b, const int __c)
{
  return (float32x2_t)__builtin_neon_vmul_lanev2sf (__a, __b, __c, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmul_lane_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vmul_lanev4hi ((int16x4_t) __a, (int16x4_t) __b, __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmul_lane_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vmul_lanev2si ((int32x2_t) __a, (int32x2_t) __b, __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmulq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vmul_lanev8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmulq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vmul_lanev4si (__a, __b, __c, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmulq_lane_f32 (float32x4_t __a, float32x2_t __b, const int __c)
{
  return (float32x4_t)__builtin_neon_vmul_lanev4sf (__a, __b, __c, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmulq_lane_u16 (uint16x8_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vmul_lanev8hi ((int16x8_t) __a, (int16x4_t) __b, __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmulq_lane_u32 (uint32x4_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vmul_lanev4si ((int32x4_t) __a, (int32x2_t) __b, __c, 0);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmla_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int16x4_t)__builtin_neon_vmla_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmla_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int32x2_t)__builtin_neon_vmla_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmla_lane_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c, const int __d)
{
  return (float32x2_t)__builtin_neon_vmla_lanev2sf (__a, __b, __c, __d, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmla_lane_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c, const int __d)
{
  return (uint16x4_t)__builtin_neon_vmla_lanev4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmla_lane_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c, const int __d)
{
  return (uint32x2_t)__builtin_neon_vmla_lanev2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlaq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
{
  return (int16x8_t)__builtin_neon_vmla_lanev8hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlaq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vmla_lanev4si (__a, __b, __c, __d, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlaq_lane_f32 (float32x4_t __a, float32x4_t __b, float32x2_t __c, const int __d)
{
  return (float32x4_t)__builtin_neon_vmla_lanev4sf (__a, __b, __c, __d, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlaq_lane_u16 (uint16x8_t __a, uint16x8_t __b, uint16x4_t __c, const int __d)
{
  return (uint16x8_t)__builtin_neon_vmla_lanev8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlaq_lane_u32 (uint32x4_t __a, uint32x4_t __b, uint32x2_t __c, const int __d)
{
  return (uint32x4_t)__builtin_neon_vmla_lanev4si ((int32x4_t) __a, (int32x4_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlal_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vmlal_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlal_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int64x2_t)__builtin_neon_vmlal_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlal_lane_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c, const int __d)
{
  return (uint32x4_t)__builtin_neon_vmlal_lanev4hi ((int32x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlal_lane_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c, const int __d)
{
  return (uint64x2_t)__builtin_neon_vmlal_lanev2si ((int64x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlal_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vqdmlal_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlal_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int64x2_t)__builtin_neon_vqdmlal_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmls_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int16x4_t)__builtin_neon_vmls_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmls_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int32x2_t)__builtin_neon_vmls_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmls_lane_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c, const int __d)
{
  return (float32x2_t)__builtin_neon_vmls_lanev2sf (__a, __b, __c, __d, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmls_lane_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c, const int __d)
{
  return (uint16x4_t)__builtin_neon_vmls_lanev4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmls_lane_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c, const int __d)
{
  return (uint32x2_t)__builtin_neon_vmls_lanev2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlsq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
{
  return (int16x8_t)__builtin_neon_vmls_lanev8hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vmls_lanev4si (__a, __b, __c, __d, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlsq_lane_f32 (float32x4_t __a, float32x4_t __b, float32x2_t __c, const int __d)
{
  return (float32x4_t)__builtin_neon_vmls_lanev4sf (__a, __b, __c, __d, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlsq_lane_u16 (uint16x8_t __a, uint16x8_t __b, uint16x4_t __c, const int __d)
{
  return (uint16x8_t)__builtin_neon_vmls_lanev8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsq_lane_u32 (uint32x4_t __a, uint32x4_t __b, uint32x2_t __c, const int __d)
{
  return (uint32x4_t)__builtin_neon_vmls_lanev4si ((int32x4_t) __a, (int32x4_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsl_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vmlsl_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlsl_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int64x2_t)__builtin_neon_vmlsl_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsl_lane_u16 (uint32x4_t __a, uint16x4_t __b, uint16x4_t __c, const int __d)
{
  return (uint32x4_t)__builtin_neon_vmlsl_lanev4hi ((int32x4_t) __a, (int16x4_t) __b, (int16x4_t) __c, __d, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlsl_lane_u32 (uint64x2_t __a, uint32x2_t __b, uint32x2_t __c, const int __d)
{
  return (uint64x2_t)__builtin_neon_vmlsl_lanev2si ((int64x2_t) __a, (int32x2_t) __b, (int32x2_t) __c, __d, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlsl_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
{
  return (int32x4_t)__builtin_neon_vqdmlsl_lanev4hi (__a, __b, __c, __d, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlsl_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
{
  return (int64x2_t)__builtin_neon_vqdmlsl_lanev2si (__a, __b, __c, __d, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmull_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vmull_lanev4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmull_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vmull_lanev2si (__a, __b, __c, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmull_lane_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vmull_lanev4hi ((int16x4_t) __a, (int16x4_t) __b, __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmull_lane_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vmull_lanev2si ((int32x2_t) __a, (int32x2_t) __b, __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmull_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vqdmull_lanev4hi (__a, __b, __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmull_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vqdmull_lanev2si (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqdmulhq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vqdmulh_lanev8hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmulhq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vqdmulh_lanev4si (__a, __b, __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqdmulh_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vqdmulh_lanev4hi (__a, __b, __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqdmulh_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vqdmulh_lanev2si (__a, __b, __c, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqrdmulhq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vqdmulh_lanev8hi (__a, __b, __c, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqrdmulhq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vqdmulh_lanev4si (__a, __b, __c, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqrdmulh_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vqdmulh_lanev4hi (__a, __b, __c, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqrdmulh_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vqdmulh_lanev2si (__a, __b, __c, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmul_n_s16 (int16x4_t __a, int16_t __b)
{
  return (int16x4_t)__builtin_neon_vmul_nv4hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmul_n_s32 (int32x2_t __a, int32_t __b)
{
  return (int32x2_t)__builtin_neon_vmul_nv2si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmul_n_f32 (float32x2_t __a, float32_t __b)
{
  return (float32x2_t)__builtin_neon_vmul_nv2sf (__a, (__builtin_neon_sf) __b, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmul_n_u16 (uint16x4_t __a, uint16_t __b)
{
  return (uint16x4_t)__builtin_neon_vmul_nv4hi ((int16x4_t) __a, (__builtin_neon_hi) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmul_n_u32 (uint32x2_t __a, uint32_t __b)
{
  return (uint32x2_t)__builtin_neon_vmul_nv2si ((int32x2_t) __a, (__builtin_neon_si) __b, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmulq_n_s16 (int16x8_t __a, int16_t __b)
{
  return (int16x8_t)__builtin_neon_vmul_nv8hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmulq_n_s32 (int32x4_t __a, int32_t __b)
{
  return (int32x4_t)__builtin_neon_vmul_nv4si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmulq_n_f32 (float32x4_t __a, float32_t __b)
{
  return (float32x4_t)__builtin_neon_vmul_nv4sf (__a, (__builtin_neon_sf) __b, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmulq_n_u16 (uint16x8_t __a, uint16_t __b)
{
  return (uint16x8_t)__builtin_neon_vmul_nv8hi ((int16x8_t) __a, (__builtin_neon_hi) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmulq_n_u32 (uint32x4_t __a, uint32_t __b)
{
  return (uint32x4_t)__builtin_neon_vmul_nv4si ((int32x4_t) __a, (__builtin_neon_si) __b, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmull_n_s16 (int16x4_t __a, int16_t __b)
{
  return (int32x4_t)__builtin_neon_vmull_nv4hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmull_n_s32 (int32x2_t __a, int32_t __b)
{
  return (int64x2_t)__builtin_neon_vmull_nv2si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmull_n_u16 (uint16x4_t __a, uint16_t __b)
{
  return (uint32x4_t)__builtin_neon_vmull_nv4hi ((int16x4_t) __a, (__builtin_neon_hi) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmull_n_u32 (uint32x2_t __a, uint32_t __b)
{
  return (uint64x2_t)__builtin_neon_vmull_nv2si ((int32x2_t) __a, (__builtin_neon_si) __b, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmull_n_s16 (int16x4_t __a, int16_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmull_nv4hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmull_n_s32 (int32x2_t __a, int32_t __b)
{
  return (int64x2_t)__builtin_neon_vqdmull_nv2si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqdmulhq_n_s16 (int16x8_t __a, int16_t __b)
{
  return (int16x8_t)__builtin_neon_vqdmulh_nv8hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmulhq_n_s32 (int32x4_t __a, int32_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmulh_nv4si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqdmulh_n_s16 (int16x4_t __a, int16_t __b)
{
  return (int16x4_t)__builtin_neon_vqdmulh_nv4hi (__a, (__builtin_neon_hi) __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqdmulh_n_s32 (int32x2_t __a, int32_t __b)
{
  return (int32x2_t)__builtin_neon_vqdmulh_nv2si (__a, (__builtin_neon_si) __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vqrdmulhq_n_s16 (int16x8_t __a, int16_t __b)
{
  return (int16x8_t)__builtin_neon_vqdmulh_nv8hi (__a, (__builtin_neon_hi) __b, 5);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqrdmulhq_n_s32 (int32x4_t __a, int32_t __b)
{
  return (int32x4_t)__builtin_neon_vqdmulh_nv4si (__a, (__builtin_neon_si) __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vqrdmulh_n_s16 (int16x4_t __a, int16_t __b)
{
  return (int16x4_t)__builtin_neon_vqdmulh_nv4hi (__a, (__builtin_neon_hi) __b, 5);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vqrdmulh_n_s32 (int32x2_t __a, int32_t __b)
{
  return (int32x2_t)__builtin_neon_vqdmulh_nv2si (__a, (__builtin_neon_si) __b, 5);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmla_n_s16 (int16x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int16x4_t)__builtin_neon_vmla_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmla_n_s32 (int32x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int32x2_t)__builtin_neon_vmla_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmla_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
{
  return (float32x2_t)__builtin_neon_vmla_nv2sf (__a, __b, (__builtin_neon_sf) __c, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmla_n_u16 (uint16x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint16x4_t)__builtin_neon_vmla_nv4hi ((int16x4_t) __a, (int16x4_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmla_n_u32 (uint32x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint32x2_t)__builtin_neon_vmla_nv2si ((int32x2_t) __a, (int32x2_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlaq_n_s16 (int16x8_t __a, int16x8_t __b, int16_t __c)
{
  return (int16x8_t)__builtin_neon_vmla_nv8hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlaq_n_s32 (int32x4_t __a, int32x4_t __b, int32_t __c)
{
  return (int32x4_t)__builtin_neon_vmla_nv4si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlaq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
{
  return (float32x4_t)__builtin_neon_vmla_nv4sf (__a, __b, (__builtin_neon_sf) __c, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlaq_n_u16 (uint16x8_t __a, uint16x8_t __b, uint16_t __c)
{
  return (uint16x8_t)__builtin_neon_vmla_nv8hi ((int16x8_t) __a, (int16x8_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlaq_n_u32 (uint32x4_t __a, uint32x4_t __b, uint32_t __c)
{
  return (uint32x4_t)__builtin_neon_vmla_nv4si ((int32x4_t) __a, (int32x4_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlal_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int32x4_t)__builtin_neon_vmlal_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlal_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int64x2_t)__builtin_neon_vmlal_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlal_n_u16 (uint32x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlal_nv4hi ((int32x4_t) __a, (int16x4_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlal_n_u32 (uint64x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint64x2_t)__builtin_neon_vmlal_nv2si ((int64x2_t) __a, (int32x2_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlal_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int32x4_t)__builtin_neon_vqdmlal_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlal_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int64x2_t)__builtin_neon_vqdmlal_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vmls_n_s16 (int16x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int16x4_t)__builtin_neon_vmls_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vmls_n_s32 (int32x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int32x2_t)__builtin_neon_vmls_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vmls_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
{
  return (float32x2_t)__builtin_neon_vmls_nv2sf (__a, __b, (__builtin_neon_sf) __c, 3);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vmls_n_u16 (uint16x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint16x4_t)__builtin_neon_vmls_nv4hi ((int16x4_t) __a, (int16x4_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vmls_n_u32 (uint32x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint32x2_t)__builtin_neon_vmls_nv2si ((int32x2_t) __a, (int32x2_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vmlsq_n_s16 (int16x8_t __a, int16x8_t __b, int16_t __c)
{
  return (int16x8_t)__builtin_neon_vmls_nv8hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsq_n_s32 (int32x4_t __a, int32x4_t __b, int32_t __c)
{
  return (int32x4_t)__builtin_neon_vmls_nv4si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vmlsq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
{
  return (float32x4_t)__builtin_neon_vmls_nv4sf (__a, __b, (__builtin_neon_sf) __c, 3);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vmlsq_n_u16 (uint16x8_t __a, uint16x8_t __b, uint16_t __c)
{
  return (uint16x8_t)__builtin_neon_vmls_nv8hi ((int16x8_t) __a, (int16x8_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsq_n_u32 (uint32x4_t __a, uint32x4_t __b, uint32_t __c)
{
  return (uint32x4_t)__builtin_neon_vmls_nv4si ((int32x4_t) __a, (int32x4_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vmlsl_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int32x4_t)__builtin_neon_vmlsl_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vmlsl_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int64x2_t)__builtin_neon_vmlsl_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vmlsl_n_u16 (uint32x4_t __a, uint16x4_t __b, uint16_t __c)
{
  return (uint32x4_t)__builtin_neon_vmlsl_nv4hi ((int32x4_t) __a, (int16x4_t) __b, (__builtin_neon_hi) __c, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vmlsl_n_u32 (uint64x2_t __a, uint32x2_t __b, uint32_t __c)
{
  return (uint64x2_t)__builtin_neon_vmlsl_nv2si ((int64x2_t) __a, (int32x2_t) __b, (__builtin_neon_si) __c, 0);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vqdmlsl_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
{
  return (int32x4_t)__builtin_neon_vqdmlsl_nv4hi (__a, __b, (__builtin_neon_hi) __c, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vqdmlsl_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
{
  return (int64x2_t)__builtin_neon_vqdmlsl_nv2si (__a, __b, (__builtin_neon_si) __c, 1);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vext_s8 (int8x8_t __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vextv8qi (__a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vext_s16 (int16x4_t __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vextv4hi (__a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vext_s32 (int32x2_t __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vextv2si (__a, __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vext_s64 (int64x1_t __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vextdi (__a, __b, __c);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vext_f32 (float32x2_t __a, float32x2_t __b, const int __c)
{
  return (float32x2_t)__builtin_neon_vextv2sf (__a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vext_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vextv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vext_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vextv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vext_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vextv2si ((int32x2_t) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vext_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vextdi ((int64x1_t) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vext_p8 (poly8x8_t __a, poly8x8_t __b, const int __c)
{
  return (poly8x8_t)__builtin_neon_vextv8qi ((int8x8_t) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vext_p16 (poly16x4_t __a, poly16x4_t __b, const int __c)
{
  return (poly16x4_t)__builtin_neon_vextv4hi ((int16x4_t) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vextq_s8 (int8x16_t __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vextv16qi (__a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vextq_s16 (int16x8_t __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vextv8hi (__a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vextq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vextv4si (__a, __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vextq_s64 (int64x2_t __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vextv2di (__a, __b, __c);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vextq_f32 (float32x4_t __a, float32x4_t __b, const int __c)
{
  return (float32x4_t)__builtin_neon_vextv4sf (__a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vextq_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vextv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vextq_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vextv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vextq_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vextv4si ((int32x4_t) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vextq_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vextv2di ((int64x2_t) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vextq_p8 (poly8x16_t __a, poly8x16_t __b, const int __c)
{
  return (poly8x16_t)__builtin_neon_vextv16qi ((int8x16_t) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vextq_p16 (poly16x8_t __a, poly16x8_t __b, const int __c)
{
  return (poly16x8_t)__builtin_neon_vextv8hi ((int16x8_t) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrev64_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vrev64v8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrev64_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vrev64v4hi (__a, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vrev64_s32 (int32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vrev64v2si (__a, 1);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vrev64_f32 (float32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vrev64v2sf (__a, 3);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrev64_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vrev64v8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrev64_u16 (uint16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vrev64v4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vrev64_u32 (uint32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vrev64v2si ((int32x2_t) __a, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vrev64_p8 (poly8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vrev64v8qi ((int8x8_t) __a, 2);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vrev64_p16 (poly16x4_t __a)
{
  return (poly16x4_t)__builtin_neon_vrev64v4hi ((int16x4_t) __a, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrev64q_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vrev64v16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrev64q_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vrev64v8hi (__a, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vrev64q_s32 (int32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vrev64v4si (__a, 1);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vrev64q_f32 (float32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vrev64v4sf (__a, 3);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrev64q_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vrev64v16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrev64q_u16 (uint16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vrev64v8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vrev64q_u32 (uint32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vrev64v4si ((int32x4_t) __a, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vrev64q_p8 (poly8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vrev64v16qi ((int8x16_t) __a, 2);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vrev64q_p16 (poly16x8_t __a)
{
  return (poly16x8_t)__builtin_neon_vrev64v8hi ((int16x8_t) __a, 2);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrev32_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vrev32v8qi (__a, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vrev32_s16 (int16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vrev32v4hi (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrev32_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vrev32v8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vrev32_u16 (uint16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vrev32v4hi ((int16x4_t) __a, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vrev32_p8 (poly8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vrev32v8qi ((int8x8_t) __a, 2);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vrev32_p16 (poly16x4_t __a)
{
  return (poly16x4_t)__builtin_neon_vrev32v4hi ((int16x4_t) __a, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrev32q_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vrev32v16qi (__a, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vrev32q_s16 (int16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vrev32v8hi (__a, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrev32q_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vrev32v16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vrev32q_u16 (uint16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vrev32v8hi ((int16x8_t) __a, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vrev32q_p8 (poly8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vrev32v16qi ((int8x16_t) __a, 2);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vrev32q_p16 (poly16x8_t __a)
{
  return (poly16x8_t)__builtin_neon_vrev32v8hi ((int16x8_t) __a, 2);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vrev16_s8 (int8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vrev16v8qi (__a, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vrev16_u8 (uint8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vrev16v8qi ((int8x8_t) __a, 0);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vrev16_p8 (poly8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vrev16v8qi ((int8x8_t) __a, 2);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vrev16q_s8 (int8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vrev16v16qi (__a, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vrev16q_u8 (uint8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vrev16v16qi ((int8x16_t) __a, 0);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vrev16q_p8 (poly8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vrev16v16qi ((int8x16_t) __a, 2);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vbsl_s8 (uint8x8_t __a, int8x8_t __b, int8x8_t __c)
{
  return (int8x8_t)__builtin_neon_vbslv8qi ((int8x8_t) __a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vbsl_s16 (uint16x4_t __a, int16x4_t __b, int16x4_t __c)
{
  return (int16x4_t)__builtin_neon_vbslv4hi ((int16x4_t) __a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vbsl_s32 (uint32x2_t __a, int32x2_t __b, int32x2_t __c)
{
  return (int32x2_t)__builtin_neon_vbslv2si ((int32x2_t) __a, __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vbsl_s64 (uint64x1_t __a, int64x1_t __b, int64x1_t __c)
{
  return (int64x1_t)__builtin_neon_vbsldi ((int64x1_t) __a, __b, __c);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vbsl_f32 (uint32x2_t __a, float32x2_t __b, float32x2_t __c)
{
  return (float32x2_t)__builtin_neon_vbslv2sf ((int32x2_t) __a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vbsl_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
{
  return (uint8x8_t)__builtin_neon_vbslv8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vbsl_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
{
  return (uint16x4_t)__builtin_neon_vbslv4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vbsl_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
{
  return (uint32x2_t)__builtin_neon_vbslv2si ((int32x2_t) __a, (int32x2_t) __b, (int32x2_t) __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vbsl_u64 (uint64x1_t __a, uint64x1_t __b, uint64x1_t __c)
{
  return (uint64x1_t)__builtin_neon_vbsldi ((int64x1_t) __a, (int64x1_t) __b, (int64x1_t) __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vbsl_p8 (uint8x8_t __a, poly8x8_t __b, poly8x8_t __c)
{
  return (poly8x8_t)__builtin_neon_vbslv8qi ((int8x8_t) __a, (int8x8_t) __b, (int8x8_t) __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vbsl_p16 (uint16x4_t __a, poly16x4_t __b, poly16x4_t __c)
{
  return (poly16x4_t)__builtin_neon_vbslv4hi ((int16x4_t) __a, (int16x4_t) __b, (int16x4_t) __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vbslq_s8 (uint8x16_t __a, int8x16_t __b, int8x16_t __c)
{
  return (int8x16_t)__builtin_neon_vbslv16qi ((int8x16_t) __a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vbslq_s16 (uint16x8_t __a, int16x8_t __b, int16x8_t __c)
{
  return (int16x8_t)__builtin_neon_vbslv8hi ((int16x8_t) __a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vbslq_s32 (uint32x4_t __a, int32x4_t __b, int32x4_t __c)
{
  return (int32x4_t)__builtin_neon_vbslv4si ((int32x4_t) __a, __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vbslq_s64 (uint64x2_t __a, int64x2_t __b, int64x2_t __c)
{
  return (int64x2_t)__builtin_neon_vbslv2di ((int64x2_t) __a, __b, __c);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vbslq_f32 (uint32x4_t __a, float32x4_t __b, float32x4_t __c)
{
  return (float32x4_t)__builtin_neon_vbslv4sf ((int32x4_t) __a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vbslq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
{
  return (uint8x16_t)__builtin_neon_vbslv16qi ((int8x16_t) __a, (int8x16_t) __b, (int8x16_t) __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vbslq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
{
  return (uint16x8_t)__builtin_neon_vbslv8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x8_t) __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vbslq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
{
  return (uint32x4_t)__builtin_neon_vbslv4si ((int32x4_t) __a, (int32x4_t) __b, (int32x4_t) __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vbslq_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
{
  return (uint64x2_t)__builtin_neon_vbslv2di ((int64x2_t) __a, (int64x2_t) __b, (int64x2_t) __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vbslq_p8 (uint8x16_t __a, poly8x16_t __b, poly8x16_t __c)
{
  return (poly8x16_t)__builtin_neon_vbslv16qi ((int8x16_t) __a, (int8x16_t) __b, (int8x16_t) __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vbslq_p16 (uint16x8_t __a, poly16x8_t __b, poly16x8_t __c)
{
  return (poly16x8_t)__builtin_neon_vbslv8hi ((int16x8_t) __a, (int16x8_t) __b, (int16x8_t) __c);
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vtrn_s8 (int8x8_t __a, int8x8_t __b)
{
  int8x8x2_t __rv;
  __builtin_neon_vtrnv8qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vtrn_s16 (int16x4_t __a, int16x4_t __b)
{
  int16x4x2_t __rv;
  __builtin_neon_vtrnv4hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vtrn_s32 (int32x2_t __a, int32x2_t __b)
{
  int32x2x2_t __rv;
  __builtin_neon_vtrnv2si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vtrn_f32 (float32x2_t __a, float32x2_t __b)
{
  float32x2x2_t __rv;
  __builtin_neon_vtrnv2sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vtrn_u8 (uint8x8_t __a, uint8x8_t __b)
{
  uint8x8x2_t __rv;
  __builtin_neon_vtrnv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vtrn_u16 (uint16x4_t __a, uint16x4_t __b)
{
  uint16x4x2_t __rv;
  __builtin_neon_vtrnv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vtrn_u32 (uint32x2_t __a, uint32x2_t __b)
{
  uint32x2x2_t __rv;
  __builtin_neon_vtrnv2si ((int32x2_t *) &__rv.val[0], (int32x2_t) __a, (int32x2_t) __b);
  return __rv;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vtrn_p8 (poly8x8_t __a, poly8x8_t __b)
{
  poly8x8x2_t __rv;
  __builtin_neon_vtrnv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vtrn_p16 (poly16x4_t __a, poly16x4_t __b)
{
  poly16x4x2_t __rv;
  __builtin_neon_vtrnv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
vtrnq_s8 (int8x16_t __a, int8x16_t __b)
{
  int8x16x2_t __rv;
  __builtin_neon_vtrnv16qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
vtrnq_s16 (int16x8_t __a, int16x8_t __b)
{
  int16x8x2_t __rv;
  __builtin_neon_vtrnv8hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
vtrnq_s32 (int32x4_t __a, int32x4_t __b)
{
  int32x4x2_t __rv;
  __builtin_neon_vtrnv4si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
vtrnq_f32 (float32x4_t __a, float32x4_t __b)
{
  float32x4x2_t __rv;
  __builtin_neon_vtrnv4sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
vtrnq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  uint8x16x2_t __rv;
  __builtin_neon_vtrnv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
vtrnq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  uint16x8x2_t __rv;
  __builtin_neon_vtrnv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
vtrnq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  uint32x4x2_t __rv;
  __builtin_neon_vtrnv4si ((int32x4_t *) &__rv.val[0], (int32x4_t) __a, (int32x4_t) __b);
  return __rv;
}

__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
vtrnq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  poly8x16x2_t __rv;
  __builtin_neon_vtrnv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
vtrnq_p16 (poly16x8_t __a, poly16x8_t __b)
{
  poly16x8x2_t __rv;
  __builtin_neon_vtrnv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vzip_s8 (int8x8_t __a, int8x8_t __b)
{
  int8x8x2_t __rv;
  __builtin_neon_vzipv8qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vzip_s16 (int16x4_t __a, int16x4_t __b)
{
  int16x4x2_t __rv;
  __builtin_neon_vzipv4hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vzip_s32 (int32x2_t __a, int32x2_t __b)
{
  int32x2x2_t __rv;
  __builtin_neon_vzipv2si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vzip_f32 (float32x2_t __a, float32x2_t __b)
{
  float32x2x2_t __rv;
  __builtin_neon_vzipv2sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vzip_u8 (uint8x8_t __a, uint8x8_t __b)
{
  uint8x8x2_t __rv;
  __builtin_neon_vzipv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vzip_u16 (uint16x4_t __a, uint16x4_t __b)
{
  uint16x4x2_t __rv;
  __builtin_neon_vzipv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vzip_u32 (uint32x2_t __a, uint32x2_t __b)
{
  uint32x2x2_t __rv;
  __builtin_neon_vzipv2si ((int32x2_t *) &__rv.val[0], (int32x2_t) __a, (int32x2_t) __b);
  return __rv;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vzip_p8 (poly8x8_t __a, poly8x8_t __b)
{
  poly8x8x2_t __rv;
  __builtin_neon_vzipv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vzip_p16 (poly16x4_t __a, poly16x4_t __b)
{
  poly16x4x2_t __rv;
  __builtin_neon_vzipv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
vzipq_s8 (int8x16_t __a, int8x16_t __b)
{
  int8x16x2_t __rv;
  __builtin_neon_vzipv16qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
vzipq_s16 (int16x8_t __a, int16x8_t __b)
{
  int16x8x2_t __rv;
  __builtin_neon_vzipv8hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
vzipq_s32 (int32x4_t __a, int32x4_t __b)
{
  int32x4x2_t __rv;
  __builtin_neon_vzipv4si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
vzipq_f32 (float32x4_t __a, float32x4_t __b)
{
  float32x4x2_t __rv;
  __builtin_neon_vzipv4sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
vzipq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  uint8x16x2_t __rv;
  __builtin_neon_vzipv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
vzipq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  uint16x8x2_t __rv;
  __builtin_neon_vzipv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
vzipq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  uint32x4x2_t __rv;
  __builtin_neon_vzipv4si ((int32x4_t *) &__rv.val[0], (int32x4_t) __a, (int32x4_t) __b);
  return __rv;
}

__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
vzipq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  poly8x16x2_t __rv;
  __builtin_neon_vzipv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
vzipq_p16 (poly16x8_t __a, poly16x8_t __b)
{
  poly16x8x2_t __rv;
  __builtin_neon_vzipv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vuzp_s8 (int8x8_t __a, int8x8_t __b)
{
  int8x8x2_t __rv;
  __builtin_neon_vuzpv8qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vuzp_s16 (int16x4_t __a, int16x4_t __b)
{
  int16x4x2_t __rv;
  __builtin_neon_vuzpv4hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vuzp_s32 (int32x2_t __a, int32x2_t __b)
{
  int32x2x2_t __rv;
  __builtin_neon_vuzpv2si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vuzp_f32 (float32x2_t __a, float32x2_t __b)
{
  float32x2x2_t __rv;
  __builtin_neon_vuzpv2sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vuzp_u8 (uint8x8_t __a, uint8x8_t __b)
{
  uint8x8x2_t __rv;
  __builtin_neon_vuzpv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vuzp_u16 (uint16x4_t __a, uint16x4_t __b)
{
  uint16x4x2_t __rv;
  __builtin_neon_vuzpv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vuzp_u32 (uint32x2_t __a, uint32x2_t __b)
{
  uint32x2x2_t __rv;
  __builtin_neon_vuzpv2si ((int32x2_t *) &__rv.val[0], (int32x2_t) __a, (int32x2_t) __b);
  return __rv;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vuzp_p8 (poly8x8_t __a, poly8x8_t __b)
{
  poly8x8x2_t __rv;
  __builtin_neon_vuzpv8qi ((int8x8_t *) &__rv.val[0], (int8x8_t) __a, (int8x8_t) __b);
  return __rv;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vuzp_p16 (poly16x4_t __a, poly16x4_t __b)
{
  poly16x4x2_t __rv;
  __builtin_neon_vuzpv4hi ((int16x4_t *) &__rv.val[0], (int16x4_t) __a, (int16x4_t) __b);
  return __rv;
}

__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
vuzpq_s8 (int8x16_t __a, int8x16_t __b)
{
  int8x16x2_t __rv;
  __builtin_neon_vuzpv16qi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
vuzpq_s16 (int16x8_t __a, int16x8_t __b)
{
  int16x8x2_t __rv;
  __builtin_neon_vuzpv8hi (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
vuzpq_s32 (int32x4_t __a, int32x4_t __b)
{
  int32x4x2_t __rv;
  __builtin_neon_vuzpv4si (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
vuzpq_f32 (float32x4_t __a, float32x4_t __b)
{
  float32x4x2_t __rv;
  __builtin_neon_vuzpv4sf (&__rv.val[0], __a, __b);
  return __rv;
}

__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
vuzpq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  uint8x16x2_t __rv;
  __builtin_neon_vuzpv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
vuzpq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  uint16x8x2_t __rv;
  __builtin_neon_vuzpv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
vuzpq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  uint32x4x2_t __rv;
  __builtin_neon_vuzpv4si ((int32x4_t *) &__rv.val[0], (int32x4_t) __a, (int32x4_t) __b);
  return __rv;
}

__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
vuzpq_p8 (poly8x16_t __a, poly8x16_t __b)
{
  poly8x16x2_t __rv;
  __builtin_neon_vuzpv16qi ((int8x16_t *) &__rv.val[0], (int8x16_t) __a, (int8x16_t) __b);
  return __rv;
}

__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
vuzpq_p16 (poly16x8_t __a, poly16x8_t __b)
{
  poly16x8x2_t __rv;
  __builtin_neon_vuzpv8hi ((int16x8_t *) &__rv.val[0], (int16x8_t) __a, (int16x8_t) __b);
  return __rv;
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vld1_s8 (const int8_t * __a)
{
  return (int8x8_t)__builtin_neon_vld1v8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vld1_s16 (const int16_t * __a)
{
  return (int16x4_t)__builtin_neon_vld1v4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vld1_s32 (const int32_t * __a)
{
  return (int32x2_t)__builtin_neon_vld1v2si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vld1_s64 (const int64_t * __a)
{
  return (int64x1_t)__builtin_neon_vld1di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vld1_f32 (const float32_t * __a)
{
  return (float32x2_t)__builtin_neon_vld1v2sf ((const __builtin_neon_sf *) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vld1_u8 (const uint8_t * __a)
{
  return (uint8x8_t)__builtin_neon_vld1v8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vld1_u16 (const uint16_t * __a)
{
  return (uint16x4_t)__builtin_neon_vld1v4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vld1_u32 (const uint32_t * __a)
{
  return (uint32x2_t)__builtin_neon_vld1v2si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vld1_u64 (const uint64_t * __a)
{
  return (uint64x1_t)__builtin_neon_vld1di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vld1_p8 (const poly8_t * __a)
{
  return (poly8x8_t)__builtin_neon_vld1v8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vld1_p16 (const poly16_t * __a)
{
  return (poly16x4_t)__builtin_neon_vld1v4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vld1q_s8 (const int8_t * __a)
{
  return (int8x16_t)__builtin_neon_vld1v16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vld1q_s16 (const int16_t * __a)
{
  return (int16x8_t)__builtin_neon_vld1v8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vld1q_s32 (const int32_t * __a)
{
  return (int32x4_t)__builtin_neon_vld1v4si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vld1q_s64 (const int64_t * __a)
{
  return (int64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vld1q_f32 (const float32_t * __a)
{
  return (float32x4_t)__builtin_neon_vld1v4sf ((const __builtin_neon_sf *) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vld1q_u8 (const uint8_t * __a)
{
  return (uint8x16_t)__builtin_neon_vld1v16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vld1q_u16 (const uint16_t * __a)
{
  return (uint16x8_t)__builtin_neon_vld1v8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vld1q_u32 (const uint32_t * __a)
{
  return (uint32x4_t)__builtin_neon_vld1v4si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vld1q_u64 (const uint64_t * __a)
{
  return (uint64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vld1q_p8 (const poly8_t * __a)
{
  return (poly8x16_t)__builtin_neon_vld1v16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vld1q_p16 (const poly16_t * __a)
{
  return (poly16x8_t)__builtin_neon_vld1v8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
{
  return (int8x8_t)__builtin_neon_vld1_lanev8qi ((const __builtin_neon_qi *) __a, __b, __c);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vld1_lane_s16 (const int16_t * __a, int16x4_t __b, const int __c)
{
  return (int16x4_t)__builtin_neon_vld1_lanev4hi ((const __builtin_neon_hi *) __a, __b, __c);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vld1_lane_s32 (const int32_t * __a, int32x2_t __b, const int __c)
{
  return (int32x2_t)__builtin_neon_vld1_lanev2si ((const __builtin_neon_si *) __a, __b, __c);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vld1_lane_f32 (const float32_t * __a, float32x2_t __b, const int __c)
{
  return (float32x2_t)__builtin_neon_vld1_lanev2sf ((const __builtin_neon_sf *) __a, __b, __c);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vld1_lane_u8 (const uint8_t * __a, uint8x8_t __b, const int __c)
{
  return (uint8x8_t)__builtin_neon_vld1_lanev8qi ((const __builtin_neon_qi *) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vld1_lane_u16 (const uint16_t * __a, uint16x4_t __b, const int __c)
{
  return (uint16x4_t)__builtin_neon_vld1_lanev4hi ((const __builtin_neon_hi *) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vld1_lane_u32 (const uint32_t * __a, uint32x2_t __b, const int __c)
{
  return (uint32x2_t)__builtin_neon_vld1_lanev2si ((const __builtin_neon_si *) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vld1_lane_p8 (const poly8_t * __a, poly8x8_t __b, const int __c)
{
  return (poly8x8_t)__builtin_neon_vld1_lanev8qi ((const __builtin_neon_qi *) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vld1_lane_p16 (const poly16_t * __a, poly16x4_t __b, const int __c)
{
  return (poly16x4_t)__builtin_neon_vld1_lanev4hi ((const __builtin_neon_hi *) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vld1_lane_s64 (const int64_t * __a, int64x1_t __b, const int __c)
{
  return (int64x1_t)__builtin_neon_vld1_lanedi ((const __builtin_neon_di *) __a, __b, __c);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vld1_lane_u64 (const uint64_t * __a, uint64x1_t __b, const int __c)
{
  return (uint64x1_t)__builtin_neon_vld1_lanedi ((const __builtin_neon_di *) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vld1q_lane_s8 (const int8_t * __a, int8x16_t __b, const int __c)
{
  return (int8x16_t)__builtin_neon_vld1_lanev16qi ((const __builtin_neon_qi *) __a, __b, __c);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vld1q_lane_s16 (const int16_t * __a, int16x8_t __b, const int __c)
{
  return (int16x8_t)__builtin_neon_vld1_lanev8hi ((const __builtin_neon_hi *) __a, __b, __c);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vld1q_lane_s32 (const int32_t * __a, int32x4_t __b, const int __c)
{
  return (int32x4_t)__builtin_neon_vld1_lanev4si ((const __builtin_neon_si *) __a, __b, __c);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vld1q_lane_f32 (const float32_t * __a, float32x4_t __b, const int __c)
{
  return (float32x4_t)__builtin_neon_vld1_lanev4sf ((const __builtin_neon_sf *) __a, __b, __c);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vld1q_lane_u8 (const uint8_t * __a, uint8x16_t __b, const int __c)
{
  return (uint8x16_t)__builtin_neon_vld1_lanev16qi ((const __builtin_neon_qi *) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vld1q_lane_u16 (const uint16_t * __a, uint16x8_t __b, const int __c)
{
  return (uint16x8_t)__builtin_neon_vld1_lanev8hi ((const __builtin_neon_hi *) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vld1q_lane_u32 (const uint32_t * __a, uint32x4_t __b, const int __c)
{
  return (uint32x4_t)__builtin_neon_vld1_lanev4si ((const __builtin_neon_si *) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vld1q_lane_p8 (const poly8_t * __a, poly8x16_t __b, const int __c)
{
  return (poly8x16_t)__builtin_neon_vld1_lanev16qi ((const __builtin_neon_qi *) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vld1q_lane_p16 (const poly16_t * __a, poly16x8_t __b, const int __c)
{
  return (poly16x8_t)__builtin_neon_vld1_lanev8hi ((const __builtin_neon_hi *) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vld1q_lane_s64 (const int64_t * __a, int64x2_t __b, const int __c)
{
  return (int64x2_t)__builtin_neon_vld1_lanev2di ((const __builtin_neon_di *) __a, __b, __c);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vld1q_lane_u64 (const uint64_t * __a, uint64x2_t __b, const int __c)
{
  return (uint64x2_t)__builtin_neon_vld1_lanev2di ((const __builtin_neon_di *) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vld1_dup_s8 (const int8_t * __a)
{
  return (int8x8_t)__builtin_neon_vld1_dupv8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vld1_dup_s16 (const int16_t * __a)
{
  return (int16x4_t)__builtin_neon_vld1_dupv4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vld1_dup_s32 (const int32_t * __a)
{
  return (int32x2_t)__builtin_neon_vld1_dupv2si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vld1_dup_f32 (const float32_t * __a)
{
  return (float32x2_t)__builtin_neon_vld1_dupv2sf ((const __builtin_neon_sf *) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vld1_dup_u8 (const uint8_t * __a)
{
  return (uint8x8_t)__builtin_neon_vld1_dupv8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vld1_dup_u16 (const uint16_t * __a)
{
  return (uint16x4_t)__builtin_neon_vld1_dupv4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vld1_dup_u32 (const uint32_t * __a)
{
  return (uint32x2_t)__builtin_neon_vld1_dupv2si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vld1_dup_p8 (const poly8_t * __a)
{
  return (poly8x8_t)__builtin_neon_vld1_dupv8qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vld1_dup_p16 (const poly16_t * __a)
{
  return (poly16x4_t)__builtin_neon_vld1_dupv4hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vld1_dup_s64 (const int64_t * __a)
{
  return (int64x1_t)__builtin_neon_vld1_dupdi ((const __builtin_neon_di *) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vld1_dup_u64 (const uint64_t * __a)
{
  return (uint64x1_t)__builtin_neon_vld1_dupdi ((const __builtin_neon_di *) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vld1q_dup_s8 (const int8_t * __a)
{
  return (int8x16_t)__builtin_neon_vld1_dupv16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vld1q_dup_s16 (const int16_t * __a)
{
  return (int16x8_t)__builtin_neon_vld1_dupv8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vld1q_dup_s32 (const int32_t * __a)
{
  return (int32x4_t)__builtin_neon_vld1_dupv4si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vld1q_dup_f32 (const float32_t * __a)
{
  return (float32x4_t)__builtin_neon_vld1_dupv4sf ((const __builtin_neon_sf *) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vld1q_dup_u8 (const uint8_t * __a)
{
  return (uint8x16_t)__builtin_neon_vld1_dupv16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vld1q_dup_u16 (const uint16_t * __a)
{
  return (uint16x8_t)__builtin_neon_vld1_dupv8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vld1q_dup_u32 (const uint32_t * __a)
{
  return (uint32x4_t)__builtin_neon_vld1_dupv4si ((const __builtin_neon_si *) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vld1q_dup_p8 (const poly8_t * __a)
{
  return (poly8x16_t)__builtin_neon_vld1_dupv16qi ((const __builtin_neon_qi *) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vld1q_dup_p16 (const poly16_t * __a)
{
  return (poly16x8_t)__builtin_neon_vld1_dupv8hi ((const __builtin_neon_hi *) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vld1q_dup_s64 (const int64_t * __a)
{
  return (int64x2_t)__builtin_neon_vld1_dupv2di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vld1q_dup_u64 (const uint64_t * __a)
{
  return (uint64x2_t)__builtin_neon_vld1_dupv2di ((const __builtin_neon_di *) __a);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_s8 (int8_t * __a, int8x8_t __b)
{
  __builtin_neon_vst1v8qi ((__builtin_neon_qi *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_s16 (int16_t * __a, int16x4_t __b)
{
  __builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_s32 (int32_t * __a, int32x2_t __b)
{
  __builtin_neon_vst1v2si ((__builtin_neon_si *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_s64 (int64_t * __a, int64x1_t __b)
{
  __builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_f32 (float32_t * __a, float32x2_t __b)
{
  __builtin_neon_vst1v2sf ((__builtin_neon_sf *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_u8 (uint8_t * __a, uint8x8_t __b)
{
  __builtin_neon_vst1v8qi ((__builtin_neon_qi *) __a, (int8x8_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_u16 (uint16_t * __a, uint16x4_t __b)
{
  __builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_u32 (uint32_t * __a, uint32x2_t __b)
{
  __builtin_neon_vst1v2si ((__builtin_neon_si *) __a, (int32x2_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_u64 (uint64_t * __a, uint64x1_t __b)
{
  __builtin_neon_vst1di ((__builtin_neon_di *) __a, (int64x1_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_p8 (poly8_t * __a, poly8x8_t __b)
{
  __builtin_neon_vst1v8qi ((__builtin_neon_qi *) __a, (int8x8_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_p16 (poly16_t * __a, poly16x4_t __b)
{
  __builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_s8 (int8_t * __a, int8x16_t __b)
{
  __builtin_neon_vst1v16qi ((__builtin_neon_qi *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_s16 (int16_t * __a, int16x8_t __b)
{
  __builtin_neon_vst1v8hi ((__builtin_neon_hi *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_s32 (int32_t * __a, int32x4_t __b)
{
  __builtin_neon_vst1v4si ((__builtin_neon_si *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_s64 (int64_t * __a, int64x2_t __b)
{
  __builtin_neon_vst1v2di ((__builtin_neon_di *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_f32 (float32_t * __a, float32x4_t __b)
{
  __builtin_neon_vst1v4sf ((__builtin_neon_sf *) __a, __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_u8 (uint8_t * __a, uint8x16_t __b)
{
  __builtin_neon_vst1v16qi ((__builtin_neon_qi *) __a, (int8x16_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_u16 (uint16_t * __a, uint16x8_t __b)
{
  __builtin_neon_vst1v8hi ((__builtin_neon_hi *) __a, (int16x8_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_u32 (uint32_t * __a, uint32x4_t __b)
{
  __builtin_neon_vst1v4si ((__builtin_neon_si *) __a, (int32x4_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_u64 (uint64_t * __a, uint64x2_t __b)
{
  __builtin_neon_vst1v2di ((__builtin_neon_di *) __a, (int64x2_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_p8 (poly8_t * __a, poly8x16_t __b)
{
  __builtin_neon_vst1v16qi ((__builtin_neon_qi *) __a, (int8x16_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_p16 (poly16_t * __a, poly16x8_t __b)
{
  __builtin_neon_vst1v8hi ((__builtin_neon_hi *) __a, (int16x8_t) __b);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_s8 (int8_t * __a, int8x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8qi ((__builtin_neon_qi *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_s16 (int16_t * __a, int16x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4hi ((__builtin_neon_hi *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_s32 (int32_t * __a, int32x2_t __b, const int __c)
{
  __builtin_neon_vst1_lanev2si ((__builtin_neon_si *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_f32 (float32_t * __a, float32x2_t __b, const int __c)
{
  __builtin_neon_vst1_lanev2sf ((__builtin_neon_sf *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_u8 (uint8_t * __a, uint8x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8qi ((__builtin_neon_qi *) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_u16 (uint16_t * __a, uint16x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_u32 (uint32_t * __a, uint32x2_t __b, const int __c)
{
  __builtin_neon_vst1_lanev2si ((__builtin_neon_si *) __a, (int32x2_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_p8 (poly8_t * __a, poly8x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8qi ((__builtin_neon_qi *) __a, (int8x8_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_p16 (poly16_t * __a, poly16x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_s64 (int64_t * __a, int64x1_t __b, const int __c)
{
  __builtin_neon_vst1_lanedi ((__builtin_neon_di *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1_lane_u64 (uint64_t * __a, uint64x1_t __b, const int __c)
{
  __builtin_neon_vst1_lanedi ((__builtin_neon_di *) __a, (int64x1_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_s8 (int8_t * __a, int8x16_t __b, const int __c)
{
  __builtin_neon_vst1_lanev16qi ((__builtin_neon_qi *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_s16 (int16_t * __a, int16x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8hi ((__builtin_neon_hi *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_s32 (int32_t * __a, int32x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4si ((__builtin_neon_si *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_f32 (float32_t * __a, float32x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4sf ((__builtin_neon_sf *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_u8 (uint8_t * __a, uint8x16_t __b, const int __c)
{
  __builtin_neon_vst1_lanev16qi ((__builtin_neon_qi *) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_u16 (uint16_t * __a, uint16x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8hi ((__builtin_neon_hi *) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_u32 (uint32_t * __a, uint32x4_t __b, const int __c)
{
  __builtin_neon_vst1_lanev4si ((__builtin_neon_si *) __a, (int32x4_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_p8 (poly8_t * __a, poly8x16_t __b, const int __c)
{
  __builtin_neon_vst1_lanev16qi ((__builtin_neon_qi *) __a, (int8x16_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_p16 (poly16_t * __a, poly16x8_t __b, const int __c)
{
  __builtin_neon_vst1_lanev8hi ((__builtin_neon_hi *) __a, (int16x8_t) __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_s64 (int64_t * __a, int64x2_t __b, const int __c)
{
  __builtin_neon_vst1_lanev2di ((__builtin_neon_di *) __a, __b, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst1q_lane_u64 (uint64_t * __a, uint64x2_t __b, const int __c)
{
  __builtin_neon_vst1_lanev2di ((__builtin_neon_di *) __a, (int64x2_t) __b, __c);
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vld2_s8 (const int8_t * __a)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vld2_s16 (const int16_t * __a)
{
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vld2_s32 (const int32_t * __a)
{
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vld2_f32 (const float32_t * __a)
{
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vld2_u8 (const uint8_t * __a)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vld2_u16 (const uint16_t * __a)
{
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vld2_u32 (const uint32_t * __a)
{
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vld2_p8 (const poly8_t * __a)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vld2_p16 (const poly16_t * __a)
{
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x2_t __attribute__ ((__always_inline__))
vld2_s64 (const int64_t * __a)
{
  union { int64x1x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x2_t __attribute__ ((__always_inline__))
vld2_u64 (const uint64_t * __a)
{
  union { uint64x1x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
vld2q_s8 (const int8_t * __a)
{
  union { int8x16x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
vld2q_s16 (const int16_t * __a)
{
  union { int16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
vld2q_s32 (const int32_t * __a)
{
  union { int32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
vld2q_f32 (const float32_t * __a)
{
  union { float32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
vld2q_u8 (const uint8_t * __a)
{
  union { uint8x16x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
vld2q_u16 (const uint16_t * __a)
{
  union { uint16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
vld2q_u32 (const uint32_t * __a)
{
  union { uint32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
vld2q_p8 (const poly8_t * __a)
{
  union { poly8x16x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
vld2q_p16 (const poly16_t * __a)
{
  union { poly16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vld2_lane_s8 (const int8_t * __a, int8x8x2_t __b, const int __c)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vld2_lane_s16 (const int16_t * __a, int16x4x2_t __b, const int __c)
{
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vld2_lane_s32 (const int32_t * __a, int32x2x2_t __b, const int __c)
{
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vld2_lane_f32 (const float32_t * __a, float32x2x2_t __b, const int __c)
{
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev2sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vld2_lane_u8 (const uint8_t * __a, uint8x8x2_t __b, const int __c)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vld2_lane_u16 (const uint16_t * __a, uint16x4x2_t __b, const int __c)
{
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vld2_lane_u32 (const uint32_t * __a, uint32x2x2_t __b, const int __c)
{
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vld2_lane_p8 (const poly8_t * __a, poly8x8x2_t __b, const int __c)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vld2_lane_p16 (const poly16_t * __a, poly16x4x2_t __b, const int __c)
{
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
vld2q_lane_s16 (const int16_t * __a, int16x8x2_t __b, const int __c)
{
  union { int16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { int16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
vld2q_lane_s32 (const int32_t * __a, int32x4x2_t __b, const int __c)
{
  union { int32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { int32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
vld2q_lane_f32 (const float32_t * __a, float32x4x2_t __b, const int __c)
{
  union { float32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { float32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
vld2q_lane_u16 (const uint16_t * __a, uint16x8x2_t __b, const int __c)
{
  union { uint16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { uint16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
vld2q_lane_u32 (const uint32_t * __a, uint32x4x2_t __b, const int __c)
{
  union { uint32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { uint32x4x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
vld2q_lane_p16 (const poly16_t * __a, poly16x8x2_t __b, const int __c)
{
  union { poly16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { poly16x8x2_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld2_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
vld2_dup_s8 (const int8_t * __a)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
vld2_dup_s16 (const int16_t * __a)
{
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
vld2_dup_s32 (const int32_t * __a)
{
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
vld2_dup_f32 (const float32_t * __a)
{
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
vld2_dup_u8 (const uint8_t * __a)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
vld2_dup_u16 (const uint16_t * __a)
{
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
vld2_dup_u32 (const uint32_t * __a)
{
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
vld2_dup_p8 (const poly8_t * __a)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
vld2_dup_p16 (const poly16_t * __a)
{
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x2_t __attribute__ ((__always_inline__))
vld2_dup_s64 (const int64_t * __a)
{
  union { int64x1x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x2_t __attribute__ ((__always_inline__))
vld2_dup_u64 (const uint64_t * __a)
{
  union { uint64x1x2_t __i; __builtin_neon_ti __o; } __rv;
  __rv.__o = __builtin_neon_vld2_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_s8 (int8_t * __a, int8x8x2_t __b)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_s16 (int16_t * __a, int16x4x2_t __b)
{
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_s32 (int32_t * __a, int32x2x2_t __b)
{
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_f32 (float32_t * __a, float32x2x2_t __b)
{
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_u8 (uint8_t * __a, uint8x8x2_t __b)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_u16 (uint16_t * __a, uint16x4x2_t __b)
{
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_u32 (uint32_t * __a, uint32x2x2_t __b)
{
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_p8 (poly8_t * __a, poly8x8x2_t __b)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_p16 (poly16_t * __a, poly16x4x2_t __b)
{
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_s64 (int64_t * __a, int64x1x2_t __b)
{
  union { int64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_u64 (uint64_t * __a, uint64x1x2_t __b)
{
  union { uint64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_s8 (int8_t * __a, int8x16x2_t __b)
{
  union { int8x16x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_s16 (int16_t * __a, int16x8x2_t __b)
{
  union { int16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_s32 (int32_t * __a, int32x4x2_t __b)
{
  union { int32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_f32 (float32_t * __a, float32x4x2_t __b)
{
  union { float32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v4sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_u8 (uint8_t * __a, uint8x16x2_t __b)
{
  union { uint8x16x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_u16 (uint16_t * __a, uint16x8x2_t __b)
{
  union { uint16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_u32 (uint32_t * __a, uint32x4x2_t __b)
{
  union { uint32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_p8 (poly8_t * __a, poly8x16x2_t __b)
{
  union { poly8x16x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_p16 (poly16_t * __a, poly16x8x2_t __b)
{
  union { poly16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_s8 (int8_t * __a, int8x8x2_t __b, const int __c)
{
  union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_s16 (int16_t * __a, int16x4x2_t __b, const int __c)
{
  union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_s32 (int32_t * __a, int32x2x2_t __b, const int __c)
{
  union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_f32 (float32_t * __a, float32x2x2_t __b, const int __c)
{
  union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev2sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_u8 (uint8_t * __a, uint8x8x2_t __b, const int __c)
{
  union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_u16 (uint16_t * __a, uint16x4x2_t __b, const int __c)
{
  union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_u32 (uint32_t * __a, uint32x2x2_t __b, const int __c)
{
  union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_p8 (poly8_t * __a, poly8x8x2_t __b, const int __c)
{
  union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2_lane_p16 (poly16_t * __a, poly16x4x2_t __b, const int __c)
{
  union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_s16 (int16_t * __a, int16x8x2_t __b, const int __c)
{
  union { int16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_s32 (int32_t * __a, int32x4x2_t __b, const int __c)
{
  union { int32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_f32 (float32_t * __a, float32x4x2_t __b, const int __c)
{
  union { float32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_u16 (uint16_t * __a, uint16x8x2_t __b, const int __c)
{
  union { uint16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_u32 (uint32_t * __a, uint32x4x2_t __b, const int __c)
{
  union { uint32x4x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst2q_lane_p16 (poly16_t * __a, poly16x8x2_t __b, const int __c)
{
  union { poly16x8x2_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst2_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline int8x8x3_t __attribute__ ((__always_inline__))
vld3_s8 (const int8_t * __a)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x3_t __attribute__ ((__always_inline__))
vld3_s16 (const int16_t * __a)
{
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x3_t __attribute__ ((__always_inline__))
vld3_s32 (const int32_t * __a)
{
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x3_t __attribute__ ((__always_inline__))
vld3_f32 (const float32_t * __a)
{
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x3_t __attribute__ ((__always_inline__))
vld3_u8 (const uint8_t * __a)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x3_t __attribute__ ((__always_inline__))
vld3_u16 (const uint16_t * __a)
{
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x3_t __attribute__ ((__always_inline__))
vld3_u32 (const uint32_t * __a)
{
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x3_t __attribute__ ((__always_inline__))
vld3_p8 (const poly8_t * __a)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x3_t __attribute__ ((__always_inline__))
vld3_p16 (const poly16_t * __a)
{
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x3_t __attribute__ ((__always_inline__))
vld3_s64 (const int64_t * __a)
{
  union { int64x1x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x3_t __attribute__ ((__always_inline__))
vld3_u64 (const uint64_t * __a)
{
  union { uint64x1x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x16x3_t __attribute__ ((__always_inline__))
vld3q_s8 (const int8_t * __a)
{
  union { int8x16x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x8x3_t __attribute__ ((__always_inline__))
vld3q_s16 (const int16_t * __a)
{
  union { int16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x4x3_t __attribute__ ((__always_inline__))
vld3q_s32 (const int32_t * __a)
{
  union { int32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x4x3_t __attribute__ ((__always_inline__))
vld3q_f32 (const float32_t * __a)
{
  union { float32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x16x3_t __attribute__ ((__always_inline__))
vld3q_u8 (const uint8_t * __a)
{
  union { uint8x16x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x8x3_t __attribute__ ((__always_inline__))
vld3q_u16 (const uint16_t * __a)
{
  union { uint16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x4x3_t __attribute__ ((__always_inline__))
vld3q_u32 (const uint32_t * __a)
{
  union { uint32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x16x3_t __attribute__ ((__always_inline__))
vld3q_p8 (const poly8_t * __a)
{
  union { poly8x16x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x8x3_t __attribute__ ((__always_inline__))
vld3q_p16 (const poly16_t * __a)
{
  union { poly16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x8x3_t __attribute__ ((__always_inline__))
vld3_lane_s8 (const int8_t * __a, int8x8x3_t __b, const int __c)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x4x3_t __attribute__ ((__always_inline__))
vld3_lane_s16 (const int16_t * __a, int16x4x3_t __b, const int __c)
{
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x2x3_t __attribute__ ((__always_inline__))
vld3_lane_s32 (const int32_t * __a, int32x2x3_t __b, const int __c)
{
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x2x3_t __attribute__ ((__always_inline__))
vld3_lane_f32 (const float32_t * __a, float32x2x3_t __b, const int __c)
{
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev2sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint8x8x3_t __attribute__ ((__always_inline__))
vld3_lane_u8 (const uint8_t * __a, uint8x8x3_t __b, const int __c)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x4x3_t __attribute__ ((__always_inline__))
vld3_lane_u16 (const uint16_t * __a, uint16x4x3_t __b, const int __c)
{
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x2x3_t __attribute__ ((__always_inline__))
vld3_lane_u32 (const uint32_t * __a, uint32x2x3_t __b, const int __c)
{
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly8x8x3_t __attribute__ ((__always_inline__))
vld3_lane_p8 (const poly8_t * __a, poly8x8x3_t __b, const int __c)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x4x3_t __attribute__ ((__always_inline__))
vld3_lane_p16 (const poly16_t * __a, poly16x4x3_t __b, const int __c)
{
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x8x3_t __attribute__ ((__always_inline__))
vld3q_lane_s16 (const int16_t * __a, int16x8x3_t __b, const int __c)
{
  union { int16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { int16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x4x3_t __attribute__ ((__always_inline__))
vld3q_lane_s32 (const int32_t * __a, int32x4x3_t __b, const int __c)
{
  union { int32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { int32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x4x3_t __attribute__ ((__always_inline__))
vld3q_lane_f32 (const float32_t * __a, float32x4x3_t __b, const int __c)
{
  union { float32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { float32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x8x3_t __attribute__ ((__always_inline__))
vld3q_lane_u16 (const uint16_t * __a, uint16x8x3_t __b, const int __c)
{
  union { uint16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { uint16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x4x3_t __attribute__ ((__always_inline__))
vld3q_lane_u32 (const uint32_t * __a, uint32x4x3_t __b, const int __c)
{
  union { uint32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { uint32x4x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x8x3_t __attribute__ ((__always_inline__))
vld3q_lane_p16 (const poly16_t * __a, poly16x8x3_t __b, const int __c)
{
  union { poly16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  union { poly16x8x3_t __i; __builtin_neon_ci __o; } __rv;
  __rv.__o = __builtin_neon_vld3_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int8x8x3_t __attribute__ ((__always_inline__))
vld3_dup_s8 (const int8_t * __a)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x3_t __attribute__ ((__always_inline__))
vld3_dup_s16 (const int16_t * __a)
{
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x3_t __attribute__ ((__always_inline__))
vld3_dup_s32 (const int32_t * __a)
{
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x3_t __attribute__ ((__always_inline__))
vld3_dup_f32 (const float32_t * __a)
{
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x3_t __attribute__ ((__always_inline__))
vld3_dup_u8 (const uint8_t * __a)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x3_t __attribute__ ((__always_inline__))
vld3_dup_u16 (const uint16_t * __a)
{
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x3_t __attribute__ ((__always_inline__))
vld3_dup_u32 (const uint32_t * __a)
{
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x3_t __attribute__ ((__always_inline__))
vld3_dup_p8 (const poly8_t * __a)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x3_t __attribute__ ((__always_inline__))
vld3_dup_p16 (const poly16_t * __a)
{
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x3_t __attribute__ ((__always_inline__))
vld3_dup_s64 (const int64_t * __a)
{
  union { int64x1x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x3_t __attribute__ ((__always_inline__))
vld3_dup_u64 (const uint64_t * __a)
{
  union { uint64x1x3_t __i; __builtin_neon_ei __o; } __rv;
  __rv.__o = __builtin_neon_vld3_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_s8 (int8_t * __a, int8x8x3_t __b)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_s16 (int16_t * __a, int16x4x3_t __b)
{
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_s32 (int32_t * __a, int32x2x3_t __b)
{
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_f32 (float32_t * __a, float32x2x3_t __b)
{
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v2sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_u8 (uint8_t * __a, uint8x8x3_t __b)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_u16 (uint16_t * __a, uint16x4x3_t __b)
{
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_u32 (uint32_t * __a, uint32x2x3_t __b)
{
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_p8 (poly8_t * __a, poly8x8x3_t __b)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_p16 (poly16_t * __a, poly16x4x3_t __b)
{
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_s64 (int64_t * __a, int64x1x3_t __b)
{
  union { int64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_u64 (uint64_t * __a, uint64x1x3_t __b)
{
  union { uint64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_s8 (int8_t * __a, int8x16x3_t __b)
{
  union { int8x16x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_s16 (int16_t * __a, int16x8x3_t __b)
{
  union { int16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_s32 (int32_t * __a, int32x4x3_t __b)
{
  union { int32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_f32 (float32_t * __a, float32x4x3_t __b)
{
  union { float32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v4sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_u8 (uint8_t * __a, uint8x16x3_t __b)
{
  union { uint8x16x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_u16 (uint16_t * __a, uint16x8x3_t __b)
{
  union { uint16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_u32 (uint32_t * __a, uint32x4x3_t __b)
{
  union { uint32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_p8 (poly8_t * __a, poly8x16x3_t __b)
{
  union { poly8x16x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_p16 (poly16_t * __a, poly16x8x3_t __b)
{
  union { poly16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_s8 (int8_t * __a, int8x8x3_t __b, const int __c)
{
  union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_s16 (int16_t * __a, int16x4x3_t __b, const int __c)
{
  union { int16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_s32 (int32_t * __a, int32x2x3_t __b, const int __c)
{
  union { int32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_f32 (float32_t * __a, float32x2x3_t __b, const int __c)
{
  union { float32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev2sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_u8 (uint8_t * __a, uint8x8x3_t __b, const int __c)
{
  union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_u16 (uint16_t * __a, uint16x4x3_t __b, const int __c)
{
  union { uint16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_u32 (uint32_t * __a, uint32x2x3_t __b, const int __c)
{
  union { uint32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_p8 (poly8_t * __a, poly8x8x3_t __b, const int __c)
{
  union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3_lane_p16 (poly16_t * __a, poly16x4x3_t __b, const int __c)
{
  union { poly16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_s16 (int16_t * __a, int16x8x3_t __b, const int __c)
{
  union { int16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_s32 (int32_t * __a, int32x4x3_t __b, const int __c)
{
  union { int32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_f32 (float32_t * __a, float32x4x3_t __b, const int __c)
{
  union { float32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_u16 (uint16_t * __a, uint16x8x3_t __b, const int __c)
{
  union { uint16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_u32 (uint32_t * __a, uint32x4x3_t __b, const int __c)
{
  union { uint32x4x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst3q_lane_p16 (poly16_t * __a, poly16x8x3_t __b, const int __c)
{
  union { poly16x8x3_t __i; __builtin_neon_ci __o; } __bu = { __b };
  __builtin_neon_vst3_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline int8x8x4_t __attribute__ ((__always_inline__))
vld4_s8 (const int8_t * __a)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x4_t __attribute__ ((__always_inline__))
vld4_s16 (const int16_t * __a)
{
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x4_t __attribute__ ((__always_inline__))
vld4_s32 (const int32_t * __a)
{
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x4_t __attribute__ ((__always_inline__))
vld4_f32 (const float32_t * __a)
{
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x4_t __attribute__ ((__always_inline__))
vld4_u8 (const uint8_t * __a)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x4_t __attribute__ ((__always_inline__))
vld4_u16 (const uint16_t * __a)
{
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x4_t __attribute__ ((__always_inline__))
vld4_u32 (const uint32_t * __a)
{
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x4_t __attribute__ ((__always_inline__))
vld4_p8 (const poly8_t * __a)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x4_t __attribute__ ((__always_inline__))
vld4_p16 (const poly16_t * __a)
{
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x4_t __attribute__ ((__always_inline__))
vld4_s64 (const int64_t * __a)
{
  union { int64x1x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x4_t __attribute__ ((__always_inline__))
vld4_u64 (const uint64_t * __a)
{
  union { uint64x1x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4di ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x16x4_t __attribute__ ((__always_inline__))
vld4q_s8 (const int8_t * __a)
{
  union { int8x16x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x8x4_t __attribute__ ((__always_inline__))
vld4q_s16 (const int16_t * __a)
{
  union { int16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x4x4_t __attribute__ ((__always_inline__))
vld4q_s32 (const int32_t * __a)
{
  union { int32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x4x4_t __attribute__ ((__always_inline__))
vld4q_f32 (const float32_t * __a)
{
  union { float32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x16x4_t __attribute__ ((__always_inline__))
vld4q_u8 (const uint8_t * __a)
{
  union { uint8x16x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x8x4_t __attribute__ ((__always_inline__))
vld4q_u16 (const uint16_t * __a)
{
  union { uint16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x4x4_t __attribute__ ((__always_inline__))
vld4q_u32 (const uint32_t * __a)
{
  union { uint32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v4si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x16x4_t __attribute__ ((__always_inline__))
vld4q_p8 (const poly8_t * __a)
{
  union { poly8x16x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v16qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x8x4_t __attribute__ ((__always_inline__))
vld4q_p16 (const poly16_t * __a)
{
  union { poly16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4v8hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int8x8x4_t __attribute__ ((__always_inline__))
vld4_lane_s8 (const int8_t * __a, int8x8x4_t __b, const int __c)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x4x4_t __attribute__ ((__always_inline__))
vld4_lane_s16 (const int16_t * __a, int16x4x4_t __b, const int __c)
{
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x2x4_t __attribute__ ((__always_inline__))
vld4_lane_s32 (const int32_t * __a, int32x2x4_t __b, const int __c)
{
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x2x4_t __attribute__ ((__always_inline__))
vld4_lane_f32 (const float32_t * __a, float32x2x4_t __b, const int __c)
{
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev2sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint8x8x4_t __attribute__ ((__always_inline__))
vld4_lane_u8 (const uint8_t * __a, uint8x8x4_t __b, const int __c)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x4x4_t __attribute__ ((__always_inline__))
vld4_lane_u16 (const uint16_t * __a, uint16x4x4_t __b, const int __c)
{
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x2x4_t __attribute__ ((__always_inline__))
vld4_lane_u32 (const uint32_t * __a, uint32x2x4_t __b, const int __c)
{
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev2si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly8x8x4_t __attribute__ ((__always_inline__))
vld4_lane_p8 (const poly8_t * __a, poly8x8x4_t __b, const int __c)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8qi ((const __builtin_neon_qi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x4x4_t __attribute__ ((__always_inline__))
vld4_lane_p16 (const poly16_t * __a, poly16x4x4_t __b, const int __c)
{
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int16x8x4_t __attribute__ ((__always_inline__))
vld4q_lane_s16 (const int16_t * __a, int16x8x4_t __b, const int __c)
{
  union { int16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { int16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int32x4x4_t __attribute__ ((__always_inline__))
vld4q_lane_s32 (const int32_t * __a, int32x4x4_t __b, const int __c)
{
  union { int32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { int32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline float32x4x4_t __attribute__ ((__always_inline__))
vld4q_lane_f32 (const float32_t * __a, float32x4x4_t __b, const int __c)
{
  union { float32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { float32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4sf ((const __builtin_neon_sf *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint16x8x4_t __attribute__ ((__always_inline__))
vld4q_lane_u16 (const uint16_t * __a, uint16x8x4_t __b, const int __c)
{
  union { uint16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { uint16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline uint32x4x4_t __attribute__ ((__always_inline__))
vld4q_lane_u32 (const uint32_t * __a, uint32x4x4_t __b, const int __c)
{
  union { uint32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { uint32x4x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev4si ((const __builtin_neon_si *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline poly16x8x4_t __attribute__ ((__always_inline__))
vld4q_lane_p16 (const poly16_t * __a, poly16x8x4_t __b, const int __c)
{
  union { poly16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  union { poly16x8x4_t __i; __builtin_neon_xi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_lanev8hi ((const __builtin_neon_hi *) __a, __bu.__o, __c);
  return __rv.__i;
}

__extension__ static __inline int8x8x4_t __attribute__ ((__always_inline__))
vld4_dup_s8 (const int8_t * __a)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline int16x4x4_t __attribute__ ((__always_inline__))
vld4_dup_s16 (const int16_t * __a)
{
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int32x2x4_t __attribute__ ((__always_inline__))
vld4_dup_s32 (const int32_t * __a)
{
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline float32x2x4_t __attribute__ ((__always_inline__))
vld4_dup_f32 (const float32_t * __a)
{
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv2sf ((const __builtin_neon_sf *) __a);
  return __rv.__i;
}

__extension__ static __inline uint8x8x4_t __attribute__ ((__always_inline__))
vld4_dup_u8 (const uint8_t * __a)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint16x4x4_t __attribute__ ((__always_inline__))
vld4_dup_u16 (const uint16_t * __a)
{
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline uint32x2x4_t __attribute__ ((__always_inline__))
vld4_dup_u32 (const uint32_t * __a)
{
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv2si ((const __builtin_neon_si *) __a);
  return __rv.__i;
}

__extension__ static __inline poly8x8x4_t __attribute__ ((__always_inline__))
vld4_dup_p8 (const poly8_t * __a)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv8qi ((const __builtin_neon_qi *) __a);
  return __rv.__i;
}

__extension__ static __inline poly16x4x4_t __attribute__ ((__always_inline__))
vld4_dup_p16 (const poly16_t * __a)
{
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupv4hi ((const __builtin_neon_hi *) __a);
  return __rv.__i;
}

__extension__ static __inline int64x1x4_t __attribute__ ((__always_inline__))
vld4_dup_s64 (const int64_t * __a)
{
  union { int64x1x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline uint64x1x4_t __attribute__ ((__always_inline__))
vld4_dup_u64 (const uint64_t * __a)
{
  union { uint64x1x4_t __i; __builtin_neon_oi __o; } __rv;
  __rv.__o = __builtin_neon_vld4_dupdi ((const __builtin_neon_di *) __a);
  return __rv.__i;
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_s8 (int8_t * __a, int8x8x4_t __b)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_s16 (int16_t * __a, int16x4x4_t __b)
{
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_s32 (int32_t * __a, int32x2x4_t __b)
{
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_f32 (float32_t * __a, float32x2x4_t __b)
{
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v2sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_u8 (uint8_t * __a, uint8x8x4_t __b)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_u16 (uint16_t * __a, uint16x4x4_t __b)
{
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_u32 (uint32_t * __a, uint32x2x4_t __b)
{
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v2si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_p8 (poly8_t * __a, poly8x8x4_t __b)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_p16 (poly16_t * __a, poly16x4x4_t __b)
{
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_s64 (int64_t * __a, int64x1x4_t __b)
{
  union { int64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_u64 (uint64_t * __a, uint64x1x4_t __b)
{
  union { uint64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4di ((__builtin_neon_di *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_s8 (int8_t * __a, int8x16x4_t __b)
{
  union { int8x16x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_s16 (int16_t * __a, int16x8x4_t __b)
{
  union { int16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_s32 (int32_t * __a, int32x4x4_t __b)
{
  union { int32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_f32 (float32_t * __a, float32x4x4_t __b)
{
  union { float32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v4sf ((__builtin_neon_sf *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_u8 (uint8_t * __a, uint8x16x4_t __b)
{
  union { uint8x16x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_u16 (uint16_t * __a, uint16x8x4_t __b)
{
  union { uint16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_u32 (uint32_t * __a, uint32x4x4_t __b)
{
  union { uint32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v4si ((__builtin_neon_si *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_p8 (poly8_t * __a, poly8x16x4_t __b)
{
  union { poly8x16x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v16qi ((__builtin_neon_qi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_p16 (poly16_t * __a, poly16x8x4_t __b)
{
  union { poly16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4v8hi ((__builtin_neon_hi *) __a, __bu.__o);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_s8 (int8_t * __a, int8x8x4_t __b, const int __c)
{
  union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_s16 (int16_t * __a, int16x4x4_t __b, const int __c)
{
  union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_s32 (int32_t * __a, int32x2x4_t __b, const int __c)
{
  union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_f32 (float32_t * __a, float32x2x4_t __b, const int __c)
{
  union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev2sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_u8 (uint8_t * __a, uint8x8x4_t __b, const int __c)
{
  union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_u16 (uint16_t * __a, uint16x4x4_t __b, const int __c)
{
  union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_u32 (uint32_t * __a, uint32x2x4_t __b, const int __c)
{
  union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev2si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_p8 (poly8_t * __a, poly8x8x4_t __b, const int __c)
{
  union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8qi ((__builtin_neon_qi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4_lane_p16 (poly16_t * __a, poly16x4x4_t __b, const int __c)
{
  union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_s16 (int16_t * __a, int16x8x4_t __b, const int __c)
{
  union { int16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_s32 (int32_t * __a, int32x4x4_t __b, const int __c)
{
  union { int32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_f32 (float32_t * __a, float32x4x4_t __b, const int __c)
{
  union { float32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4sf ((__builtin_neon_sf *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_u16 (uint16_t * __a, uint16x8x4_t __b, const int __c)
{
  union { uint16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_u32 (uint32_t * __a, uint32x4x4_t __b, const int __c)
{
  union { uint32x4x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev4si ((__builtin_neon_si *) __a, __bu.__o, __c);
}

__extension__ static __inline void __attribute__ ((__always_inline__))
vst4q_lane_p16 (poly16_t * __a, poly16x8x4_t __b, const int __c)
{
  union { poly16x8x4_t __i; __builtin_neon_xi __o; } __bu = { __b };
  __builtin_neon_vst4_lanev8hi ((__builtin_neon_hi *) __a, __bu.__o, __c);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vand_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vandv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vand_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vandv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vand_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vandv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vand_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vandv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vand_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vandv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vand_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vandv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vand_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vanddi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vand_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vanddi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vandq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vandv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vandq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vandv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vandq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vandv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vandq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vandv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vandq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vandv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vandq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vandv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vandq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vandv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vandq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vandv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vorr_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vorrv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vorr_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vorrv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vorr_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vorrv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vorr_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vorrv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vorr_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vorrv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vorr_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vorrv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vorr_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vorrdi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vorr_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vorrdi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vorrq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vorrv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vorrq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vorrv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vorrq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vorrv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vorrq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vorrv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vorrq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vorrv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vorrq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vorrv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vorrq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vorrv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vorrq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vorrv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
veor_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_veorv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
veor_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_veorv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
veor_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_veorv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
veor_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_veorv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
veor_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_veorv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
veor_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_veorv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
veor_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_veordi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
veor_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_veordi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
veorq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_veorv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
veorq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_veorv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
veorq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_veorv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
veorq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_veorv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
veorq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_veorv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
veorq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_veorv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
veorq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_veorv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
veorq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_veorv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vbic_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vbicv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vbic_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vbicv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vbic_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vbicv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vbic_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vbicv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vbic_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vbicv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vbic_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vbicv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vbic_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vbicdi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vbic_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vbicdi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vbicq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vbicv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vbicq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vbicv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vbicq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vbicv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vbicq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vbicv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vbicq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vbicv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vbicq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vbicv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vbicq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vbicv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vbicq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vbicv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vorn_s8 (int8x8_t __a, int8x8_t __b)
{
  return (int8x8_t)__builtin_neon_vornv8qi (__a, __b, 1);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vorn_s16 (int16x4_t __a, int16x4_t __b)
{
  return (int16x4_t)__builtin_neon_vornv4hi (__a, __b, 1);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vorn_s32 (int32x2_t __a, int32x2_t __b)
{
  return (int32x2_t)__builtin_neon_vornv2si (__a, __b, 1);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vorn_u8 (uint8x8_t __a, uint8x8_t __b)
{
  return (uint8x8_t)__builtin_neon_vornv8qi ((int8x8_t) __a, (int8x8_t) __b, 0);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vorn_u16 (uint16x4_t __a, uint16x4_t __b)
{
  return (uint16x4_t)__builtin_neon_vornv4hi ((int16x4_t) __a, (int16x4_t) __b, 0);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vorn_u32 (uint32x2_t __a, uint32x2_t __b)
{
  return (uint32x2_t)__builtin_neon_vornv2si ((int32x2_t) __a, (int32x2_t) __b, 0);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vorn_s64 (int64x1_t __a, int64x1_t __b)
{
  return (int64x1_t)__builtin_neon_vorndi (__a, __b, 1);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vorn_u64 (uint64x1_t __a, uint64x1_t __b)
{
  return (uint64x1_t)__builtin_neon_vorndi ((int64x1_t) __a, (int64x1_t) __b, 0);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vornq_s8 (int8x16_t __a, int8x16_t __b)
{
  return (int8x16_t)__builtin_neon_vornv16qi (__a, __b, 1);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vornq_s16 (int16x8_t __a, int16x8_t __b)
{
  return (int16x8_t)__builtin_neon_vornv8hi (__a, __b, 1);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vornq_s32 (int32x4_t __a, int32x4_t __b)
{
  return (int32x4_t)__builtin_neon_vornv4si (__a, __b, 1);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vornq_s64 (int64x2_t __a, int64x2_t __b)
{
  return (int64x2_t)__builtin_neon_vornv2di (__a, __b, 1);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vornq_u8 (uint8x16_t __a, uint8x16_t __b)
{
  return (uint8x16_t)__builtin_neon_vornv16qi ((int8x16_t) __a, (int8x16_t) __b, 0);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vornq_u16 (uint16x8_t __a, uint16x8_t __b)
{
  return (uint16x8_t)__builtin_neon_vornv8hi ((int16x8_t) __a, (int16x8_t) __b, 0);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vornq_u32 (uint32x4_t __a, uint32x4_t __b)
{
  return (uint32x4_t)__builtin_neon_vornv4si ((int32x4_t) __a, (int32x4_t) __b, 0);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vornq_u64 (uint64x2_t __a, uint64x2_t __b)
{
  return (uint64x2_t)__builtin_neon_vornv2di ((int64x2_t) __a, (int64x2_t) __b, 0);
}


__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_s8 (int8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv8qi (__a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_s16 (int16x4_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv4hi (__a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_s32 (int32x2_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv2si (__a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_s64 (int64x1_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qidi (__a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_f32 (float32x2_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv2sf (__a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_u8 (uint8x8_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv8qi ((int8x8_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_u16 (uint16x4_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_u32 (uint32x2_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv2si ((int32x2_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_u64 (uint64x1_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qidi ((int64x1_t) __a);
}

__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
vreinterpret_p8_p16 (poly16x4_t __a)
{
  return (poly8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_s8 (int8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv16qi (__a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_s16 (int16x8_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv8hi (__a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_s32 (int32x4_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv4si (__a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_s64 (int64x2_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv2di (__a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_f32 (float32x4_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv4sf (__a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_u8 (uint8x16_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv16qi ((int8x16_t) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_u16 (uint16x8_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_u32 (uint32x4_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv4si ((int32x4_t) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_u64 (uint64x2_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv2di ((int64x2_t) __a);
}

__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
vreinterpretq_p8_p16 (poly16x8_t __a)
{
  return (poly8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_s8 (int8x8_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv8qi (__a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_s16 (int16x4_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv4hi (__a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_s32 (int32x2_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv2si (__a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_s64 (int64x1_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hidi (__a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_f32 (float32x2_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv2sf (__a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_u8 (uint8x8_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_u16 (uint16x4_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv4hi ((int16x4_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_u32 (uint32x2_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv2si ((int32x2_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_u64 (uint64x1_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hidi ((int64x1_t) __a);
}

__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
vreinterpret_p16_p8 (poly8x8_t __a)
{
  return (poly16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_s8 (int8x16_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv16qi (__a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_s16 (int16x8_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv8hi (__a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_s32 (int32x4_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv4si (__a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_s64 (int64x2_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv2di (__a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_f32 (float32x4_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv4sf (__a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_u8 (uint8x16_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_u16 (uint16x8_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv8hi ((int16x8_t) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_u32 (uint32x4_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv4si ((int32x4_t) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_u64 (uint64x2_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv2di ((int64x2_t) __a);
}

__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
vreinterpretq_p16_p8 (poly8x16_t __a)
{
  return (poly16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_s8 (int8x8_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv8qi (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_s16 (int16x4_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv4hi (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_s32 (int32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv2si (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_s64 (int64x1_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfdi (__a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_u8 (uint8x8_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv8qi ((int8x8_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_u16 (uint16x4_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv4hi ((int16x4_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_u32 (uint32x2_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv2si ((int32x2_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_u64 (uint64x1_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfdi ((int64x1_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_p8 (poly8x8_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv8qi ((int8x8_t) __a);
}

__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
vreinterpret_f32_p16 (poly16x4_t __a)
{
  return (float32x2_t)__builtin_neon_vreinterpretv2sfv4hi ((int16x4_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_s8 (int8x16_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv16qi (__a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_s16 (int16x8_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv8hi (__a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_s32 (int32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv4si (__a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_s64 (int64x2_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv2di (__a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_u8 (uint8x16_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv16qi ((int8x16_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_u16 (uint16x8_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv8hi ((int16x8_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_u32 (uint32x4_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv4si ((int32x4_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_u64 (uint64x2_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv2di ((int64x2_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_p8 (poly8x16_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv16qi ((int8x16_t) __a);
}

__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
vreinterpretq_f32_p16 (poly16x8_t __a)
{
  return (float32x4_t)__builtin_neon_vreinterpretv4sfv8hi ((int16x8_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_s8 (int8x8_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv8qi (__a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_s16 (int16x4_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv4hi (__a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_s32 (int32x2_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv2si (__a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_f32 (float32x2_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv2sf (__a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_u8 (uint8x8_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_u16 (uint16x4_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_u32 (uint32x2_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv2si ((int32x2_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_u64 (uint64x1_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdidi ((int64x1_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_p8 (poly8x8_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vreinterpret_s64_p16 (poly16x4_t __a)
{
  return (int64x1_t)__builtin_neon_vreinterpretdiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_s8 (int8x16_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div16qi (__a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_s16 (int16x8_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div8hi (__a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_s32 (int32x4_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div4si (__a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_f32 (float32x4_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div4sf (__a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_u8 (uint8x16_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div16qi ((int8x16_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_u16 (uint16x8_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div8hi ((int16x8_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_u32 (uint32x4_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div4si ((int32x4_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_u64 (uint64x2_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div2di ((int64x2_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_p8 (poly8x16_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div16qi ((int8x16_t) __a);
}

__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
vreinterpretq_s64_p16 (poly16x8_t __a)
{
  return (int64x2_t)__builtin_neon_vreinterpretv2div8hi ((int16x8_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_s8 (int8x8_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv8qi (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_s16 (int16x4_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv4hi (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_s32 (int32x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv2si (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_s64 (int64x1_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdidi (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_f32 (float32x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv2sf (__a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_u8 (uint8x8_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_u16 (uint16x4_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_u32 (uint32x2_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv2si ((int32x2_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_p8 (poly8x8_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
vreinterpret_u64_p16 (poly16x4_t __a)
{
  return (uint64x1_t)__builtin_neon_vreinterpretdiv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_s8 (int8x16_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div16qi (__a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_s16 (int16x8_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div8hi (__a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_s32 (int32x4_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div4si (__a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_s64 (int64x2_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div2di (__a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_f32 (float32x4_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div4sf (__a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_u8 (uint8x16_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div16qi ((int8x16_t) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_u16 (uint16x8_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div8hi ((int16x8_t) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_u32 (uint32x4_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div4si ((int32x4_t) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_p8 (poly8x16_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div16qi ((int8x16_t) __a);
}

__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
vreinterpretq_u64_p16 (poly16x8_t __a)
{
  return (uint64x2_t)__builtin_neon_vreinterpretv2div8hi ((int16x8_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_s16 (int16x4_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv4hi (__a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_s32 (int32x2_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv2si (__a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_s64 (int64x1_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qidi (__a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_f32 (float32x2_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv2sf (__a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_u8 (uint8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_u16 (uint16x4_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_u32 (uint32x2_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv2si ((int32x2_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_u64 (uint64x1_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qidi ((int64x1_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_p8 (poly8x8_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
vreinterpret_s8_p16 (poly16x4_t __a)
{
  return (int8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_s16 (int16x8_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv8hi (__a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_s32 (int32x4_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv4si (__a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_s64 (int64x2_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv2di (__a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_f32 (float32x4_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv4sf (__a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_u8 (uint8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv16qi ((int8x16_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_u16 (uint16x8_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_u32 (uint32x4_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv4si ((int32x4_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_u64 (uint64x2_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv2di ((int64x2_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_p8 (poly8x16_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv16qi ((int8x16_t) __a);
}

__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
vreinterpretq_s8_p16 (poly16x8_t __a)
{
  return (int8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_s8 (int8x8_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv8qi (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_s32 (int32x2_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv2si (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_s64 (int64x1_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hidi (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_f32 (float32x2_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv2sf (__a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_u8 (uint8x8_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_u16 (uint16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_u32 (uint32x2_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv2si ((int32x2_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_u64 (uint64x1_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hidi ((int64x1_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_p8 (poly8x8_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
vreinterpret_s16_p16 (poly16x4_t __a)
{
  return (int16x4_t)__builtin_neon_vreinterpretv4hiv4hi ((int16x4_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_s8 (int8x16_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv16qi (__a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_s32 (int32x4_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv4si (__a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_s64 (int64x2_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv2di (__a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_f32 (float32x4_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv4sf (__a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_u8 (uint8x16_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_u16 (uint16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv8hi ((int16x8_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_u32 (uint32x4_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv4si ((int32x4_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_u64 (uint64x2_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv2di ((int64x2_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_p8 (poly8x16_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
vreinterpretq_s16_p16 (poly16x8_t __a)
{
  return (int16x8_t)__builtin_neon_vreinterpretv8hiv8hi ((int16x8_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_s8 (int8x8_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv8qi (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_s16 (int16x4_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv4hi (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_s64 (int64x1_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2sidi (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_f32 (float32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv2sf (__a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_u8 (uint8x8_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv8qi ((int8x8_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_u16 (uint16x4_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv4hi ((int16x4_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_u32 (uint32x2_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv2si ((int32x2_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_u64 (uint64x1_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2sidi ((int64x1_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_p8 (poly8x8_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv8qi ((int8x8_t) __a);
}

__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
vreinterpret_s32_p16 (poly16x4_t __a)
{
  return (int32x2_t)__builtin_neon_vreinterpretv2siv4hi ((int16x4_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_s8 (int8x16_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv16qi (__a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_s16 (int16x8_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv8hi (__a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_s64 (int64x2_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv2di (__a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_f32 (float32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv4sf (__a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_u8 (uint8x16_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv16qi ((int8x16_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_u16 (uint16x8_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv8hi ((int16x8_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_u32 (uint32x4_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv4si ((int32x4_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_u64 (uint64x2_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv2di ((int64x2_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_p8 (poly8x16_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv16qi ((int8x16_t) __a);
}

__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
vreinterpretq_s32_p16 (poly16x8_t __a)
{
  return (int32x4_t)__builtin_neon_vreinterpretv4siv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_s8 (int8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv8qi (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_s16 (int16x4_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv4hi (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_s32 (int32x2_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv2si (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_s64 (int64x1_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qidi (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_f32 (float32x2_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv2sf (__a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_u16 (uint16x4_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_u32 (uint32x2_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv2si ((int32x2_t) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_u64 (uint64x1_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qidi ((int64x1_t) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_p8 (poly8x8_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
vreinterpret_u8_p16 (poly16x4_t __a)
{
  return (uint8x8_t)__builtin_neon_vreinterpretv8qiv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_s8 (int8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv16qi (__a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_s16 (int16x8_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv8hi (__a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_s32 (int32x4_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv4si (__a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_s64 (int64x2_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv2di (__a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_f32 (float32x4_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv4sf (__a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_u16 (uint16x8_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_u32 (uint32x4_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv4si ((int32x4_t) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_u64 (uint64x2_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv2di ((int64x2_t) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_p8 (poly8x16_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
vreinterpretq_u8_p16 (poly16x8_t __a)
{
  return (uint8x16_t)__builtin_neon_vreinterpretv16qiv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_s8 (int8x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv8qi (__a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_s16 (int16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv4hi (__a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_s32 (int32x2_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv2si (__a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_s64 (int64x1_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hidi (__a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_f32 (float32x2_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv2sf (__a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_u8 (uint8x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_u32 (uint32x2_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv2si ((int32x2_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_u64 (uint64x1_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hidi ((int64x1_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_p8 (poly8x8_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
vreinterpret_u16_p16 (poly16x4_t __a)
{
  return (uint16x4_t)__builtin_neon_vreinterpretv4hiv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_s8 (int8x16_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv16qi (__a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_s16 (int16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv8hi (__a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_s32 (int32x4_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv4si (__a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_s64 (int64x2_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv2di (__a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_f32 (float32x4_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv4sf (__a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_u8 (uint8x16_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_u32 (uint32x4_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv4si ((int32x4_t) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_u64 (uint64x2_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv2di ((int64x2_t) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_p8 (poly8x16_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
vreinterpretq_u16_p16 (poly16x8_t __a)
{
  return (uint16x8_t)__builtin_neon_vreinterpretv8hiv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_s8 (int8x8_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv8qi (__a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_s16 (int16x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv4hi (__a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_s32 (int32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv2si (__a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_s64 (int64x1_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2sidi (__a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_f32 (float32x2_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv2sf (__a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_u8 (uint8x8_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_u16 (uint16x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_u64 (uint64x1_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2sidi ((int64x1_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_p8 (poly8x8_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv8qi ((int8x8_t) __a);
}

__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
vreinterpret_u32_p16 (poly16x4_t __a)
{
  return (uint32x2_t)__builtin_neon_vreinterpretv2siv4hi ((int16x4_t) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_s8 (int8x16_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv16qi (__a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_s16 (int16x8_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv8hi (__a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_s32 (int32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv4si (__a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_s64 (int64x2_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv2di (__a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_f32 (float32x4_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv4sf (__a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_u8 (uint8x16_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_u16 (uint16x8_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv8hi ((int16x8_t) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_u64 (uint64x2_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv2di ((int64x2_t) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_p8 (poly8x16_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv16qi ((int8x16_t) __a);
}

__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
vreinterpretq_u32_p16 (poly16x8_t __a)
{
  return (uint32x4_t)__builtin_neon_vreinterpretv4siv8hi ((int16x8_t) __a);
}
# 19 "jsimd_arm.c" 2


# 1 "../jinclude.h" 1
# 20 "../jinclude.h"
# 1 "../jconfig.h" 1
# 21 "../jinclude.h" 2
# 35 "../jinclude.h"
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 150 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 3 4
typedef int ptrdiff_t;
# 212 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 3 4
typedef unsigned int size_t;
# 324 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 3 4
typedef unsigned int wchar_t;
# 36 "../jinclude.h" 2



# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 1 3 4
# 33 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 34 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 2 3 4








# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/waitflags.h" 1 3 4
# 43 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 2 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/waitstatus.h" 1 3 4
# 65 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/waitstatus.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/endian.h" 1 3 4
# 37 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/endian.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/endian.h" 1 3 4
# 38 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/endian.h" 2 3 4
# 61 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/endian.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/byteswap.h" 1 3 4
# 62 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/endian.h" 2 3 4
# 66 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/waitstatus.h" 2 3 4

union wait
  {
    int w_status;
    struct
      {

 unsigned int __w_termsig:7;
 unsigned int __w_coredump:1;
 unsigned int __w_retcode:8;
 unsigned int:16;







      } __wait_terminated;
    struct
      {

 unsigned int __w_stopval:8;
 unsigned int __w_stopsig:8;
 unsigned int:16;






      } __wait_stopped;
  };
# 44 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 2 3 4
# 68 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
typedef union
  {
    union wait *__uptr;
    int *__iptr;
  } __WAIT_STATUS __attribute__ ((__transparent_union__));
# 96 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4


typedef struct
  {
    int quot;
    int rem;
  } div_t;



typedef struct
  {
    long int quot;
    long int rem;
  } ldiv_t;







__extension__ typedef struct
  {
    long long int quot;
    long long int rem;
  } lldiv_t;


# 140 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern size_t __ctype_get_mb_cur_max (void) __attribute__ ((__nothrow__)) ;




extern double atof (__const char *__nptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1))) ;

extern int atoi (__const char *__nptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1))) ;

extern long int atol (__const char *__nptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1))) ;





__extension__ extern long long int atoll (__const char *__nptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1))) ;





extern double strtod (__const char *__restrict __nptr,
        char **__restrict __endptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;





extern float strtof (__const char *__restrict __nptr,
       char **__restrict __endptr) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;

extern long double strtold (__const char *__restrict __nptr,
       char **__restrict __endptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;





extern long int strtol (__const char *__restrict __nptr,
   char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;

extern unsigned long int strtoul (__const char *__restrict __nptr,
      char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;




__extension__
extern long long int strtoq (__const char *__restrict __nptr,
        char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;

__extension__
extern unsigned long long int strtouq (__const char *__restrict __nptr,
           char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;





__extension__
extern long long int strtoll (__const char *__restrict __nptr,
         char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;

__extension__
extern unsigned long long int strtoull (__const char *__restrict __nptr,
     char **__restrict __endptr, int __base)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;

# 277 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4

extern __inline double
__attribute__ ((__nothrow__)) atof (__const char *__nptr)
{
  return strtod (__nptr, (char **) ((void *)0));
}
extern __inline int
__attribute__ ((__nothrow__)) atoi (__const char *__nptr)
{
  return (int) strtol (__nptr, (char **) ((void *)0), 10);
}
extern __inline long int
__attribute__ ((__nothrow__)) atol (__const char *__nptr)
{
  return strtol (__nptr, (char **) ((void *)0), 10);
}




__extension__ extern __inline long long int
__attribute__ ((__nothrow__)) atoll (__const char *__nptr)
{
  return strtoll (__nptr, (char **) ((void *)0), 10);
}

# 311 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern char *l64a (long int __n) __attribute__ ((__nothrow__)) ;


extern long int a64l (__const char *__s)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1))) ;




# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 1 3 4
# 28 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4


# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/types.h" 1 3 4
# 28 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/types.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/wordsize.h" 1 3 4
# 29 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/types.h" 2 3 4


typedef unsigned char __u_char;
typedef unsigned short int __u_short;
typedef unsigned int __u_int;
typedef unsigned long int __u_long;


typedef signed char __int8_t;
typedef unsigned char __uint8_t;
typedef signed short int __int16_t;
typedef unsigned short int __uint16_t;
typedef signed int __int32_t;
typedef unsigned int __uint32_t;




__extension__ typedef signed long long int __int64_t;
__extension__ typedef unsigned long long int __uint64_t;







__extension__ typedef long long int __quad_t;
__extension__ typedef unsigned long long int __u_quad_t;
# 131 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/types.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/typesizes.h" 1 3 4
# 132 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/types.h" 2 3 4


__extension__ typedef __u_quad_t __dev_t;
__extension__ typedef unsigned int __uid_t;
__extension__ typedef unsigned int __gid_t;
__extension__ typedef unsigned long int __ino_t;
__extension__ typedef __u_quad_t __ino64_t;
__extension__ typedef unsigned int __mode_t;
__extension__ typedef unsigned int __nlink_t;
__extension__ typedef long int __off_t;
__extension__ typedef __quad_t __off64_t;
__extension__ typedef int __pid_t;
__extension__ typedef struct { int __val[2]; } __fsid_t;
__extension__ typedef long int __clock_t;
__extension__ typedef unsigned long int __rlim_t;
__extension__ typedef __u_quad_t __rlim64_t;
__extension__ typedef unsigned int __id_t;
__extension__ typedef long int __time_t;
__extension__ typedef unsigned int __useconds_t;
__extension__ typedef long int __suseconds_t;

__extension__ typedef int __daddr_t;
__extension__ typedef long int __swblk_t;
__extension__ typedef int __key_t;


__extension__ typedef int __clockid_t;


__extension__ typedef void * __timer_t;


__extension__ typedef long int __blksize_t;




__extension__ typedef long int __blkcnt_t;
__extension__ typedef __quad_t __blkcnt64_t;


__extension__ typedef unsigned long int __fsblkcnt_t;
__extension__ typedef __u_quad_t __fsblkcnt64_t;


__extension__ typedef unsigned long int __fsfilcnt_t;
__extension__ typedef __u_quad_t __fsfilcnt64_t;

__extension__ typedef int __ssize_t;



typedef __off64_t __loff_t;
typedef __quad_t *__qaddr_t;
typedef char *__caddr_t;


__extension__ typedef int __intptr_t;


__extension__ typedef unsigned int __socklen_t;
# 31 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4



typedef __u_char u_char;
typedef __u_short u_short;
typedef __u_int u_int;
typedef __u_long u_long;
typedef __quad_t quad_t;
typedef __u_quad_t u_quad_t;
typedef __fsid_t fsid_t;




typedef __loff_t loff_t;



typedef __ino_t ino_t;
# 61 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
typedef __dev_t dev_t;




typedef __gid_t gid_t;




typedef __mode_t mode_t;




typedef __nlink_t nlink_t;




typedef __uid_t uid_t;





typedef __off_t off_t;
# 99 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
typedef __pid_t pid_t;





typedef __id_t id_t;




typedef __ssize_t ssize_t;





typedef __daddr_t daddr_t;
typedef __caddr_t caddr_t;





typedef __key_t key_t;
# 133 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 1 3 4
# 58 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 3 4


typedef __clock_t clock_t;



# 74 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 3 4


typedef __time_t time_t;



# 92 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 3 4
typedef __clockid_t clockid_t;
# 104 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 3 4
typedef __timer_t timer_t;
# 134 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4
# 147 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 148 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4



typedef unsigned long int ulong;
typedef unsigned short int ushort;
typedef unsigned int uint;
# 201 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
typedef unsigned int u_int8_t __attribute__ ((__mode__ (__QI__)));
typedef unsigned int u_int16_t __attribute__ ((__mode__ (__HI__)));
typedef unsigned int u_int32_t __attribute__ ((__mode__ (__SI__)));
typedef unsigned int u_int64_t __attribute__ ((__mode__ (__DI__)));

typedef int register_t __attribute__ ((__mode__ (__word__)));
# 220 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 1 3 4
# 31 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/select.h" 1 3 4
# 32 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 2 3 4


# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/sigset.h" 1 3 4
# 24 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/sigset.h" 3 4
typedef int __sig_atomic_t;




typedef struct
  {
    unsigned long int __val[(1024 / (8 * sizeof (unsigned long int)))];
  } __sigset_t;
# 35 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 2 3 4



typedef __sigset_t sigset_t;





# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 1 3 4
# 120 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/time.h" 3 4
struct timespec
  {
    __time_t tv_sec;
    long int tv_nsec;
  };
# 45 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 2 3 4

# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/time.h" 1 3 4
# 75 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/time.h" 3 4
struct timeval
  {
    __time_t tv_sec;
    __suseconds_t tv_usec;
  };
# 47 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 2 3 4


typedef __suseconds_t suseconds_t;





typedef long int __fd_mask;
# 67 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 3 4
typedef struct
  {






    __fd_mask __fds_bits[1024 / (8 * (int) sizeof (__fd_mask))];


  } fd_set;






typedef __fd_mask fd_mask;
# 99 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 3 4

# 109 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 3 4
extern int select (int __nfds, fd_set *__restrict __readfds,
     fd_set *__restrict __writefds,
     fd_set *__restrict __exceptfds,
     struct timeval *__restrict __timeout);
# 121 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/select.h" 3 4
extern int pselect (int __nfds, fd_set *__restrict __readfds,
      fd_set *__restrict __writefds,
      fd_set *__restrict __exceptfds,
      const struct timespec *__restrict __timeout,
      const __sigset_t *__restrict __sigmask);



# 221 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4


# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/sysmacros.h" 1 3 4
# 30 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/sysmacros.h" 3 4
__extension__
extern unsigned int gnu_dev_major (unsigned long long int __dev)
     __attribute__ ((__nothrow__));
__extension__
extern unsigned int gnu_dev_minor (unsigned long long int __dev)
     __attribute__ ((__nothrow__));
__extension__
extern unsigned long long int gnu_dev_makedev (unsigned int __major,
            unsigned int __minor)
     __attribute__ ((__nothrow__));


__extension__ extern __inline unsigned int
__attribute__ ((__nothrow__)) gnu_dev_major (unsigned long long int __dev)
{
  return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
}

__extension__ extern __inline unsigned int
__attribute__ ((__nothrow__)) gnu_dev_minor (unsigned long long int __dev)
{
  return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
}

__extension__ extern __inline unsigned long long int
__attribute__ ((__nothrow__)) gnu_dev_makedev (unsigned int __major, unsigned int __minor)
{
  return ((__minor & 0xff) | ((__major & 0xfff) << 8)
   | (((unsigned long long int) (__minor & ~0xff)) << 12)
   | (((unsigned long long int) (__major & ~0xfff)) << 32));
}
# 224 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4





typedef __blksize_t blksize_t;






typedef __blkcnt_t blkcnt_t;



typedef __fsblkcnt_t fsblkcnt_t;



typedef __fsfilcnt_t fsfilcnt_t;
# 271 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/pthreadtypes.h" 1 3 4
# 38 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/pthreadtypes.h" 3 4
typedef unsigned long int pthread_t;


typedef union
{
  char __size[36];
  long int __align;
} pthread_attr_t;


typedef struct __pthread_internal_slist
{
  struct __pthread_internal_slist *__next;
} __pthread_slist_t;




typedef union
{
  struct __pthread_mutex_s
  {
    int __lock;
    unsigned int __count;
    int __owner;


    int __kind;
    unsigned int __nusers;
    __extension__ union
    {
      int __spins;
      __pthread_slist_t __list;
    };
  } __data;
  char __size[24];
  long int __align;
} pthread_mutex_t;

typedef union
{
  char __size[4];
  long int __align;
} pthread_mutexattr_t;




typedef union
{
  struct
  {
    int __lock;
    unsigned int __futex;
    __extension__ unsigned long long int __total_seq;
    __extension__ unsigned long long int __wakeup_seq;
    __extension__ unsigned long long int __woken_seq;
    void *__mutex;
    unsigned int __nwaiters;
    unsigned int __broadcast_seq;
  } __data;
  char __size[48];
  __extension__ long long int __align;
} pthread_cond_t;

typedef union
{
  char __size[4];
  long int __align;
} pthread_condattr_t;



typedef unsigned int pthread_key_t;



typedef int pthread_once_t;





typedef union
{
  struct
  {
    int __lock;
    unsigned int __nr_readers;
    unsigned int __readers_wakeup;
    unsigned int __writer_wakeup;
    unsigned int __nr_readers_queued;
    unsigned int __nr_writers_queued;
# 141 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/pthreadtypes.h" 3 4
    unsigned char __flags;
    unsigned char __shared;
    unsigned char __pad1;
    unsigned char __pad2;

    int __writer;
  } __data;
  char __size[32];
  long int __align;
} pthread_rwlock_t;

typedef union
{
  char __size[8];
  long int __align;
} pthread_rwlockattr_t;





typedef volatile int pthread_spinlock_t;




typedef union
{
  char __size[20];
  long int __align;
} pthread_barrier_t;

typedef union
{
  char __size[4];
  int __align;
} pthread_barrierattr_t;
# 272 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/sys/types.h" 2 3 4



# 321 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 2 3 4






extern long int random (void) __attribute__ ((__nothrow__));


extern void srandom (unsigned int __seed) __attribute__ ((__nothrow__));





extern char *initstate (unsigned int __seed, char *__statebuf,
   size_t __statelen) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));



extern char *setstate (char *__statebuf) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));







struct random_data
  {
    int32_t *fptr;
    int32_t *rptr;
    int32_t *state;
    int rand_type;
    int rand_deg;
    int rand_sep;
    int32_t *end_ptr;
  };

extern int random_r (struct random_data *__restrict __buf,
       int32_t *__restrict __result) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));

extern int srandom_r (unsigned int __seed, struct random_data *__buf)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));

extern int initstate_r (unsigned int __seed, char *__restrict __statebuf,
   size_t __statelen,
   struct random_data *__restrict __buf)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2, 4)));

extern int setstate_r (char *__restrict __statebuf,
         struct random_data *__restrict __buf)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));






extern int rand (void) __attribute__ ((__nothrow__));

extern void srand (unsigned int __seed) __attribute__ ((__nothrow__));




extern int rand_r (unsigned int *__seed) __attribute__ ((__nothrow__));







extern double drand48 (void) __attribute__ ((__nothrow__));
extern double erand48 (unsigned short int __xsubi[3]) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));


extern long int lrand48 (void) __attribute__ ((__nothrow__));
extern long int nrand48 (unsigned short int __xsubi[3])
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));


extern long int mrand48 (void) __attribute__ ((__nothrow__));
extern long int jrand48 (unsigned short int __xsubi[3])
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));


extern void srand48 (long int __seedval) __attribute__ ((__nothrow__));
extern unsigned short int *seed48 (unsigned short int __seed16v[3])
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));
extern void lcong48 (unsigned short int __param[7]) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));





struct drand48_data
  {
    unsigned short int __x[3];
    unsigned short int __old_x[3];
    unsigned short int __c;
    unsigned short int __init;
    unsigned long long int __a;
  };


extern int drand48_r (struct drand48_data *__restrict __buffer,
        double *__restrict __result) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
extern int erand48_r (unsigned short int __xsubi[3],
        struct drand48_data *__restrict __buffer,
        double *__restrict __result) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern int lrand48_r (struct drand48_data *__restrict __buffer,
        long int *__restrict __result)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
extern int nrand48_r (unsigned short int __xsubi[3],
        struct drand48_data *__restrict __buffer,
        long int *__restrict __result)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern int mrand48_r (struct drand48_data *__restrict __buffer,
        long int *__restrict __result)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
extern int jrand48_r (unsigned short int __xsubi[3],
        struct drand48_data *__restrict __buffer,
        long int *__restrict __result)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern int srand48_r (long int __seedval, struct drand48_data *__buffer)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));

extern int seed48_r (unsigned short int __seed16v[3],
       struct drand48_data *__buffer) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));

extern int lcong48_r (unsigned short int __param[7],
        struct drand48_data *__buffer)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));









extern void *malloc (size_t __size) __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) ;

extern void *calloc (size_t __nmemb, size_t __size)
     __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) ;










extern void *realloc (void *__ptr, size_t __size)
     __attribute__ ((__nothrow__)) __attribute__ ((__warn_unused_result__));

extern void free (void *__ptr) __attribute__ ((__nothrow__));




extern void cfree (void *__ptr) __attribute__ ((__nothrow__));



# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/alloca.h" 1 3 4
# 25 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/alloca.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 26 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/alloca.h" 2 3 4







extern void *alloca (size_t __size) __attribute__ ((__nothrow__));






# 498 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 2 3 4





extern void *valloc (size_t __size) __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) ;




extern int posix_memalign (void **__memptr, size_t __alignment, size_t __size)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;




extern void abort (void) __attribute__ ((__nothrow__)) __attribute__ ((__noreturn__));



extern int atexit (void (*__func) (void)) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));
# 531 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4





extern int on_exit (void (*__func) (int __status, void *__arg), void *__arg)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));






extern void exit (int __status) __attribute__ ((__nothrow__)) __attribute__ ((__noreturn__));
# 554 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4






extern void _Exit (int __status) __attribute__ ((__nothrow__)) __attribute__ ((__noreturn__));






extern char *getenv (__const char *__name) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;




extern char *__secure_getenv (__const char *__name)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;





extern int putenv (char *__string) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));





extern int setenv (__const char *__name, __const char *__value, int __replace)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));


extern int unsetenv (__const char *__name) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));






extern int clearenv (void) __attribute__ ((__nothrow__));
# 606 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern char *mktemp (char *__template) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;
# 620 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern int mkstemp (char *__template) __attribute__ ((__nonnull__ (1))) ;
# 642 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern int mkstemps (char *__template, int __suffixlen) __attribute__ ((__nonnull__ (1))) ;
# 663 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern char *mkdtemp (char *__template) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;
# 712 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4





extern int system (__const char *__command) ;

# 734 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern char *realpath (__const char *__restrict __name,
         char *__restrict __resolved) __attribute__ ((__nothrow__)) ;






typedef int (*__compar_fn_t) (__const void *, __const void *);
# 752 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4



extern void *bsearch (__const void *__key, __const void *__base,
        size_t __nmemb, size_t __size, __compar_fn_t __compar)
     __attribute__ ((__nonnull__ (1, 2, 5))) ;



extern void qsort (void *__base, size_t __nmemb, size_t __size,
     __compar_fn_t __compar) __attribute__ ((__nonnull__ (1, 4)));
# 771 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern int abs (int __x) __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;
extern long int labs (long int __x) __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;



__extension__ extern long long int llabs (long long int __x)
     __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;







extern div_t div (int __numer, int __denom)
     __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;
extern ldiv_t ldiv (long int __numer, long int __denom)
     __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;




__extension__ extern lldiv_t lldiv (long long int __numer,
        long long int __denom)
     __attribute__ ((__nothrow__)) __attribute__ ((__const__)) ;

# 808 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern char *ecvt (double __value, int __ndigit, int *__restrict __decpt,
     int *__restrict __sign) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4))) ;




extern char *fcvt (double __value, int __ndigit, int *__restrict __decpt,
     int *__restrict __sign) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4))) ;




extern char *gcvt (double __value, int __ndigit, char *__buf)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3))) ;




extern char *qecvt (long double __value, int __ndigit,
      int *__restrict __decpt, int *__restrict __sign)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4))) ;
extern char *qfcvt (long double __value, int __ndigit,
      int *__restrict __decpt, int *__restrict __sign)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4))) ;
extern char *qgcvt (long double __value, int __ndigit, char *__buf)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3))) ;




extern int ecvt_r (double __value, int __ndigit, int *__restrict __decpt,
     int *__restrict __sign, char *__restrict __buf,
     size_t __len) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4, 5)));
extern int fcvt_r (double __value, int __ndigit, int *__restrict __decpt,
     int *__restrict __sign, char *__restrict __buf,
     size_t __len) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4, 5)));

extern int qecvt_r (long double __value, int __ndigit,
      int *__restrict __decpt, int *__restrict __sign,
      char *__restrict __buf, size_t __len)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4, 5)));
extern int qfcvt_r (long double __value, int __ndigit,
      int *__restrict __decpt, int *__restrict __sign,
      char *__restrict __buf, size_t __len)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (3, 4, 5)));







extern int mblen (__const char *__s, size_t __n) __attribute__ ((__nothrow__)) ;


extern int mbtowc (wchar_t *__restrict __pwc,
     __const char *__restrict __s, size_t __n) __attribute__ ((__nothrow__)) ;


extern int wctomb (char *__s, wchar_t __wchar) __attribute__ ((__nothrow__)) ;



extern size_t mbstowcs (wchar_t *__restrict __pwcs,
   __const char *__restrict __s, size_t __n) __attribute__ ((__nothrow__));

extern size_t wcstombs (char *__restrict __s,
   __const wchar_t *__restrict __pwcs, size_t __n)
     __attribute__ ((__nothrow__));








extern int rpmatch (__const char *__response) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1))) ;
# 896 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern int getsubopt (char **__restrict __optionp,
        char *__const *__restrict __tokens,
        char **__restrict __valuep)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2, 3))) ;
# 948 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4
extern int getloadavg (double __loadavg[], int __nelem)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));
# 964 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdlib.h" 3 4

# 40 "../jinclude.h" 2






# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 1 3 4
# 30 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4




# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 35 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 2 3 4
# 45 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
struct _IO_FILE;



typedef struct _IO_FILE FILE;





# 65 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
typedef struct _IO_FILE __FILE;
# 75 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 1 3 4
# 32 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/_G_config.h" 1 3 4
# 15 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/_G_config.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 16 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/_G_config.h" 2 3 4




# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/wchar.h" 1 3 4
# 83 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/wchar.h" 3 4
typedef struct
{
  int __count;
  union
  {

    unsigned int __wch;



    char __wchb[4];
  } __value;
} __mbstate_t;
# 21 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/_G_config.h" 2 3 4

typedef struct
{
  __off_t __pos;
  __mbstate_t __state;
} _G_fpos_t;
typedef struct
{
  __off64_t __pos;
  __mbstate_t __state;
} _G_fpos64_t;
# 53 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/_G_config.h" 3 4
typedef int _G_int16_t __attribute__ ((__mode__ (__HI__)));
typedef int _G_int32_t __attribute__ ((__mode__ (__SI__)));
typedef unsigned int _G_uint16_t __attribute__ ((__mode__ (__HI__)));
typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
# 33 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 2 3 4
# 53 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stdarg.h" 1 3 4
# 40 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stdarg.h" 3 4
typedef __builtin_va_list __gnuc_va_list;
# 54 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 2 3 4
# 170 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
struct _IO_jump_t; struct _IO_FILE;
# 180 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
typedef void _IO_lock_t;





struct _IO_marker {
  struct _IO_marker *_next;
  struct _IO_FILE *_sbuf;



  int _pos;
# 203 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
};


enum __codecvt_result
{
  __codecvt_ok,
  __codecvt_partial,
  __codecvt_error,
  __codecvt_noconv
};
# 271 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
struct _IO_FILE {
  int _flags;




  char* _IO_read_ptr;
  char* _IO_read_end;
  char* _IO_read_base;
  char* _IO_write_base;
  char* _IO_write_ptr;
  char* _IO_write_end;
  char* _IO_buf_base;
  char* _IO_buf_end;

  char *_IO_save_base;
  char *_IO_backup_base;
  char *_IO_save_end;

  struct _IO_marker *_markers;

  struct _IO_FILE *_chain;

  int _fileno;



  int _flags2;

  __off_t _old_offset;



  unsigned short _cur_column;
  signed char _vtable_offset;
  char _shortbuf[1];



  _IO_lock_t *_lock;
# 319 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
  __off64_t _offset;
# 328 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
  void *__pad1;
  void *__pad2;
  void *__pad3;
  void *__pad4;
  size_t __pad5;

  int _mode;

  char _unused2[15 * sizeof (int) - 4 * sizeof (void *) - sizeof (size_t)];

};


typedef struct _IO_FILE _IO_FILE;


struct _IO_FILE_plus;

extern struct _IO_FILE_plus _IO_2_1_stdin_;
extern struct _IO_FILE_plus _IO_2_1_stdout_;
extern struct _IO_FILE_plus _IO_2_1_stderr_;
# 364 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
typedef __ssize_t __io_read_fn (void *__cookie, char *__buf, size_t __nbytes);







typedef __ssize_t __io_write_fn (void *__cookie, __const char *__buf,
     size_t __n);







typedef int __io_seek_fn (void *__cookie, __off64_t *__pos, int __w);


typedef int __io_close_fn (void *__cookie);
# 416 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
extern int __underflow (_IO_FILE *);
extern int __uflow (_IO_FILE *);
extern int __overflow (_IO_FILE *, int);
# 460 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
extern int _IO_getc (_IO_FILE *__fp);
extern int _IO_putc (int __c, _IO_FILE *__fp);
extern int _IO_feof (_IO_FILE *__fp) __attribute__ ((__nothrow__));
extern int _IO_ferror (_IO_FILE *__fp) __attribute__ ((__nothrow__));

extern int _IO_peekc_locked (_IO_FILE *__fp);





extern void _IO_flockfile (_IO_FILE *) __attribute__ ((__nothrow__));
extern void _IO_funlockfile (_IO_FILE *) __attribute__ ((__nothrow__));
extern int _IO_ftrylockfile (_IO_FILE *) __attribute__ ((__nothrow__));
# 490 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/libio.h" 3 4
extern int _IO_vfscanf (_IO_FILE * __restrict, const char * __restrict,
   __gnuc_va_list, int *__restrict);
extern int _IO_vfprintf (_IO_FILE *__restrict, const char *__restrict,
    __gnuc_va_list);
extern __ssize_t _IO_padn (_IO_FILE *, int, __ssize_t);
extern size_t _IO_sgetn (_IO_FILE *, void *, size_t);

extern __off64_t _IO_seekoff (_IO_FILE *, __off64_t, int, int);
extern __off64_t _IO_seekpos (_IO_FILE *, __off64_t, int);

extern void _IO_free_backup_area (_IO_FILE *) __attribute__ ((__nothrow__));
# 76 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 2 3 4




typedef __gnuc_va_list va_list;
# 109 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4


typedef _G_fpos_t fpos_t;




# 161 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/stdio_lim.h" 1 3 4
# 162 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 2 3 4



extern struct _IO_FILE *stdin;
extern struct _IO_FILE *stdout;
extern struct _IO_FILE *stderr;







extern int remove (__const char *__filename) __attribute__ ((__nothrow__));

extern int rename (__const char *__old, __const char *__new) __attribute__ ((__nothrow__));




extern int renameat (int __oldfd, __const char *__old, int __newfd,
       __const char *__new) __attribute__ ((__nothrow__));








extern FILE *tmpfile (void) ;
# 206 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern char *tmpnam (char *__s) __attribute__ ((__nothrow__)) ;





extern char *tmpnam_r (char *__s) __attribute__ ((__nothrow__)) ;
# 224 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern char *tempnam (__const char *__dir, __const char *__pfx)
     __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) ;








extern int fclose (FILE *__stream);




extern int fflush (FILE *__stream);

# 249 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int fflush_unlocked (FILE *__stream);
# 263 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4






extern FILE *fopen (__const char *__restrict __filename,
      __const char *__restrict __modes) ;




extern FILE *freopen (__const char *__restrict __filename,
        __const char *__restrict __modes,
        FILE *__restrict __stream) ;
# 292 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4

# 303 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern FILE *fdopen (int __fd, __const char *__modes) __attribute__ ((__nothrow__)) ;
# 316 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern FILE *fmemopen (void *__s, size_t __len, __const char *__modes)
  __attribute__ ((__nothrow__)) ;




extern FILE *open_memstream (char **__bufloc, size_t *__sizeloc) __attribute__ ((__nothrow__)) ;






extern void setbuf (FILE *__restrict __stream, char *__restrict __buf) __attribute__ ((__nothrow__));



extern int setvbuf (FILE *__restrict __stream, char *__restrict __buf,
      int __modes, size_t __n) __attribute__ ((__nothrow__));





extern void setbuffer (FILE *__restrict __stream, char *__restrict __buf,
         size_t __size) __attribute__ ((__nothrow__));


extern void setlinebuf (FILE *__stream) __attribute__ ((__nothrow__));








extern int fprintf (FILE *__restrict __stream,
      __const char *__restrict __format, ...);




extern int printf (__const char *__restrict __format, ...);

extern int sprintf (char *__restrict __s,
      __const char *__restrict __format, ...) __attribute__ ((__nothrow__));





extern int vfprintf (FILE *__restrict __s, __const char *__restrict __format,
       __gnuc_va_list __arg);




extern int vprintf (__const char *__restrict __format, __gnuc_va_list __arg);

extern int vsprintf (char *__restrict __s, __const char *__restrict __format,
       __gnuc_va_list __arg) __attribute__ ((__nothrow__));





extern int snprintf (char *__restrict __s, size_t __maxlen,
       __const char *__restrict __format, ...)
     __attribute__ ((__nothrow__)) __attribute__ ((__format__ (__printf__, 3, 4)));

extern int vsnprintf (char *__restrict __s, size_t __maxlen,
        __const char *__restrict __format, __gnuc_va_list __arg)
     __attribute__ ((__nothrow__)) __attribute__ ((__format__ (__printf__, 3, 0)));

# 414 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int vdprintf (int __fd, __const char *__restrict __fmt,
       __gnuc_va_list __arg)
     __attribute__ ((__format__ (__printf__, 2, 0)));
extern int dprintf (int __fd, __const char *__restrict __fmt, ...)
     __attribute__ ((__format__ (__printf__, 2, 3)));








extern int fscanf (FILE *__restrict __stream,
     __const char *__restrict __format, ...) ;




extern int scanf (__const char *__restrict __format, ...) ;

extern int sscanf (__const char *__restrict __s,
     __const char *__restrict __format, ...) __attribute__ ((__nothrow__));
# 445 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int fscanf (FILE *__restrict __stream, __const char *__restrict __format, ...) __asm__ ("" "__isoc99_fscanf")

                               ;
extern int scanf (__const char *__restrict __format, ...) __asm__ ("" "__isoc99_scanf")
                              ;
extern int sscanf (__const char *__restrict __s, __const char *__restrict __format, ...) __asm__ ("" "__isoc99_sscanf") __attribute__ ((__nothrow__))

                      ;
# 465 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4








extern int vfscanf (FILE *__restrict __s, __const char *__restrict __format,
      __gnuc_va_list __arg)
     __attribute__ ((__format__ (__scanf__, 2, 0))) ;





extern int vscanf (__const char *__restrict __format, __gnuc_va_list __arg)
     __attribute__ ((__format__ (__scanf__, 1, 0))) ;


extern int vsscanf (__const char *__restrict __s,
      __const char *__restrict __format, __gnuc_va_list __arg)
     __attribute__ ((__nothrow__)) __attribute__ ((__format__ (__scanf__, 2, 0)));
# 496 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int vfscanf (FILE *__restrict __s, __const char *__restrict __format, __gnuc_va_list __arg) __asm__ ("" "__isoc99_vfscanf")



     __attribute__ ((__format__ (__scanf__, 2, 0))) ;
extern int vscanf (__const char *__restrict __format, __gnuc_va_list __arg) __asm__ ("" "__isoc99_vscanf")

     __attribute__ ((__format__ (__scanf__, 1, 0))) ;
extern int vsscanf (__const char *__restrict __s, __const char *__restrict __format, __gnuc_va_list __arg) __asm__ ("" "__isoc99_vsscanf") __attribute__ ((__nothrow__))



     __attribute__ ((__format__ (__scanf__, 2, 0)));
# 524 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4









extern int fgetc (FILE *__stream);
extern int getc (FILE *__stream);





extern int getchar (void);

# 552 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int getc_unlocked (FILE *__stream);
extern int getchar_unlocked (void);
# 563 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int fgetc_unlocked (FILE *__stream);











extern int fputc (int __c, FILE *__stream);
extern int putc (int __c, FILE *__stream);





extern int putchar (int __c);

# 596 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int fputc_unlocked (int __c, FILE *__stream);







extern int putc_unlocked (int __c, FILE *__stream);
extern int putchar_unlocked (int __c);






extern int getw (FILE *__stream);


extern int putw (int __w, FILE *__stream);








extern char *fgets (char *__restrict __s, int __n, FILE *__restrict __stream)
     ;






extern char *gets (char *__s) ;

# 658 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern __ssize_t __getdelim (char **__restrict __lineptr,
          size_t *__restrict __n, int __delimiter,
          FILE *__restrict __stream) ;
extern __ssize_t getdelim (char **__restrict __lineptr,
        size_t *__restrict __n, int __delimiter,
        FILE *__restrict __stream) ;







extern __ssize_t getline (char **__restrict __lineptr,
       size_t *__restrict __n,
       FILE *__restrict __stream) ;








extern int fputs (__const char *__restrict __s, FILE *__restrict __stream);





extern int puts (__const char *__s);






extern int ungetc (int __c, FILE *__stream);






extern size_t fread (void *__restrict __ptr, size_t __size,
       size_t __n, FILE *__restrict __stream) ;




extern size_t fwrite (__const void *__restrict __ptr, size_t __size,
        size_t __n, FILE *__restrict __s) ;

# 730 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern size_t fread_unlocked (void *__restrict __ptr, size_t __size,
         size_t __n, FILE *__restrict __stream) ;
extern size_t fwrite_unlocked (__const void *__restrict __ptr, size_t __size,
          size_t __n, FILE *__restrict __stream) ;








extern int fseek (FILE *__stream, long int __off, int __whence);




extern long int ftell (FILE *__stream) ;




extern void rewind (FILE *__stream);

# 766 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern int fseeko (FILE *__stream, __off_t __off, int __whence);




extern __off_t ftello (FILE *__stream) ;
# 785 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4






extern int fgetpos (FILE *__restrict __stream, fpos_t *__restrict __pos);




extern int fsetpos (FILE *__stream, __const fpos_t *__pos);
# 808 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4

# 817 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4


extern void clearerr (FILE *__stream) __attribute__ ((__nothrow__));

extern int feof (FILE *__stream) __attribute__ ((__nothrow__)) ;

extern int ferror (FILE *__stream) __attribute__ ((__nothrow__)) ;




extern void clearerr_unlocked (FILE *__stream) __attribute__ ((__nothrow__));
extern int feof_unlocked (FILE *__stream) __attribute__ ((__nothrow__)) ;
extern int ferror_unlocked (FILE *__stream) __attribute__ ((__nothrow__)) ;








extern void perror (__const char *__s);






# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/sys_errlist.h" 1 3 4
# 27 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/sys_errlist.h" 3 4
extern int sys_nerr;
extern __const char *__const sys_errlist[];
# 847 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 2 3 4




extern int fileno (FILE *__stream) __attribute__ ((__nothrow__)) ;




extern int fileno_unlocked (FILE *__stream) __attribute__ ((__nothrow__)) ;
# 866 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern FILE *popen (__const char *__command, __const char *__modes) ;





extern int pclose (FILE *__stream);





extern char *ctermid (char *__s) __attribute__ ((__nothrow__));
# 906 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
extern void flockfile (FILE *__stream) __attribute__ ((__nothrow__));



extern int ftrylockfile (FILE *__stream) __attribute__ ((__nothrow__)) ;


extern void funlockfile (FILE *__stream) __attribute__ ((__nothrow__));
# 927 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/stdio.h" 1 3 4
# 36 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/stdio.h" 3 4
extern __inline int
vprintf (__const char *__restrict __fmt, __gnuc_va_list __arg)
{
  return vfprintf (stdout, __fmt, __arg);
}



extern __inline int
getchar (void)
{
  return _IO_getc (stdin);
}




extern __inline int
fgetc_unlocked (FILE *__fp)
{
  return (__builtin_expect (((__fp)->_IO_read_ptr >= (__fp)->_IO_read_end), 0) ? __uflow (__fp) : *(unsigned char *) (__fp)->_IO_read_ptr++);
}





extern __inline int
getc_unlocked (FILE *__fp)
{
  return (__builtin_expect (((__fp)->_IO_read_ptr >= (__fp)->_IO_read_end), 0) ? __uflow (__fp) : *(unsigned char *) (__fp)->_IO_read_ptr++);
}


extern __inline int
getchar_unlocked (void)
{
  return (__builtin_expect (((stdin)->_IO_read_ptr >= (stdin)->_IO_read_end), 0) ? __uflow (stdin) : *(unsigned char *) (stdin)->_IO_read_ptr++);
}




extern __inline int
putchar (int __c)
{
  return _IO_putc (__c, stdout);
}




extern __inline int
fputc_unlocked (int __c, FILE *__stream)
{
  return (__builtin_expect (((__stream)->_IO_write_ptr >= (__stream)->_IO_write_end), 0) ? __overflow (__stream, (unsigned char) (__c)) : (unsigned char) (*(__stream)->_IO_write_ptr++ = (__c)));
}





extern __inline int
putc_unlocked (int __c, FILE *__stream)
{
  return (__builtin_expect (((__stream)->_IO_write_ptr >= (__stream)->_IO_write_end), 0) ? __overflow (__stream, (unsigned char) (__c)) : (unsigned char) (*(__stream)->_IO_write_ptr++ = (__c)));
}


extern __inline int
putchar_unlocked (int __c)
{
  return (__builtin_expect (((stdout)->_IO_write_ptr >= (stdout)->_IO_write_end), 0) ? __overflow (stdout, (unsigned char) (__c)) : (unsigned char) (*(stdout)->_IO_write_ptr++ = (__c)));
}
# 125 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/stdio.h" 3 4
extern __inline int
__attribute__ ((__nothrow__)) feof_unlocked (FILE *__stream)
{
  return (((__stream)->_flags & 0x10) != 0);
}


extern __inline int
__attribute__ ((__nothrow__)) ferror_unlocked (FILE *__stream)
{
  return (((__stream)->_flags & 0x20) != 0);
}
# 928 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 2 3 4
# 936 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/stdio.h" 3 4

# 47 "../jinclude.h" 2
# 66 "../jinclude.h"
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 1 3 4
# 29 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4





# 1 "/opt/x-tools/arm-cortex_a9-linux-gnueabi/lib/gcc/arm-cortex_a9-linux-gnueabi/4.6.1/include/stddef.h" 1 3 4
# 35 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 2 3 4









extern void *memcpy (void *__restrict __dest,
       __const void *__restrict __src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern void *memmove (void *__dest, __const void *__src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));






extern void *memccpy (void *__restrict __dest, __const void *__restrict __src,
        int __c, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));





extern void *memset (void *__s, int __c, size_t __n) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));


extern int memcmp (__const void *__s1, __const void *__s2, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));
# 95 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern void *memchr (__const void *__s, int __c, size_t __n)
      __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));


# 126 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4


extern char *strcpy (char *__restrict __dest, __const char *__restrict __src)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));

extern char *strncpy (char *__restrict __dest,
        __const char *__restrict __src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern char *strcat (char *__restrict __dest, __const char *__restrict __src)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));

extern char *strncat (char *__restrict __dest, __const char *__restrict __src,
        size_t __n) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern int strcmp (__const char *__s1, __const char *__s2)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));

extern int strncmp (__const char *__s1, __const char *__s2, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));


extern int strcoll (__const char *__s1, __const char *__s2)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));

extern size_t strxfrm (char *__restrict __dest,
         __const char *__restrict __src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));






# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/xlocale.h" 1 3 4
# 28 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/xlocale.h" 3 4
typedef struct __locale_struct
{

  struct __locale_data *__locales[13];


  const unsigned short int *__ctype_b;
  const int *__ctype_tolower;
  const int *__ctype_toupper;


  const char *__names[13];
} *__locale_t;


typedef __locale_t locale_t;
# 163 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 2 3 4


extern int strcoll_l (__const char *__s1, __const char *__s2, __locale_t __l)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2, 3)));

extern size_t strxfrm_l (char *__dest, __const char *__src, size_t __n,
    __locale_t __l) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2, 4)));





extern char *strdup (__const char *__s)
     __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) __attribute__ ((__nonnull__ (1)));






extern char *strndup (__const char *__string, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__malloc__)) __attribute__ ((__nonnull__ (1)));
# 210 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4

# 235 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strchr (__const char *__s, int __c)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));
# 262 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strrchr (__const char *__s, int __c)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));


# 281 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4



extern size_t strcspn (__const char *__s, __const char *__reject)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));


extern size_t strspn (__const char *__s, __const char *__accept)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));
# 314 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strpbrk (__const char *__s, __const char *__accept)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));
# 342 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strstr (__const char *__haystack, __const char *__needle)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));




extern char *strtok (char *__restrict __s, __const char *__restrict __delim)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2)));




extern char *__strtok_r (char *__restrict __s,
    __const char *__restrict __delim,
    char **__restrict __save_ptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2, 3)));

extern char *strtok_r (char *__restrict __s, __const char *__restrict __delim,
         char **__restrict __save_ptr)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (2, 3)));
# 397 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4


extern size_t strlen (__const char *__s)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));





extern size_t strnlen (__const char *__string, size_t __maxlen)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));





extern char *strerror (int __errnum) __attribute__ ((__nothrow__));

# 427 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern int strerror_r (int __errnum, char *__buf, size_t __buflen) __asm__ ("" "__xpg_strerror_r") __attribute__ ((__nothrow__))

                        __attribute__ ((__nonnull__ (2)));
# 445 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strerror_l (int __errnum, __locale_t __l) __attribute__ ((__nothrow__));





extern void __bzero (void *__s, size_t __n) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));



extern void bcopy (__const void *__src, void *__dest, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));


extern void bzero (void *__s, size_t __n) __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1)));


extern int bcmp (__const void *__s1, __const void *__s2, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));
# 489 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *index (__const char *__s, int __c)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));
# 517 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *rindex (__const char *__s, int __c)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1)));




extern int ffs (int __i) __attribute__ ((__nothrow__)) __attribute__ ((__const__));
# 536 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern int strcasecmp (__const char *__s1, __const char *__s2)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));


extern int strncasecmp (__const char *__s1, __const char *__s2, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__pure__)) __attribute__ ((__nonnull__ (1, 2)));
# 559 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
extern char *strsep (char **__restrict __stringp,
       __const char *__restrict __delim)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));




extern char *strsignal (int __sig) __attribute__ ((__nothrow__));


extern char *__stpcpy (char *__restrict __dest, __const char *__restrict __src)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
extern char *stpcpy (char *__restrict __dest, __const char *__restrict __src)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));



extern char *__stpncpy (char *__restrict __dest,
   __const char *__restrict __src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
extern char *stpncpy (char *__restrict __dest,
        __const char *__restrict __src, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__nonnull__ (1, 2)));
# 634 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4
# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string.h" 1 3 4
# 635 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 2 3 4


# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 1 3 4
# 80 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
typedef struct { unsigned char __arr[2]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR2;
typedef struct { unsigned char __arr[3]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR3;
typedef struct { unsigned char __arr[4]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR4;
typedef struct { unsigned char __arr[5]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR5;
typedef struct { unsigned char __arr[6]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR6;
typedef struct { unsigned char __arr[7]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR7;
typedef struct { unsigned char __arr[8]; } __attribute__ ((__packed__)) __STRING2_COPY_ARR8;
# 394 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern void *__rawmemchr (const void *__s, int __c);
# 969 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern __inline size_t __strcspn_c1 (__const char *__s, int __reject);
extern __inline size_t
__strcspn_c1 (__const char *__s, int __reject)
{
  register size_t __result = 0;
  while (__s[__result] != '\0' && __s[__result] != __reject)
    ++__result;
  return __result;
}

extern __inline size_t __strcspn_c2 (__const char *__s, int __reject1,
         int __reject2);
extern __inline size_t
__strcspn_c2 (__const char *__s, int __reject1, int __reject2)
{
  register size_t __result = 0;
  while (__s[__result] != '\0' && __s[__result] != __reject1
  && __s[__result] != __reject2)
    ++__result;
  return __result;
}

extern __inline size_t __strcspn_c3 (__const char *__s, int __reject1,
         int __reject2, int __reject3);
extern __inline size_t
__strcspn_c3 (__const char *__s, int __reject1, int __reject2,
       int __reject3)
{
  register size_t __result = 0;
  while (__s[__result] != '\0' && __s[__result] != __reject1
  && __s[__result] != __reject2 && __s[__result] != __reject3)
    ++__result;
  return __result;
}
# 1045 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern __inline size_t __strspn_c1 (__const char *__s, int __accept);
extern __inline size_t
__strspn_c1 (__const char *__s, int __accept)
{
  register size_t __result = 0;

  while (__s[__result] == __accept)
    ++__result;
  return __result;
}

extern __inline size_t __strspn_c2 (__const char *__s, int __accept1,
        int __accept2);
extern __inline size_t
__strspn_c2 (__const char *__s, int __accept1, int __accept2)
{
  register size_t __result = 0;

  while (__s[__result] == __accept1 || __s[__result] == __accept2)
    ++__result;
  return __result;
}

extern __inline size_t __strspn_c3 (__const char *__s, int __accept1,
        int __accept2, int __accept3);
extern __inline size_t
__strspn_c3 (__const char *__s, int __accept1, int __accept2, int __accept3)
{
  register size_t __result = 0;

  while (__s[__result] == __accept1 || __s[__result] == __accept2
  || __s[__result] == __accept3)
    ++__result;
  return __result;
}
# 1121 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern __inline char *__strpbrk_c2 (__const char *__s, int __accept1,
         int __accept2);
extern __inline char *
__strpbrk_c2 (__const char *__s, int __accept1, int __accept2)
{

  while (*__s != '\0' && *__s != __accept1 && *__s != __accept2)
    ++__s;
  return *__s == '\0' ? ((void *)0) : (char *) (size_t) __s;
}

extern __inline char *__strpbrk_c3 (__const char *__s, int __accept1,
         int __accept2, int __accept3);
extern __inline char *
__strpbrk_c3 (__const char *__s, int __accept1, int __accept2,
       int __accept3)
{

  while (*__s != '\0' && *__s != __accept1 && *__s != __accept2
  && *__s != __accept3)
    ++__s;
  return *__s == '\0' ? ((void *)0) : (char *) (size_t) __s;
}
# 1172 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern __inline char *__strtok_r_1c (char *__s, char __sep, char **__nextp);
extern __inline char *
__strtok_r_1c (char *__s, char __sep, char **__nextp)
{
  char *__result;
  if (__s == ((void *)0))
    __s = *__nextp;
  while (*__s == __sep)
    ++__s;
  __result = ((void *)0);
  if (*__s != '\0')
    {
      __result = __s++;
      while (*__s != '\0')
 if (*__s++ == __sep)
   {
     __s[-1] = '\0';
     break;
   }
    }
  *__nextp = __s;
  return __result;
}
# 1204 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern char *__strsep_g (char **__stringp, __const char *__delim);
# 1222 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern __inline char *__strsep_1c (char **__s, char __reject);
extern __inline char *
__strsep_1c (char **__s, char __reject)
{
  register char *__retval = *__s;
  if (__retval != ((void *)0) && (*__s = (__extension__ (__builtin_constant_p (__reject) && !__builtin_constant_p (__retval) && (__reject) == '\0' ? (char *) __rawmemchr (__retval, __reject) : __builtin_strchr (__retval, __reject)))) != ((void *)0))
    *(*__s)++ = '\0';
  return __retval;
}

extern __inline char *__strsep_2c (char **__s, char __reject1, char __reject2);
extern __inline char *
__strsep_2c (char **__s, char __reject1, char __reject2)
{
  register char *__retval = *__s;
  if (__retval != ((void *)0))
    {
      register char *__cp = __retval;
      while (1)
 {
   if (*__cp == '\0')
     {
       __cp = ((void *)0);
   break;
     }
   if (*__cp == __reject1 || *__cp == __reject2)
     {
       *__cp++ = '\0';
       break;
     }
   ++__cp;
 }
      *__s = __cp;
    }
  return __retval;
}

extern __inline char *__strsep_3c (char **__s, char __reject1, char __reject2,
       char __reject3);
extern __inline char *
__strsep_3c (char **__s, char __reject1, char __reject2, char __reject3)
{
  register char *__retval = *__s;
  if (__retval != ((void *)0))
    {
      register char *__cp = __retval;
      while (1)
 {
   if (*__cp == '\0')
     {
       __cp = ((void *)0);
   break;
     }
   if (*__cp == __reject1 || *__cp == __reject2 || *__cp == __reject3)
     {
       *__cp++ = '\0';
       break;
     }
   ++__cp;
 }
      *__s = __cp;
    }
  return __retval;
}
# 1303 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern char *__strdup (__const char *__string) __attribute__ ((__nothrow__)) __attribute__ ((__malloc__));
# 1322 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/bits/string2.h" 3 4
extern char *__strndup (__const char *__string, size_t __n)
     __attribute__ ((__nothrow__)) __attribute__ ((__malloc__));
# 638 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 2 3 4
# 646 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/string.h" 3 4

# 67 "../jinclude.h" 2
# 22 "jsimd_arm.c" 2
# 1 "../jpeglib.h" 1
# 28 "../jpeglib.h"
# 1 "../jmorecfg.h" 1
# 60 "../jmorecfg.h"
typedef unsigned char JSAMPLE;
# 100 "../jmorecfg.h"
typedef short JCOEF;
# 111 "../jmorecfg.h"
typedef unsigned char JOCTET;
# 136 "../jmorecfg.h"
typedef unsigned char UINT8;
# 148 "../jmorecfg.h"
typedef unsigned short UINT16;







typedef short INT16;





typedef long INT32;
# 172 "../jmorecfg.h"
typedef unsigned int JDIMENSION;
# 228 "../jmorecfg.h"
typedef int boolean;
# 318 "../jmorecfg.h"
static const int rgb_red[12] = {
 -1, -1, 0, -1, -1, -1, 0, 0, 2, 2, 3, 1
};

static const int rgb_green[12] = {
 -1, -1, 1, -1, -1, -1, 1, 1, 1, 1, 2, 2
};

static const int rgb_blue[12] = {
 -1, -1, 2, -1, -1, -1, 2, 2, 0, 0, 1, 3
};

static const int rgb_pixelsize[12] = {
 -1, -1, 3, -1, -1, -1, 3, 4, 3, 4, 4, 4
};
# 29 "../jpeglib.h" 2
# 68 "../jpeglib.h"
typedef JSAMPLE *JSAMPROW;
typedef JSAMPROW *JSAMPARRAY;
typedef JSAMPARRAY *JSAMPIMAGE;

typedef JCOEF JBLOCK[64];
typedef JBLOCK *JBLOCKROW;
typedef JBLOCKROW *JBLOCKARRAY;
typedef JBLOCKARRAY *JBLOCKIMAGE;

typedef JCOEF *JCOEFPTR;







typedef struct {




  UINT16 quantval[64];





  boolean sent_table;
} JQUANT_TBL;




typedef struct {

  UINT8 bits[17];

  UINT8 huffval[256];





  boolean sent_table;
} JHUFF_TBL;




typedef struct {



  int component_id;
  int component_index;
  int h_samp_factor;
  int v_samp_factor;
  int quant_tbl_no;




  int dc_tbl_no;
  int ac_tbl_no;
# 141 "../jpeglib.h"
  JDIMENSION width_in_blocks;
  JDIMENSION height_in_blocks;
# 153 "../jpeglib.h"
  int DCT_scaled_size;







  JDIMENSION downsampled_width;
  JDIMENSION downsampled_height;




  boolean component_needed;



  int MCU_width;
  int MCU_height;
  int MCU_blocks;
  int MCU_sample_width;
  int last_col_width;
  int last_row_height;





  JQUANT_TBL * quant_table;


  void * dct_table;
} jpeg_component_info;




typedef struct {
  int comps_in_scan;
  int component_index[4];
  int Ss, Se;
  int Ah, Al;
} jpeg_scan_info;



typedef struct jpeg_marker_struct * jpeg_saved_marker_ptr;

struct jpeg_marker_struct {
  jpeg_saved_marker_ptr next;
  UINT8 marker;
  unsigned int original_length;
  unsigned int data_length;
  JOCTET * data;

};





typedef enum {
 JCS_UNKNOWN,
 JCS_GRAYSCALE,
 JCS_RGB,

 JCS_YCbCr,
 JCS_CMYK,
 JCS_YCCK,
 JCS_EXT_RGB,
 JCS_EXT_RGBX,
 JCS_EXT_BGR,
 JCS_EXT_BGRX,
 JCS_EXT_XBGR,
 JCS_EXT_XRGB
} J_COLOR_SPACE;



typedef enum {
 JDCT_ISLOW,
 JDCT_IFAST,
 JDCT_FLOAT,
 JDCT_IFAST2
} J_DCT_METHOD;
# 249 "../jpeglib.h"
typedef enum {
 JDITHER_NONE,
 JDITHER_ORDERED,
 JDITHER_FS
} J_DITHER_MODE;
# 270 "../jpeglib.h"
struct jpeg_common_struct {
  struct jpeg_error_mgr * err; struct jpeg_memory_mgr * mem; struct jpeg_progress_mgr * progress; void * client_data; boolean is_decompressor; int global_state;




};

typedef struct jpeg_common_struct * j_common_ptr;
typedef struct jpeg_compress_struct * j_compress_ptr;
typedef struct jpeg_decompress_struct * j_decompress_ptr;




struct jpeg_compress_struct {
  struct jpeg_error_mgr * err; struct jpeg_memory_mgr * mem; struct jpeg_progress_mgr * progress; void * client_data; boolean is_decompressor; int global_state;


  struct jpeg_destination_mgr * dest;






  JDIMENSION image_width;
  JDIMENSION image_height;
  int input_components;
  J_COLOR_SPACE in_color_space;

  double input_gamma;
# 324 "../jpeglib.h"
  int data_precision;

  int num_components;
  J_COLOR_SPACE jpeg_color_space;

  jpeg_component_info * comp_info;


  JQUANT_TBL * quant_tbl_ptrs[4];







  JHUFF_TBL * dc_huff_tbl_ptrs[4];
  JHUFF_TBL * ac_huff_tbl_ptrs[4];


  UINT8 arith_dc_L[16];
  UINT8 arith_dc_U[16];
  UINT8 arith_ac_K[16];

  int num_scans;
  const jpeg_scan_info * scan_info;





  boolean raw_data_in;
  boolean arith_code;
  boolean optimize_coding;
  boolean CCIR601_sampling;



  int smoothing_factor;
  J_DCT_METHOD dct_method;






  unsigned int restart_interval;
  int restart_in_rows;



  boolean write_JFIF_header;
  UINT8 JFIF_major_version;
  UINT8 JFIF_minor_version;




  UINT8 density_unit;
  UINT16 X_density;
  UINT16 Y_density;
  boolean write_Adobe_marker;






  JDIMENSION next_scanline;
# 401 "../jpeglib.h"
  boolean progressive_mode;
  int max_h_samp_factor;
  int max_v_samp_factor;






  JDIMENSION total_iMCU_rows;
# 421 "../jpeglib.h"
  int comps_in_scan;
  jpeg_component_info * cur_comp_info[4];


  JDIMENSION MCUs_per_row;
  JDIMENSION MCU_rows_in_scan;

  int blocks_in_MCU;
  int MCU_membership[10];



  int Ss, Se, Ah, Al;
# 444 "../jpeglib.h"
  struct jpeg_comp_master * master;
  struct jpeg_c_main_controller * main;
  struct jpeg_c_prep_controller * prep;
  struct jpeg_c_coef_controller * coef;
  struct jpeg_marker_writer * marker;
  struct jpeg_color_converter * cconvert;
  struct jpeg_downsampler * downsample;
  struct jpeg_forward_dct * fdct;
  struct jpeg_entropy_encoder * entropy;
  jpeg_scan_info * script_space;
  int script_space_size;
};




struct jpeg_decompress_struct {
  struct jpeg_error_mgr * err; struct jpeg_memory_mgr * mem; struct jpeg_progress_mgr * progress; void * client_data; boolean is_decompressor; int global_state;


  struct jpeg_source_mgr * src;




  JDIMENSION image_width;
  JDIMENSION image_height;
  int num_components;
  J_COLOR_SPACE jpeg_color_space;






  J_COLOR_SPACE out_color_space;

  unsigned int scale_num, scale_denom;

  double output_gamma;

  boolean buffered_image;
  boolean raw_data_out;

  J_DCT_METHOD dct_method;
  boolean do_fancy_upsampling;
  boolean do_block_smoothing;

  boolean quantize_colors;

  J_DITHER_MODE dither_mode;
  boolean two_pass_quantize;
  int desired_number_of_colors;

  boolean enable_1pass_quant;
  boolean enable_external_quant;
  boolean enable_2pass_quant;







  JDIMENSION output_width;
  JDIMENSION output_height;
  int out_color_components;
  int output_components;



  int rec_outbuf_height;
# 527 "../jpeglib.h"
  int actual_number_of_colors;
  JSAMPARRAY colormap;
# 538 "../jpeglib.h"
  JDIMENSION output_scanline;




  int input_scan_number;
  JDIMENSION input_iMCU_row;





  int output_scan_number;
  JDIMENSION output_iMCU_row;
# 560 "../jpeglib.h"
  int (*coef_bits)[64];
# 571 "../jpeglib.h"
  JQUANT_TBL * quant_tbl_ptrs[4];


  JHUFF_TBL * dc_huff_tbl_ptrs[4];
  JHUFF_TBL * ac_huff_tbl_ptrs[4];






  int data_precision;

  jpeg_component_info * comp_info;





  boolean progressive_mode;
  boolean arith_code;

  UINT8 arith_dc_L[16];
  UINT8 arith_dc_U[16];
  UINT8 arith_ac_K[16];

  unsigned int restart_interval;




  boolean saw_JFIF_marker;

  UINT8 JFIF_major_version;
  UINT8 JFIF_minor_version;
  UINT8 density_unit;
  UINT16 X_density;
  UINT16 Y_density;
  boolean saw_Adobe_marker;
  UINT8 Adobe_transform;

  boolean CCIR601_sampling;





  jpeg_saved_marker_ptr marker_list;
# 627 "../jpeglib.h"
  int max_h_samp_factor;
  int max_v_samp_factor;





  int min_DCT_scaled_size;


  JDIMENSION total_iMCU_rows;
# 646 "../jpeglib.h"
  JSAMPLE * sample_range_limit;






  int comps_in_scan;
  jpeg_component_info * cur_comp_info[4];


  JDIMENSION MCUs_per_row;
  JDIMENSION MCU_rows_in_scan;

  int blocks_in_MCU;
  int MCU_membership[10];



  int Ss, Se, Ah, Al;
# 679 "../jpeglib.h"
  int unread_marker;




  struct jpeg_decomp_master * master;
  struct jpeg_d_main_controller * main;
  struct jpeg_d_coef_controller * coef;
  struct jpeg_d_post_controller * post;
  struct jpeg_input_controller * inputctl;
  struct jpeg_marker_reader * marker;
  struct jpeg_entropy_decoder * entropy;
  struct jpeg_inverse_dct * idct;
  struct jpeg_upsampler * upsample;
  struct jpeg_color_deconverter * cconvert;
  struct jpeg_color_quantizer * cquantize;
};
# 708 "../jpeglib.h"
struct jpeg_error_mgr {

  void (*error_exit) (j_common_ptr cinfo);

  void (*emit_message) (j_common_ptr cinfo, int msg_level);

  void (*output_message) (j_common_ptr cinfo);

  void (*format_message) (j_common_ptr cinfo, char * buffer);


  void (*reset_error_mgr) (j_common_ptr cinfo);




  int msg_code;

  union {
    int i[8];
    char s[80];
  } msg_parm;



  int trace_level;







  long num_warnings;
# 753 "../jpeglib.h"
  const char * const * jpeg_message_table;
  int last_jpeg_message;



  const char * const * addon_message_table;
  int first_addon_message;
  int last_addon_message;
};




struct jpeg_progress_mgr {
  void (*progress_monitor) (j_common_ptr cinfo);

  long pass_counter;
  long pass_limit;
  int completed_passes;
  int total_passes;
};




struct jpeg_destination_mgr {
  JOCTET * next_output_byte;
  size_t free_in_buffer;

  void (*init_destination) (j_compress_ptr cinfo);
  boolean (*empty_output_buffer) (j_compress_ptr cinfo);
  void (*term_destination) (j_compress_ptr cinfo);
};




struct jpeg_source_mgr {
  const JOCTET * next_input_byte;
  size_t bytes_in_buffer;

  void (*init_source) (j_decompress_ptr cinfo);
  boolean (*fill_input_buffer) (j_decompress_ptr cinfo);
  void (*skip_input_data) (j_decompress_ptr cinfo, long num_bytes);
  boolean (*resync_to_restart) (j_decompress_ptr cinfo, int desired);
  void (*term_source) (j_decompress_ptr cinfo);
};
# 817 "../jpeglib.h"
typedef struct jvirt_sarray_control * jvirt_sarray_ptr;
typedef struct jvirt_barray_control * jvirt_barray_ptr;


struct jpeg_memory_mgr {

  void * (*alloc_small) (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
                         ;
  void * (*alloc_large) (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
                              ;
  JSAMPARRAY (*alloc_sarray) (j_common_ptr cinfo, int pool_id, JDIMENSION samplesperrow, JDIMENSION numrows)

                             ;
  JBLOCKARRAY (*alloc_barray) (j_common_ptr cinfo, int pool_id, JDIMENSION blocksperrow, JDIMENSION numrows)

                              ;
  jvirt_sarray_ptr (*request_virt_sarray) (j_common_ptr cinfo, int pool_id, boolean pre_zero, JDIMENSION samplesperrow, JDIMENSION numrows, JDIMENSION maxaccess)




                              ;
  jvirt_barray_ptr (*request_virt_barray) (j_common_ptr cinfo, int pool_id, boolean pre_zero, JDIMENSION blocksperrow, JDIMENSION numrows, JDIMENSION maxaccess)




                              ;
  void (*realize_virt_arrays) (j_common_ptr cinfo);
  JSAMPARRAY (*access_virt_sarray) (j_common_ptr cinfo, jvirt_sarray_ptr ptr, JDIMENSION start_row, JDIMENSION num_rows, boolean writable)



                          ;
  JBLOCKARRAY (*access_virt_barray) (j_common_ptr cinfo, jvirt_barray_ptr ptr, JDIMENSION start_row, JDIMENSION num_rows, boolean writable)



                           ;
  void (*free_pool) (j_common_ptr cinfo, int pool_id);
  void (*self_destruct) (j_common_ptr cinfo);






  long max_memory_to_use;


  long max_alloc_chunk;
};





typedef boolean (*jpeg_marker_parser_method) (j_decompress_ptr cinfo);
# 962 "../jpeglib.h"
extern struct jpeg_error_mgr * jpeg_std_error
 (struct jpeg_error_mgr * err);
# 978 "../jpeglib.h"
extern void jpeg_CreateCompress (j_compress_ptr cinfo, int version, size_t structsize)
                                          ;
extern void jpeg_CreateDecompress (j_decompress_ptr cinfo, int version, size_t structsize)
                                     ;

extern void jpeg_destroy_compress (j_compress_ptr cinfo);
extern void jpeg_destroy_decompress (j_decompress_ptr cinfo);



extern void jpeg_stdio_dest (j_compress_ptr cinfo, FILE * outfile);
extern void jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile);
# 1002 "../jpeglib.h"
extern void jpeg_set_defaults (j_compress_ptr cinfo);

extern void jpeg_set_colorspace (j_compress_ptr cinfo, J_COLOR_SPACE colorspace)
                                    ;
extern void jpeg_default_colorspace (j_compress_ptr cinfo);
extern void jpeg_set_quality (j_compress_ptr cinfo, int quality, boolean force_baseline)
                               ;
extern void jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor, boolean force_baseline)

                               ;




extern void jpeg_add_quant_table (j_compress_ptr cinfo, int which_tbl, const unsigned int *basic_table, int scale_factor, boolean force_baseline)


                                   ;
extern int jpeg_quality_scaling (int quality);
extern void jpeg_simple_progression (j_compress_ptr cinfo);
extern void jpeg_suppress_tables (j_compress_ptr cinfo, boolean suppress)
                             ;
extern JQUANT_TBL * jpeg_alloc_quant_table (j_common_ptr cinfo);
extern JHUFF_TBL * jpeg_alloc_huff_table (j_common_ptr cinfo);


extern void jpeg_start_compress (j_compress_ptr cinfo, boolean write_all_tables)
                                    ;
extern JDIMENSION jpeg_write_scanlines (j_compress_ptr cinfo, JSAMPARRAY scanlines, JDIMENSION num_lines)

                                ;
extern void jpeg_finish_compress (j_compress_ptr cinfo);







extern JDIMENSION jpeg_write_raw_data (j_compress_ptr cinfo, JSAMPIMAGE data, JDIMENSION num_lines)

                               ;


extern void jpeg_write_marker
 (j_compress_ptr cinfo, int marker, const JOCTET * dataptr, unsigned int datalen)
                                                    ;

extern void jpeg_write_m_header
 (j_compress_ptr cinfo, int marker, unsigned int datalen);
extern void jpeg_write_m_byte
 (j_compress_ptr cinfo, int val);


extern void jpeg_write_tables (j_compress_ptr cinfo);


extern int jpeg_read_header (j_decompress_ptr cinfo, boolean require_image)
                             ;
# 1072 "../jpeglib.h"
extern boolean jpeg_start_decompress (j_decompress_ptr cinfo);
extern JDIMENSION jpeg_read_scanlines (j_decompress_ptr cinfo, JSAMPARRAY scanlines, JDIMENSION max_lines)

                               ;
extern boolean jpeg_finish_decompress (j_decompress_ptr cinfo);


extern JDIMENSION jpeg_read_raw_data (j_decompress_ptr cinfo, JSAMPIMAGE data, JDIMENSION max_lines)

                              ;


extern boolean jpeg_has_multiple_scans (j_decompress_ptr cinfo);
extern boolean jpeg_start_output (j_decompress_ptr cinfo, int scan_number)
                            ;
extern boolean jpeg_finish_output (j_decompress_ptr cinfo);
extern boolean jpeg_input_complete (j_decompress_ptr cinfo);
extern void jpeg_new_colormap (j_decompress_ptr cinfo);
extern int jpeg_consume_input (j_decompress_ptr cinfo);
# 1102 "../jpeglib.h"
extern void jpeg_calc_output_dimensions (j_decompress_ptr cinfo);


extern void jpeg_save_markers
 (j_decompress_ptr cinfo, int marker_code, unsigned int length_limit)
                                 ;


extern void jpeg_set_marker_processor
 (j_decompress_ptr cinfo, int marker_code, jpeg_marker_parser_method routine)
                                         ;


extern jvirt_barray_ptr * jpeg_read_coefficients (j_decompress_ptr cinfo);
extern void jpeg_write_coefficients (j_compress_ptr cinfo, jvirt_barray_ptr * coef_arrays)
                                       ;
extern void jpeg_copy_critical_parameters (j_decompress_ptr srcinfo, j_compress_ptr dstinfo)
                              ;







extern void jpeg_abort_compress (j_compress_ptr cinfo);
extern void jpeg_abort_decompress (j_decompress_ptr cinfo);




extern void jpeg_abort (j_common_ptr cinfo);
extern void jpeg_destroy (j_common_ptr cinfo);


extern boolean jpeg_resync_to_restart (j_decompress_ptr cinfo, int desired)
                      ;
# 1192 "../jpeglib.h"
# 1 "../jpegint.h" 1
# 17 "../jpegint.h"
typedef enum {
 JBUF_PASS_THRU,

 JBUF_SAVE_SOURCE,
 JBUF_CRANK_DEST,
 JBUF_SAVE_AND_PASS
} J_BUF_MODE;
# 46 "../jpegint.h"
struct jpeg_comp_master {
  void (*prepare_for_pass) (j_compress_ptr cinfo);
  void (*pass_startup) (j_compress_ptr cinfo);
  void (*finish_pass) (j_compress_ptr cinfo);


  boolean call_pass_startup;
  boolean is_last_pass;
};


struct jpeg_c_main_controller {
  void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode);
  void (*process_data) (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail)

                                    ;
};


struct jpeg_c_prep_controller {
  void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode);
  void (*pre_process_data) (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail, JSAMPIMAGE output_buf, JDIMENSION *out_row_group_ctr, JDIMENSION out_row_groups_avail)





                                        ;
};


struct jpeg_c_coef_controller {
  void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode);
  boolean (*compress_data) (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
                             ;
};


struct jpeg_color_converter {
  void (*start_pass) (j_compress_ptr cinfo);
  void (*color_convert) (j_compress_ptr cinfo, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                         ;
};


struct jpeg_downsampler {
  void (*start_pass) (j_compress_ptr cinfo);
  void (*downsample) (j_compress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION in_row_index, JSAMPIMAGE output_buf, JDIMENSION out_row_group_index)


                                        ;

  boolean need_context_rows;
};


struct jpeg_forward_dct {
  void (*start_pass) (j_compress_ptr cinfo);

  void (*forward_DCT) (j_compress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY sample_data, JBLOCKROW coef_blocks, JDIMENSION start_row, JDIMENSION start_col, JDIMENSION num_blocks)



                                ;
};


struct jpeg_entropy_encoder {
  void (*start_pass) (j_compress_ptr cinfo, boolean gather_statistics);
  boolean (*encode_mcu) (j_compress_ptr cinfo, JBLOCKROW *MCU_data);
  void (*finish_pass) (j_compress_ptr cinfo);
};


struct jpeg_marker_writer {
  void (*write_file_header) (j_compress_ptr cinfo);
  void (*write_frame_header) (j_compress_ptr cinfo);
  void (*write_scan_header) (j_compress_ptr cinfo);
  void (*write_file_trailer) (j_compress_ptr cinfo);
  void (*write_tables_only) (j_compress_ptr cinfo);


  void (*write_marker_header) (j_compress_ptr cinfo, int marker, unsigned int datalen)
                                ;
  void (*write_marker_byte) (j_compress_ptr cinfo, int val);
};





struct jpeg_decomp_master {
  void (*prepare_for_output_pass) (j_decompress_ptr cinfo);
  void (*finish_output_pass) (j_decompress_ptr cinfo);


  boolean is_dummy_pass;
};


struct jpeg_input_controller {
  int (*consume_input) (j_decompress_ptr cinfo);
  void (*reset_input_controller) (j_decompress_ptr cinfo);
  void (*start_input_pass) (j_decompress_ptr cinfo);
  void (*finish_input_pass) (j_decompress_ptr cinfo);


  boolean has_multiple_scans;
  boolean eoi_reached;
};


struct jpeg_d_main_controller {
  void (*start_pass) (j_decompress_ptr cinfo, J_BUF_MODE pass_mode);
  void (*process_data) (j_decompress_ptr cinfo, JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)

                                     ;
};


struct jpeg_d_coef_controller {
  void (*start_input_pass) (j_decompress_ptr cinfo);
  int (*consume_data) (j_decompress_ptr cinfo);
  void (*start_output_pass) (j_decompress_ptr cinfo);
  int (*decompress_data) (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
                            ;

  jvirt_barray_ptr *coef_arrays;
};


struct jpeg_d_post_controller {
  void (*start_pass) (j_decompress_ptr cinfo, J_BUF_MODE pass_mode);
  void (*post_process_data) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr, JDIMENSION in_row_groups_avail, JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)





                                   ;
};


struct jpeg_marker_reader {
  void (*reset_marker_reader) (j_decompress_ptr cinfo);




  int (*read_markers) (j_decompress_ptr cinfo);

  jpeg_marker_parser_method read_restart_marker;




  boolean saw_SOI;
  boolean saw_SOF;
  int next_restart_num;
  unsigned int discarded_bytes;
};


struct jpeg_entropy_decoder {
  void (*start_pass) (j_decompress_ptr cinfo);
  boolean (*decode_mcu) (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
                         ;



  boolean insufficient_data;
};


typedef void (*inverse_DCT_method_ptr) (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                 ;

struct jpeg_inverse_dct {
  void (*start_pass) (j_decompress_ptr cinfo);

  inverse_DCT_method_ptr inverse_DCT[10];
};


struct jpeg_upsampler {
  void (*start_pass) (j_decompress_ptr cinfo);
  void (*upsample) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr, JDIMENSION in_row_groups_avail, JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)





                                 ;

  boolean need_context_rows;
};


struct jpeg_color_deconverter {
  void (*start_pass) (j_decompress_ptr cinfo);
  void (*color_convert) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                         ;
};


struct jpeg_color_quantizer {
  void (*start_pass) (j_decompress_ptr cinfo, boolean is_pre_scan);
  void (*color_quantize) (j_decompress_ptr cinfo, JSAMPARRAY input_buf, JSAMPARRAY output_buf, int num_rows)

                   ;
  void (*finish_pass) (j_decompress_ptr cinfo);
  void (*new_color_map) (j_decompress_ptr cinfo);
};
# 338 "../jpegint.h"
extern void jinit_compress_master (j_compress_ptr cinfo);
extern void jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
                              ;
extern void jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_c_prep_controller (j_compress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_color_converter (j_compress_ptr cinfo);
extern void jinit_downsampler (j_compress_ptr cinfo);
extern void jinit_forward_dct (j_compress_ptr cinfo);
extern void jinit_huff_encoder (j_compress_ptr cinfo);
extern void jinit_phuff_encoder (j_compress_ptr cinfo);
extern void jinit_arith_encoder (j_compress_ptr cinfo);
extern void jinit_marker_writer (j_compress_ptr cinfo);

extern void jinit_master_decompress (j_decompress_ptr cinfo);
extern void jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_d_post_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
                                 ;
extern void jinit_input_controller (j_decompress_ptr cinfo);
extern void jinit_marker_reader (j_decompress_ptr cinfo);
extern void jinit_huff_decoder (j_decompress_ptr cinfo);
extern void jinit_phuff_decoder (j_decompress_ptr cinfo);
extern void jinit_arith_decoder (j_decompress_ptr cinfo);
extern void jinit_inverse_dct (j_decompress_ptr cinfo);
extern void jinit_upsampler (j_decompress_ptr cinfo);
extern void jinit_color_deconverter (j_decompress_ptr cinfo);
extern void jinit_1pass_quantizer (j_decompress_ptr cinfo);
extern void jinit_2pass_quantizer (j_decompress_ptr cinfo);
extern void jinit_merged_upsampler (j_decompress_ptr cinfo);

extern void jinit_memory_mgr (j_common_ptr cinfo);


extern long jdiv_round_up (long a, long b);
extern long jround_up (long a, long b);
extern void jcopy_sample_rows (JSAMPARRAY input_array, int source_row, JSAMPARRAY output_array, int dest_row, int num_rows, JDIMENSION num_cols)

                                           ;
extern void jcopy_block_row (JBLOCKROW input_row, JBLOCKROW output_row, JDIMENSION num_blocks)
                             ;
extern void jzero_far (void * target, size_t bytestozero);




extern const int jpeg_natural_order[];


extern const INT32 jpeg_aritab[];
# 1193 "../jpeglib.h" 2
# 1 "../jerror.h" 1
# 34 "../jerror.h"
typedef enum {





JMSG_NOMESSAGE ,



JERR_ARITH_NOTIMPL ,


JERR_BAD_ALIGN_TYPE ,
JERR_BAD_ALLOC_CHUNK ,
JERR_BAD_BUFFER_MODE ,
JERR_BAD_COMPONENT_ID ,



JERR_BAD_DCT_COEF ,
JERR_BAD_DCTSIZE ,




JERR_BAD_HUFF_TABLE ,
JERR_BAD_IN_COLORSPACE ,
JERR_BAD_J_COLORSPACE ,
JERR_BAD_LENGTH ,
JERR_BAD_LIB_VERSION ,

JERR_BAD_MCU_SIZE ,
JERR_BAD_POOL_ID ,
JERR_BAD_PRECISION ,
JERR_BAD_PROGRESSION ,

JERR_BAD_PROG_SCRIPT ,

JERR_BAD_SAMPLING ,
JERR_BAD_SCAN_SCRIPT ,
JERR_BAD_STATE ,
JERR_BAD_STRUCT_SIZE ,

JERR_BAD_VIRTUAL_ACCESS ,
JERR_BUFFER_SIZE ,
JERR_CANT_SUSPEND ,
JERR_CCIR601_NOTIMPL ,
JERR_COMPONENT_COUNT ,
JERR_CONVERSION_NOTIMPL ,
JERR_DAC_INDEX ,
JERR_DAC_VALUE ,
JERR_DHT_INDEX ,
JERR_DQT_INDEX ,
JERR_EMPTY_IMAGE ,
JERR_EMS_READ ,
JERR_EMS_WRITE ,
JERR_EOI_EXPECTED ,
JERR_FILE_READ ,
JERR_FILE_WRITE ,
JERR_FRACT_SAMPLE_NOTIMPL ,
JERR_HUFF_CLEN_OVERFLOW ,
JERR_HUFF_MISSING_CODE ,
JERR_IMAGE_TOO_BIG ,
JERR_INPUT_EMPTY ,
JERR_INPUT_EOF ,
JERR_MISMATCHED_QUANT_TABLE ,

JERR_MISSING_DATA ,
JERR_MODE_CHANGE ,
JERR_NOTIMPL ,
JERR_NOT_COMPILED ,



JERR_NO_BACKING_STORE ,
JERR_NO_HUFF_TABLE ,
JERR_NO_IMAGE ,
JERR_NO_QUANT_TABLE ,
JERR_NO_SOI ,
JERR_OUT_OF_MEMORY ,
JERR_QUANT_COMPONENTS ,

JERR_QUANT_FEW_COLORS ,
JERR_QUANT_MANY_COLORS ,
JERR_SOF_DUPLICATE ,
JERR_SOF_NO_SOS ,
JERR_SOF_UNSUPPORTED ,
JERR_SOI_DUPLICATE ,
JERR_SOS_NO_SOF ,
JERR_TFILE_CREATE ,
JERR_TFILE_READ ,
JERR_TFILE_SEEK ,
JERR_TFILE_WRITE ,

JERR_TOO_LITTLE_DATA ,
JERR_UNKNOWN_MARKER ,
JERR_VIRTUAL_BUG ,
JERR_WIDTH_OVERFLOW ,
JERR_XMS_READ ,
JERR_XMS_WRITE ,
JMSG_COPYRIGHT ,
JMSG_VERSION ,
JTRC_16BIT_TABLES ,

JTRC_ADOBE ,

JTRC_APP0 ,
JTRC_APP14 ,
JTRC_DAC ,
JTRC_DHT ,
JTRC_DQT ,
JTRC_DRI ,
JTRC_EMS_CLOSE ,
JTRC_EMS_OPEN ,
JTRC_EOI ,
JTRC_HUFFBITS ,
JTRC_JFIF ,
JTRC_JFIF_BADTHUMBNAILSIZE ,

JTRC_JFIF_EXTENSION ,

JTRC_JFIF_THUMBNAIL ,
JTRC_MISC_MARKER ,
JTRC_PARMLESS_MARKER ,
JTRC_QUANTVALS ,
JTRC_QUANT_3_NCOLORS ,
JTRC_QUANT_NCOLORS ,
JTRC_QUANT_SELECTED ,
JTRC_RECOVERY_ACTION ,
JTRC_RST ,
JTRC_SMOOTH_NOTIMPL ,

JTRC_SOF ,
JTRC_SOF_COMPONENT ,
JTRC_SOI ,
JTRC_SOS ,
JTRC_SOS_COMPONENT ,
JTRC_SOS_PARAMS ,
JTRC_TFILE_CLOSE ,
JTRC_TFILE_OPEN ,
JTRC_THUMB_JPEG ,

JTRC_THUMB_PALETTE ,

JTRC_THUMB_RGB ,

JTRC_UNKNOWN_IDS ,

JTRC_XMS_CLOSE ,
JTRC_XMS_OPEN ,
JWRN_ADOBE_XFORM ,



JWRN_BOGUS_PROGRESSION ,

JWRN_EXTRANEOUS_DATA ,

JWRN_HIT_MARKER ,
JWRN_HUFF_BAD_CODE ,
JWRN_JFIF_MAJOR ,
JWRN_JPEG_EOF ,
JWRN_MUST_RESYNC ,

JWRN_NOT_SEQUENTIAL ,
JWRN_TOO_MUCH_DATA ,

JERR_BAD_CROP_SPEC ,

JERR_NO_ARITH_TABLE ,
JWRN_ARITH_BAD_CODE ,





  JMSG_LASTMSGCODE
} J_MESSAGE_CODE;
# 1194 "../jpeglib.h" 2
# 23 "jsimd_arm.c" 2
# 1 "../jsimd.h" 1
# 40 "../jsimd.h"
extern int jsimd_can_rgb_ycc (void);
extern int jsimd_can_rgb_gray (void);
extern int jsimd_can_ycc_rgb (void);

extern void jsimd_rgb_ycc_convert
        (j_compress_ptr cinfo, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_rgb_gray_convert
        (j_compress_ptr cinfo, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_ycc_rgb_convert
        (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;

extern int jsimd_can_h2v2_downsample (void);
extern int jsimd_can_h2v1_downsample (void);

extern void jsimd_h2v2_downsample
        (j_compress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY output_data)
                                                            ;
extern void jsimd_h2v1_downsample
        (j_compress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY output_data)
                                                            ;

extern int jsimd_can_h2v2_upsample (void);
extern int jsimd_can_h2v1_upsample (void);

extern void jsimd_h2v2_upsample
        (j_decompress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_upsample
        (j_decompress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern int jsimd_can_h2v2_fancy_upsample (void);
extern int jsimd_can_h2v1_fancy_upsample (void);

extern void jsimd_h2v2_fancy_upsample
        (j_decompress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_fancy_upsample
        (j_decompress_ptr cinfo, jpeg_component_info * compptr, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern int jsimd_can_h2v2_merged_upsample (void);
extern int jsimd_can_h2v1_merged_upsample (void);

extern void jsimd_h2v2_merged_upsample
        (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)

                                    ;
extern void jsimd_h2v1_merged_upsample
        (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)

                                    ;
# 24 "jsimd_arm.c" 2
# 1 "../jdct.h" 1
# 37 "../jdct.h"
typedef short DCTELEM;
typedef unsigned short UDCTELEM;
typedef unsigned int UDCTELEM2;
# 65 "../jdct.h"
typedef short ISLOW_MULT_TYPE;

typedef short IFAST_MULT_TYPE;







typedef float FLOAT_MULT_TYPE;
# 108 "../jdct.h"
extern void jpeg_fdct_islow (DCTELEM * data);
extern void jpeg_fdct_ifast (DCTELEM * data);
extern void jpeg_fdct_float (float * data);

extern void jpeg_idct_islow
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
extern void jpeg_idct_ifast
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
extern void jpeg_idct_float
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
extern void jpeg_idct_4x4
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
extern void jpeg_idct_2x2
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
extern void jpeg_idct_1x1
    (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)
                                                                     ;
# 25 "jsimd_arm.c" 2
# 1 "../jsimddct.h" 1
# 41 "../jsimddct.h"
extern int jsimd_can_convsamp (void);
extern int jsimd_can_convsamp_float (void);

extern void jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace)

                                                      ;
extern void jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col, float * workspace)

                                                               ;

extern int jsimd_can_fdct_islow (void);
extern int jsimd_can_fdct_ifast (void);
extern int jsimd_can_fdct_float (void);

extern void jsimd_fdct_islow (DCTELEM * data);
extern void jsimd_fdct_ifast (DCTELEM * data);
extern void jsimd_fdct_float (float * data);

extern int jsimd_can_quantize (void);
extern int jsimd_can_quantize_float (void);

extern void jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace)

                                                      ;
extern void jsimd_quantize_float (JCOEFPTR coef_block, float * divisors, float * workspace)

                                                               ;

extern int jsimd_can_idct_2x2 (void);
extern int jsimd_can_idct_4x4 (void);

extern void jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)



                                                        ;
extern void jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)



                                                        ;

extern int jsimd_can_idct_islow (void);
extern int jsimd_can_idct_ifast (void);
extern int jsimd_can_idct_float (void);

extern void jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)



                                                          ;
extern void jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)



                                                          ;
extern void jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)



                                                          ;
# 26 "jsimd_arm.c" 2
# 1 "jsimd.h" 1
# 152 "jsimd.h"
extern unsigned int jpeg_simd_cpu_support (void);


extern void jsimd_rgb_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgb_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgbx_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgr_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgrx_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxbgr_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxrgb_ycc_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;

extern void jsimd_rgb_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgb_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgbx_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgr_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgrx_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxbgr_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxrgb_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;

extern void jsimd_ycc_rgb_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgb_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgbx_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgr_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgrx_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxbgr_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxrgb_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;

extern const int jconst_rgb_ycc_convert_sse2[];
extern void jsimd_rgb_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgb_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgbx_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgr_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgrx_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxbgr_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxrgb_ycc_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;

extern const int jconst_rgb_gray_convert_sse2[];
extern void jsimd_rgb_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgb_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extrgbx_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgr_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extbgrx_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxbgr_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;
extern void jsimd_extxrgb_gray_convert_sse2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)

                                                  ;

extern const int jconst_ycc_rgb_convert_sse2[];
extern void jsimd_ycc_rgb_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgb_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgbx_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgr_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgrx_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxbgr_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxrgb_convert_sse2
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;

extern void jsimd_ycc_rgb_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgb_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extrgbx_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgr_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extbgrx_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxbgr_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;
extern void jsimd_ycc_extxrgb_convert_neon
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, JSAMPARRAY output_buf, int num_rows)

                                                  ;


extern void jsimd_h2v2_downsample_mmx
        (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, JDIMENSION width_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data)

                                                            ;
extern void jsimd_h2v1_downsample_mmx
        (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, JDIMENSION width_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data)

                                                            ;

extern void jsimd_h2v2_downsample_sse2
        (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, JDIMENSION width_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data)

                                                            ;
extern void jsimd_h2v1_downsample_sse2
        (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, JDIMENSION width_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data)

                                                            ;


extern void jsimd_h2v2_upsample_mmx
        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_upsample_mmx
        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern void jsimd_h2v2_fancy_upsample_mmx
        (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_fancy_upsample_mmx
        (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern void jsimd_h2v2_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extrgb_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extrgbx_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extbgr_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extbgrx_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extxbgr_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extxrgb_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extrgb_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extrgbx_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extbgr_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extbgrx_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extxbgr_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extxrgb_merged_upsample_mmx
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;

extern void jsimd_h2v2_upsample_sse2
        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_upsample_sse2
        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern const int jconst_fancy_upsample_sse2[];
extern void jsimd_h2v2_fancy_upsample_sse2
        (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;
extern void jsimd_h2v1_fancy_upsample_sse2
        (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
                                                                  ;

extern const int jconst_merged_upsample_sse2[];
extern void jsimd_h2v2_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extrgb_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extrgbx_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extbgr_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extbgrx_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extxbgr_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v2_extxrgb_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extrgb_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extrgbx_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extbgr_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extbgrx_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extxbgr_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;
extern void jsimd_h2v1_extxrgb_merged_upsample_sse2
        (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
                                                                 ;


extern void jsimd_convsamp_mmx (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace)

                                                          ;

extern void jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace)

                                                           ;

extern void jsimd_convsamp_float_3dnow (JSAMPARRAY sample_data, JDIMENSION start_col, float * workspace)

                                                                     ;

extern void jsimd_convsamp_float_sse (JSAMPARRAY sample_data, JDIMENSION start_col, float * workspace)

                                                                   ;

extern void jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, float * workspace)

                                                                    ;


extern void jsimd_fdct_islow_mmx (DCTELEM * data);
extern void jsimd_fdct_ifast_mmx (DCTELEM * data);

extern const int jconst_fdct_ifast_sse2[];
extern void jsimd_fdct_islow_sse2 (DCTELEM * data);
extern const int jconst_fdct_islow_sse2[];
extern void jsimd_fdct_ifast_sse2 (DCTELEM * data);

extern void jsimd_fdct_float_3dnow (float * data);

extern const int jconst_fdct_float_sse[];
extern void jsimd_fdct_float_sse (float * data);


extern void jsimd_quantize_mmx (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace)

                                                          ;

extern void jsimd_quantize_sse2 (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace)

                                                           ;

extern void jsimd_quantize_float_3dnow (JCOEFPTR coef_block, float * divisors, float * workspace)

                                                                     ;

extern void jsimd_quantize_float_sse (JCOEFPTR coef_block, float * divisors, float * workspace)

                                                                   ;

extern void jsimd_quantize_float_sse2 (JCOEFPTR coef_block, float * divisors, float * workspace)

                                                                    ;


extern void jsimd_idct_2x2_mmx (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                            ;
extern void jsimd_idct_4x4_mmx (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                            ;

extern const int jconst_idct_red_sse2[];
extern void jsimd_idct_2x2_sse2 (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                             ;
extern void jsimd_idct_4x4_sse2 (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                             ;

extern void jsimd_idct_2x2_neon (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                             ;
extern void jsimd_idct_4x4_neon (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                             ;


extern void jsimd_idct_islow_mmx (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                              ;
extern void jsimd_idct_ifast_mmx (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                              ;

extern const int jconst_idct_islow_sse2[];
extern void jsimd_idct_islow_sse2 (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                               ;
extern const int jconst_idct_ifast_sse2[];
extern void jsimd_idct_ifast_sse2 (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                               ;

extern void jsimd_idct_ifast_neon (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                               ;

extern void jsimd_idct_float_3dnow (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                                ;

extern const int jconst_idct_float_sse[];
extern void jsimd_idct_float_sse (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                              ;

extern const int jconst_idct_float_sse2[];
extern void jsimd_idct_float_sse2 (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)


                                                               ;
# 27 "jsimd_arm.c" 2



# 1 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 1 3 4
# 30 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4

# 48 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4
enum
{
  _ISupper = ((0) < 8 ? ((1 << (0)) << 8) : ((1 << (0)) >> 8)),
  _ISlower = ((1) < 8 ? ((1 << (1)) << 8) : ((1 << (1)) >> 8)),
  _ISalpha = ((2) < 8 ? ((1 << (2)) << 8) : ((1 << (2)) >> 8)),
  _ISdigit = ((3) < 8 ? ((1 << (3)) << 8) : ((1 << (3)) >> 8)),
  _ISxdigit = ((4) < 8 ? ((1 << (4)) << 8) : ((1 << (4)) >> 8)),
  _ISspace = ((5) < 8 ? ((1 << (5)) << 8) : ((1 << (5)) >> 8)),
  _ISprint = ((6) < 8 ? ((1 << (6)) << 8) : ((1 << (6)) >> 8)),
  _ISgraph = ((7) < 8 ? ((1 << (7)) << 8) : ((1 << (7)) >> 8)),
  _ISblank = ((8) < 8 ? ((1 << (8)) << 8) : ((1 << (8)) >> 8)),
  _IScntrl = ((9) < 8 ? ((1 << (9)) << 8) : ((1 << (9)) >> 8)),
  _ISpunct = ((10) < 8 ? ((1 << (10)) << 8) : ((1 << (10)) >> 8)),
  _ISalnum = ((11) < 8 ? ((1 << (11)) << 8) : ((1 << (11)) >> 8))
};
# 81 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4
extern __const unsigned short int **__ctype_b_loc (void)
     __attribute__ ((__nothrow__)) __attribute__ ((__const));
extern __const __int32_t **__ctype_tolower_loc (void)
     __attribute__ ((__nothrow__)) __attribute__ ((__const));
extern __const __int32_t **__ctype_toupper_loc (void)
     __attribute__ ((__nothrow__)) __attribute__ ((__const));
# 96 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4






extern int isalnum (int) __attribute__ ((__nothrow__));
extern int isalpha (int) __attribute__ ((__nothrow__));
extern int iscntrl (int) __attribute__ ((__nothrow__));
extern int isdigit (int) __attribute__ ((__nothrow__));
extern int islower (int) __attribute__ ((__nothrow__));
extern int isgraph (int) __attribute__ ((__nothrow__));
extern int isprint (int) __attribute__ ((__nothrow__));
extern int ispunct (int) __attribute__ ((__nothrow__));
extern int isspace (int) __attribute__ ((__nothrow__));
extern int isupper (int) __attribute__ ((__nothrow__));
extern int isxdigit (int) __attribute__ ((__nothrow__));



extern int tolower (int __c) __attribute__ ((__nothrow__));


extern int toupper (int __c) __attribute__ ((__nothrow__));








extern int isblank (int) __attribute__ ((__nothrow__));


# 142 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4
extern int isascii (int __c) __attribute__ ((__nothrow__));



extern int toascii (int __c) __attribute__ ((__nothrow__));



extern int _toupper (int) __attribute__ ((__nothrow__));
extern int _tolower (int) __attribute__ ((__nothrow__));
# 190 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4
extern __inline int
__attribute__ ((__nothrow__)) tolower (int __c)
{
  return __c >= -128 && __c < 256 ? (*__ctype_tolower_loc ())[__c] : __c;
}

extern __inline int
__attribute__ ((__nothrow__)) toupper (int __c)
{
  return __c >= -128 && __c < 256 ? (*__ctype_toupper_loc ())[__c] : __c;
}
# 247 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4
extern int isalnum_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isalpha_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int iscntrl_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isdigit_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int islower_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isgraph_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isprint_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int ispunct_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isspace_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isupper_l (int, __locale_t) __attribute__ ((__nothrow__));
extern int isxdigit_l (int, __locale_t) __attribute__ ((__nothrow__));

extern int isblank_l (int, __locale_t) __attribute__ ((__nothrow__));



extern int __tolower_l (int __c, __locale_t __l) __attribute__ ((__nothrow__));
extern int tolower_l (int __c, __locale_t __l) __attribute__ ((__nothrow__));


extern int __toupper_l (int __c, __locale_t __l) __attribute__ ((__nothrow__));
extern int toupper_l (int __c, __locale_t __l) __attribute__ ((__nothrow__));
# 323 "/home/build/CTBU/MonoLake-foreign/opensource-core/build-trees/root/usr/include/ctype.h" 3 4

# 31 "jsimd_arm.c" 2
# 44 "jsimd_arm.c"
typedef struct {
  int16x8x4_t half[2];
} int16_8x8_t;

inline static int16_8x8_t load_transpose_8x8(const int16x8x4_t* cp) __attribute__ ((__always_inline__));

inline static int16_8x8_t load_transpose_8x8(const int16x8x4_t* cp)
{
  int16_8x8_t coef;

  {
    int16x8x2_t unzipped;
    const int16x8x4_t* coefp = cp;

    coef.half[0] = vld4q_s16((const int16_t*) (coefp++));
    coef.half[1] = vld4q_s16((const int16_t*) (coefp++));

    unzipped = vuzpq_s16(coef.half[0].val[0], coef.half[1].val[0]);
    coef.half[0].val[0] = unzipped.val[0];
    coef.half[1].val[0] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[1], coef.half[1].val[1]);
    coef.half[0].val[1] = unzipped.val[0];
    coef.half[1].val[1] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[2], coef.half[1].val[2]);
    coef.half[0].val[2] = unzipped.val[0];
    coef.half[1].val[2] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[3], coef.half[1].val[3]);
    coef.half[0].val[3] = unzipped.val[0];
    coef.half[1].val[3] = unzipped.val[1];
# 108 "jsimd_arm.c"
  }

  return coef;
}

inline static int16_8x8_t dq_transpose_8x8(const int16x8x4_t* cp, const int16x8x4_t* dp) __attribute__ ((__always_inline__));

inline static int16_8x8_t dq_transpose_8x8(const int16x8x4_t* cp, const int16x8x4_t* dp)
{
  int16_8x8_t coef;

  {
    int16_8x8_t dct;
    int16x8x2_t unzipped;
    const int16x8x4_t* coefp = cp;
    const int16x8x4_t* dctp = dp;

    coef.half[0] = vld4q_s16((const int16_t*) (coefp++));
    dct.half[0] = vld4q_s16((const int16_t*) (dctp++));
    coef.half[1] = vld4q_s16((const int16_t*) (coefp++));
    dct.half[1] = vld4q_s16((const int16_t*) (dctp++));

    coef.half[0].val[0] = vmulq_s16(coef.half[0].val[0], dct.half[0].val[0]);
    coef.half[0].val[1] = vmulq_s16(coef.half[0].val[1], dct.half[0].val[1]);
    coef.half[0].val[2] = vmulq_s16(coef.half[0].val[2], dct.half[0].val[2]);
    coef.half[0].val[3] = vmulq_s16(coef.half[0].val[3], dct.half[0].val[3]);

    coef.half[1].val[0] = vmulq_s16(coef.half[1].val[0], dct.half[1].val[0]);
    coef.half[1].val[1] = vmulq_s16(coef.half[1].val[1], dct.half[1].val[1]);
    coef.half[1].val[2] = vmulq_s16(coef.half[1].val[2], dct.half[1].val[2]);
    coef.half[1].val[3] = vmulq_s16(coef.half[1].val[3], dct.half[1].val[3]);

    unzipped = vuzpq_s16(coef.half[0].val[0], coef.half[1].val[0]);
    coef.half[0].val[0] = unzipped.val[0];
    coef.half[1].val[0] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[1], coef.half[1].val[1]);
    coef.half[0].val[1] = unzipped.val[0];
    coef.half[1].val[1] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[2], coef.half[1].val[2]);
    coef.half[0].val[2] = unzipped.val[0];
    coef.half[1].val[2] = unzipped.val[1];
    unzipped = vuzpq_s16(coef.half[0].val[3], coef.half[1].val[3]);
    coef.half[0].val[3] = unzipped.val[0];
    coef.half[1].val[3] = unzipped.val[1];
# 203 "jsimd_arm.c"
  }

  return coef;
}

inline static int16_8x8_t fdct_helper_8x8(int16x4_t fdct_constants, int16_8x8_t a) __attribute__ ((__always_inline__));

inline static int16_8x8_t fdct_helper_8x8(int16x4_t fdct_constants, int16_8x8_t a)
{
  int16x8x4_t even_part, odd_part;
  int16_8x8_t b;
# 235 "jsimd_arm.c"
  even_part.val[0] = vaddq_s16(a.half[0].val[0], a.half[1].val[3]);
  even_part.val[1] = vaddq_s16(a.half[0].val[1], a.half[1].val[2]);
  even_part.val[2] = vaddq_s16(a.half[0].val[2], a.half[1].val[1]);
  even_part.val[3] = vaddq_s16(a.half[0].val[3], a.half[1].val[0]);
  odd_part.val[0] = vsubq_s16(a.half[0].val[0], a.half[1].val[3]);
  odd_part.val[1] = vsubq_s16(a.half[0].val[1], a.half[1].val[2]);
  odd_part.val[2] = vsubq_s16(a.half[0].val[2], a.half[1].val[1]);
  odd_part.val[3] = vsubq_s16(a.half[0].val[3], a.half[1].val[0]);

  {
    int16x8_t diff0 = vsubq_s16(even_part.val[0], even_part.val[3]);
    int16x8_t diff1 = vsubq_s16(even_part.val[1], even_part.val[2]);
    int16x8_t sum0 = vaddq_s16(even_part.val[0], even_part.val[3]);
    int16x8_t sum1 = vaddq_s16(even_part.val[1], even_part.val[2]);

    int16x8_t even_a = vaddq_s16(diff0, diff1);
    int16x8_t even_b = vqrdmulhq_lane_s16(even_a, fdct_constants, 0);
    even_b = vaddq_s16(even_a, even_b);
    even_a = vshlq_n_s16(diff0, 1);

    b.half[0].val[0] = vshlq_n_s16(vaddq_s16(sum0, sum1), 1);
    b.half[1].val[0] = vshlq_n_s16(vsubq_s16(sum0, sum1), 1);
    b.half[0].val[2] = vaddq_s16(even_a, even_b);
    b.half[1].val[2] = vsubq_s16(even_a, even_b);
  }

  {
    int16x8_t sum12 = vaddq_s16(odd_part.val[1], odd_part.val[2]);
    int16x8_t odd_d = vqrdmulhq_lane_s16(sum12, fdct_constants, 0);
    odd_d = vaddq_s16(odd_d, sum12);
    int16x8_t t1 = vshlq_n_s16(odd_part.val[0], 1);
    int16x8_t odd_x = vaddq_s16(t1, odd_d);
    int16x8_t odd_y = vsubq_s16(t1, odd_d);

    int16x8_t sum01 = vaddq_s16(odd_part.val[0], odd_part.val[1]);
    int16x8_t sum23 = vaddq_s16(odd_part.val[2], odd_part.val[3]);

    t1 = vsubq_s16(sum23, sum01);
    int16x8_t odd_b = vqrdmulhq_lane_s16(t1, fdct_constants, 1);

    int16x8_t odd_a = vqrdmulhq_lane_s16(sum01, fdct_constants, 3);
    odd_a = vaddq_s16(odd_a, sum01);
    odd_a = vaddq_s16(odd_a, sum01);
    odd_a = vaddq_s16(odd_a, odd_b);

    int16x8_t odd_c = vqrdmulhq_lane_s16(sum23, fdct_constants, 2);
    odd_c = vaddq_s16(odd_c, sum23);
    odd_c = vaddq_s16(odd_c, odd_b);

    b.half[0].val[1] = vaddq_s16(odd_x, odd_a);
    b.half[1].val[3] = vsubq_s16(odd_x, odd_a);
    b.half[1].val[1] = vaddq_s16(odd_y, odd_c);
    b.half[0].val[3] = vsubq_s16(odd_y, odd_c);
  }
# 307 "jsimd_arm.c"
  return b;
}

inline static int16_8x8_t idct_helper_8x8(int16x4_t idct_constants, int16_8x8_t a) __attribute__ ((__always_inline__));

inline static int16_8x8_t idct_helper_8x8(int16x4_t idct_constants, int16_8x8_t a)
{
  int16x8x4_t even_part, odd_part;
# 336 "jsimd_arm.c"
  {
    int16x8_t sum0 = vaddq_s16(a.half[0].val[0], a.half[1].val[0]);
    int16x8_t sum2 = vaddq_s16(a.half[0].val[2], a.half[1].val[2]);
    int16x8_t diff0 = vsubq_s16(a.half[0].val[0], a.half[1].val[0]);
    int16x8_t diff2 = vsubq_s16(a.half[0].val[2], a.half[1].val[2]);

    int16x8_t even_a = vqrdmulhq_lane_s16(diff2, idct_constants, 0);
    even_a = vaddq_s16(even_a, diff2);
    even_a = vsubq_s16(even_a, sum2);

    even_part.val[0] = vaddq_s16(sum0, sum2);
    even_part.val[1] = vaddq_s16(diff0, even_a);
    even_part.val[2] = vsubq_s16(diff0, even_a);
    even_part.val[3] = vsubq_s16(sum0, sum2);
  }

  {
    int16x8_t sum1 = vaddq_s16(a.half[0].val[1], a.half[1].val[3]);
    int16x8_t sum3 = vaddq_s16(a.half[1].val[1], a.half[0].val[3]);
    int16x8_t diff1 = vsubq_s16(a.half[0].val[1], a.half[1].val[3]);
    int16x8_t diff3 = vsubq_s16(a.half[1].val[1], a.half[0].val[3]);

    int16x8_t t1 = vsubq_s16(sum1, sum3);
    int16x8_t odd_b = vqrdmulhq_lane_s16(t1, idct_constants, 0);
    odd_b = vaddq_s16(odd_b, t1);

    t1 = vaddq_s16(diff1, diff3);
    int16x8_t odd_d = vqrdmulhq_lane_s16(t1, idct_constants, 1);
    odd_d = vaddq_s16(odd_d, t1);

    int16x8_t odd_a = vqrdmulhq_lane_s16(diff3, idct_constants, 3);
    odd_a = vaddq_s16(odd_a, diff3);
    odd_a = vaddq_s16(odd_a, diff3);
    odd_a = vsubq_s16(odd_d, odd_a);

    int16x8_t odd_c = vqrdmulhq_lane_s16(diff1, idct_constants, 2);
    odd_c = vaddq_s16(odd_c, diff1);
    odd_c = vsubq_s16(odd_d, odd_c);

    odd_part.val[0] = vaddq_s16(sum1, sum3);
    odd_part.val[1] = vsubq_s16(odd_a, odd_part.val[0]);
    odd_part.val[2] = vsubq_s16(odd_b, odd_part.val[1]);
    odd_part.val[3] = vsubq_s16(odd_c, odd_part.val[2]);
  }

  int16_8x8_t b;
  b.half[0].val[0] = vaddq_s16(even_part.val[0], odd_part.val[0]);
  b.half[1].val[3] = vsubq_s16(even_part.val[0], odd_part.val[0]);
  b.half[0].val[1] = vaddq_s16(even_part.val[1], odd_part.val[1]);
  b.half[1].val[2] = vsubq_s16(even_part.val[1], odd_part.val[1]);
  b.half[0].val[2] = vaddq_s16(even_part.val[2], odd_part.val[2]);
  b.half[1].val[1] = vsubq_s16(even_part.val[2], odd_part.val[2]);
  b.half[0].val[3] = vaddq_s16(even_part.val[3], odd_part.val[3]);
  b.half[1].val[0] = vsubq_s16(even_part.val[3], odd_part.val[3]);
# 408 "jsimd_arm.c"
  return b;
}

inline static int16x8x2_t vswpq_cheat(uint32x4_t a, uint32x4_t b) __attribute__ ((__always_inline__));

inline static int16x8x2_t vswpq_cheat(uint32x4_t a, uint32x4_t b)
{
  int16x8x2_t r;
  asm ( "vswp %f0, %e1 \n\t" : "+w" (a), "+w" (b) : );
  r.val[0] = vreinterpretq_s16_u32(a);
  r.val[1] = vreinterpretq_s16_u32(b);
  return r;
}

inline static int16_8x8_t transpose_8x8(int16_8x8_t a) __attribute__ ((__always_inline__));

inline static int16_8x8_t transpose_8x8(int16_8x8_t a)
{
# 446 "jsimd_arm.c"
  int16x8x2_t t0 = vtrnq_s16(a.half[0].val[0], a.half[0].val[1]);
  int16x8x2_t t1 = vtrnq_s16(a.half[0].val[2], a.half[0].val[3]);
  int16x8x2_t t2 = vtrnq_s16(a.half[1].val[0], a.half[1].val[1]);
  int16x8x2_t t3 = vtrnq_s16(a.half[1].val[2], a.half[1].val[3]);

  uint32x4x2_t u0 = vtrnq_u32(vreinterpretq_u32_s16(t0.val[0]), vreinterpretq_u32_s16(t1.val[0]));
  uint32x4x2_t u1 = vtrnq_u32(vreinterpretq_u32_s16(t0.val[1]), vreinterpretq_u32_s16(t1.val[1]));
  uint32x4x2_t u2 = vtrnq_u32(vreinterpretq_u32_s16(t2.val[0]), vreinterpretq_u32_s16(t3.val[0]));
  uint32x4x2_t u3 = vtrnq_u32(vreinterpretq_u32_s16(t2.val[1]), vreinterpretq_u32_s16(t3.val[1]));

  int16_8x8_t b;
  int16x8x2_t r;
  r = vswpq_cheat(u0.val[0], u2.val[0]);
  b.half[0].val[0] = r.val[0];
  b.half[1].val[0] = r.val[1];
  r = vswpq_cheat(u1.val[0], u3.val[0]);
  b.half[0].val[1] = r.val[0];
  b.half[1].val[1] = r.val[1];
  r = vswpq_cheat(u0.val[1], u2.val[1]);
  b.half[0].val[2] = r.val[0];
  b.half[1].val[2] = r.val[1];
  r = vswpq_cheat(u1.val[1], u3.val[1]);
  b.half[0].val[3] = r.val[0];
  b.half[1].val[3] = r.val[1];
# 488 "jsimd_arm.c"
  return b;
}

static unsigned int simd_support = ~0;





static int
check_feature (char *buffer, char *feature)
{
  char *p;
  if (*feature == 0)
    return 0;
  if ((__extension__ (__builtin_constant_p (8) && ((__builtin_constant_p (buffer) && strlen (buffer) < ((size_t) (8))) || (__builtin_constant_p ("Features") && strlen ("Features") < ((size_t) (8)))) ? __extension__ ({ size_t __s1_len, __s2_len; (__builtin_constant_p (buffer) && __builtin_constant_p ("Features") && (__s1_len = __builtin_strlen (buffer), __s2_len = __builtin_strlen ("Features"), (!((size_t)(const void *)((buffer) + 1) - (size_t)(const void *)(buffer) == 1) || __s1_len >= 4) && (!((size_t)(const void *)(("Features") + 1) - (size_t)(const void *)("Features") == 1) || __s2_len >= 4)) ? __builtin_strcmp (buffer, "Features") : (__builtin_constant_p (buffer) && ((size_t)(const void *)((buffer) + 1) - (size_t)(const void *)(buffer) == 1) && (__s1_len = __builtin_strlen (buffer), __s1_len < 4) ? (__builtin_constant_p ("Features") && ((size_t)(const void *)(("Features") + 1) - (size_t)(const void *)("Features") == 1) ? __builtin_strcmp (buffer, "Features") : (__extension__ ({ __const unsigned char *__s2 = (__const unsigned char *) (__const char *) ("Features"); register int __result = (((__const unsigned char *) (__const char *) (buffer))[0] - __s2[0]); if (__s1_len > 0 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (buffer))[1] - __s2[1]); if (__s1_len > 1 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (buffer))[2] - __s2[2]); if (__s1_len > 2 && __result == 0) __result = (((__const unsigned char *) (__const char *) (buffer))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p ("Features") && ((size_t)(const void *)(("Features") + 1) - (size_t)(const void *)("Features") == 1) && (__s2_len = __builtin_strlen ("Features"), __s2_len < 4) ? (__builtin_constant_p (buffer) && ((size_t)(const void *)((buffer) + 1) - (size_t)(const void *)(buffer) == 1) ? __builtin_strcmp (buffer, "Features") : (__extension__ ({ __const unsigned char *__s1 = (__const unsigned char *) (__const char *) (buffer); register int __result = __s1[0] - ((__const unsigned char *) (__const char *) ("Features"))[0]; if (__s2_len > 0 && __result == 0) { __result = (__s1[1] - ((__const unsigned char *) (__const char *) ("Features"))[1]); if (__s2_len > 1 && __result == 0) { __result = (__s1[2] - ((__const unsigned char *) (__const char *) ("Features"))[2]); if (__s2_len > 2 && __result == 0) __result = (__s1[3] - ((__const unsigned char *) (__const char *) ("Features"))[3]); } } __result; }))) : __builtin_strcmp (buffer, "Features")))); }) : strncmp (buffer, "Features", 8))) != 0)
    return 0;
  buffer += 8;
  while (((*__ctype_b_loc ())[(int) ((*buffer))] & (unsigned short int) _ISspace))
    buffer++;


  while ((p = strstr(buffer, feature))) {
    if (p > buffer && !((*__ctype_b_loc ())[(int) ((*(p - 1)))] & (unsigned short int) _ISspace)) {
      buffer++;
      continue;
    }
    p += strlen(feature);
    if (*p != 0 && !((*__ctype_b_loc ())[(int) ((*p))] & (unsigned short int) _ISspace)) {
      buffer++;
      continue;
    }
    return 1;
  }
  return 0;
}

static int
parse_proc_cpuinfo (int bufsize)
{
  char *buffer = (char *)malloc(bufsize);
  FILE *fd;
  simd_support = 0;

  if (!buffer)
    return 0;

  fd = fopen("/proc/cpuinfo", "r");
  if (fd) {
    while (fgets(buffer, bufsize, fd)) {
      if (!(__extension__ (__builtin_constant_p ('\n') && !__builtin_constant_p (buffer) && ('\n') == '\0' ? (char *) __rawmemchr (buffer, '\n') : __builtin_strchr (buffer, '\n'))) && !feof(fd)) {

        fclose(fd);
        free(buffer);
        return 0;
      }
      if (check_feature(buffer, "neon"))
        simd_support |= 0x10;
    }
    fclose(fd);
  }
  free(buffer);
  return 1;
}
# 560 "jsimd_arm.c"
static void
init_simd (void)
{
  char *env = ((void *)0);




  if (simd_support != ~0)
    return;

  simd_support = 0;


  simd_support |= 0x10;
# 587 "jsimd_arm.c"
  env = getenv("JSIMD_FORCE_ARM_NEON");
  if ((env != ((void *)0)) && (__extension__ ({ size_t __s1_len, __s2_len; (__builtin_constant_p (env) && __builtin_constant_p ("1") && (__s1_len = __builtin_strlen (env), __s2_len = __builtin_strlen ("1"), (!((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) || __s1_len >= 4) && (!((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) || __s2_len >= 4)) ? __builtin_strcmp (env, "1") : (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) && (__s1_len = __builtin_strlen (env), __s1_len < 4) ? (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s2 = (__const unsigned char *) (__const char *) ("1"); register int __result = (((__const unsigned char *) (__const char *) (env))[0] - __s2[0]); if (__s1_len > 0 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[1] - __s2[1]); if (__s1_len > 1 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[2] - __s2[2]); if (__s1_len > 2 && __result == 0) __result = (((__const unsigned char *) (__const char *) (env))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) && (__s2_len = __builtin_strlen ("1"), __s2_len < 4) ? (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s1 = (__const unsigned char *) (__const char *) (env); register int __result = __s1[0] - ((__const unsigned char *) (__const char *) ("1"))[0]; if (__s2_len > 0 && __result == 0) { __result = (__s1[1] - ((__const unsigned char *) (__const char *) ("1"))[1]); if (__s2_len > 1 && __result == 0) { __result = (__s1[2] - ((__const unsigned char *) (__const char *) ("1"))[2]); if (__s2_len > 2 && __result == 0) __result = (__s1[3] - ((__const unsigned char *) (__const char *) ("1"))[3]); } } __result; }))) : __builtin_strcmp (env, "1")))); }) == 0))
    simd_support |= 0x10;
  env = getenv("JSIMD_FORCE_ARM_NEON_ASM");
  if ((env != ((void *)0)) && (__extension__ ({ size_t __s1_len, __s2_len; (__builtin_constant_p (env) && __builtin_constant_p ("1") && (__s1_len = __builtin_strlen (env), __s2_len = __builtin_strlen ("1"), (!((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) || __s1_len >= 4) && (!((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) || __s2_len >= 4)) ? __builtin_strcmp (env, "1") : (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) && (__s1_len = __builtin_strlen (env), __s1_len < 4) ? (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s2 = (__const unsigned char *) (__const char *) ("1"); register int __result = (((__const unsigned char *) (__const char *) (env))[0] - __s2[0]); if (__s1_len > 0 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[1] - __s2[1]); if (__s1_len > 1 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[2] - __s2[2]); if (__s1_len > 2 && __result == 0) __result = (((__const unsigned char *) (__const char *) (env))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) && (__s2_len = __builtin_strlen ("1"), __s2_len < 4) ? (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s1 = (__const unsigned char *) (__const char *) (env); register int __result = __s1[0] - ((__const unsigned char *) (__const char *) ("1"))[0]; if (__s2_len > 0 && __result == 0) { __result = (__s1[1] - ((__const unsigned char *) (__const char *) ("1"))[1]); if (__s2_len > 1 && __result == 0) { __result = (__s1[2] - ((__const unsigned char *) (__const char *) ("1"))[2]); if (__s2_len > 2 && __result == 0) __result = (__s1[3] - ((__const unsigned char *) (__const char *) ("1"))[3]); } } __result; }))) : __builtin_strcmp (env, "1")))); }) == 0))
    simd_support |= 0x20;
  env = getenv("JSIMD_FORCE_NO_SIMD");
  if ((env != ((void *)0)) && (__extension__ ({ size_t __s1_len, __s2_len; (__builtin_constant_p (env) && __builtin_constant_p ("1") && (__s1_len = __builtin_strlen (env), __s2_len = __builtin_strlen ("1"), (!((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) || __s1_len >= 4) && (!((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) || __s2_len >= 4)) ? __builtin_strcmp (env, "1") : (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) && (__s1_len = __builtin_strlen (env), __s1_len < 4) ? (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s2 = (__const unsigned char *) (__const char *) ("1"); register int __result = (((__const unsigned char *) (__const char *) (env))[0] - __s2[0]); if (__s1_len > 0 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[1] - __s2[1]); if (__s1_len > 1 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (env))[2] - __s2[2]); if (__s1_len > 2 && __result == 0) __result = (((__const unsigned char *) (__const char *) (env))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p ("1") && ((size_t)(const void *)(("1") + 1) - (size_t)(const void *)("1") == 1) && (__s2_len = __builtin_strlen ("1"), __s2_len < 4) ? (__builtin_constant_p (env) && ((size_t)(const void *)((env) + 1) - (size_t)(const void *)(env) == 1) ? __builtin_strcmp (env, "1") : (__extension__ ({ __const unsigned char *__s1 = (__const unsigned char *) (__const char *) (env); register int __result = __s1[0] - ((__const unsigned char *) (__const char *) ("1"))[0]; if (__s2_len > 0 && __result == 0) { __result = (__s1[1] - ((__const unsigned char *) (__const char *) ("1"))[1]); if (__s2_len > 1 && __result == 0) { __result = (__s1[2] - ((__const unsigned char *) (__const char *) ("1"))[2]); if (__s2_len > 2 && __result == 0) __result = (__s1[3] - ((__const unsigned char *) (__const char *) ("1"))[3]); } } __result; }))) : __builtin_strcmp (env, "1")))); }) == 0))
    simd_support = 0;
}

int
jsimd_can_rgb_ycc (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_rgb_gray (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_ycc_rgb (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if ((3 != 3) && (3 != 4))
    return 0;
  if (simd_support & 0x10)
    return 1;

  return 0;
}

void
jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                       JDIMENSION output_row, int num_rows)
{
}

void
jsimd_rgb_gray_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                        JDIMENSION output_row, int num_rows)
{
}

inline static uint8x8x3_t vswp_rgb(uint8x8x3_t a, int flags) __attribute__ ((__always_inline__));

inline static uint8x8x3_t vswp_rgb(uint8x8x3_t a, int flags)
{
  asm ( "teq %2, #1\n\tit eq \n\tvswpeq %P0, %P1 \n\t" : "+w" (a.val[0]), "+w" (a.val[2]) : "r" (flags) );
  return a;
}

inline static uint8x8x3_t ycbcr2rgb(uint8x8x3_t ycbcr, int16x4_t ycc_constants, uint8x8_t v128) __attribute__ ((__always_inline__));

inline static uint8x8x3_t ycbcr2rgb(uint8x8x3_t ycbcr, int16x4_t ycc_constants, uint8x8_t v128)
{
  uint8x8x3_t rgb;
  int16x8_t cb16 = vshll_n_s8(vreinterpret_s8_u8(vadd_u8(ycbcr.val[1], v128)), 7);
  int16x8_t cr16 = vshll_n_s8(vreinterpret_s8_u8(vadd_u8(ycbcr.val[2], v128)), 7);
  int16x8_t y16 = vshll_n_s8(vreinterpret_s8_u8(vadd_u8(ycbcr.val[0], v128)), 6);
  int16x8_t cg16 = vaddq_s16(cr16, vqrdmulhq_lane_s16(cb16, ycc_constants, 0));
  int16x8_t b16 = vaddq_s16(y16, vqdmulhq_lane_s16(cb16, ycc_constants, 2));
  int16x8_t r16 = vaddq_s16(y16, vqdmulhq_lane_s16(cr16, ycc_constants, 3));
  int16x8_t g16 = vsubq_s16(y16, vqdmulhq_lane_s16(cg16, ycc_constants, 1));
  rgb.val[2] = vadd_u8(vreinterpret_u8_s8(vqrshrn_n_s16(b16, 6)), v128);
  rgb.val[0] = vadd_u8(vreinterpret_u8_s8(vqrshrn_n_s16(r16, 6)), v128);
  rgb.val[1] = vadd_u8(vreinterpret_u8_s8(vqrshrn_n_s16(g16, 6)), v128);
  return rgb;
}

inline static void store_rgb(uint8_t* outptr, uint8x8x3_t rgb, int flags, uint8x8_t alpha) __attribute__ ((__always_inline__));

inline static void store_rgb(uint8_t* outptr, uint8x8x3_t rgb, int flags, uint8x8_t alpha)
{
  rgb = vswp_rgb(rgb, flags);
  if (flags & 2) {
    vst3_u8(outptr, rgb);
    return;
  }
  if (flags & 4) {
    uint8x8x4_t rgba;
    rgba.val[0] = alpha;
    rgba.val[1] = rgb.val[0];
    rgba.val[2] = rgb.val[1];
    rgba.val[3] = rgb.val[2];
    vst4_u8(outptr, rgba);
  } else {
    uint8x8x4_t rgba;
    rgba.val[0] = rgb.val[0];
    rgba.val[1] = rgb.val[1];
    rgba.val[2] = rgb.val[2];
    rgba.val[3] = alpha;
    vst4_u8(outptr, rgba);
  }
}
# 710 "jsimd_arm.c"
void
jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
                       JSAMPIMAGE input_buf, JDIMENSION input_row,
                       JSAMPARRAY output_buf, int num_rows)
{
  if (!(simd_support & 0x10))
    return;

  if (!(simd_support & 0x20))
  {
    static int16_t ycc_constants[4] __attribute__ ((aligned (8)))
      = { 15791, 11700, 29032, 22970 };
    int16x4_t c = vld1_s16(ycc_constants);
    uint8x8_t v128 = vdup_n_u8(128);
    uint8x8_t alpha = vdup_n_u8(255);
    JDIMENSION num_cols = cinfo->output_width;
    int rgbstride = rgb_pixelsize[cinfo->out_color_space];
    int rindex = rgb_red[cinfo->out_color_space];
    int bindex = rgb_blue[cinfo->out_color_space];
    int flags = 0;
    if (rgb_red[cinfo->out_color_space] > rgb_blue[cinfo->out_color_space])
      flags |= 1;
    if (rgbstride == 3)
      flags |= 2;
    if (rindex && bindex)
      flags |= 4;

    int i, j;
    for (i = 0; i < num_rows; ++i)
    {
      JSAMPROW yrow = input_buf[0][input_row + i];
      JSAMPROW cbrow = input_buf[1][input_row + i];
      JSAMPROW crrow = input_buf[2][input_row + i];
      uint8_t* outptr = (uint8_t*) output_buf[i];

      int trailing_cols = num_cols % 8;
      for (j = 0; j < num_cols - trailing_cols; j += 8)
      {
        uint8x8x3_t ycbcr;
        ycbcr.val[0] = vld1_u8((uint8_t*) &yrow[j]);
        ycbcr.val[1] = vld1_u8((uint8_t*) &cbrow[j]);
        ycbcr.val[2] = vld1_u8((uint8_t*) &crrow[j]);
        uint8x8x3_t rgb = ycbcr2rgb(ycbcr, c, v128);
        store_rgb(outptr, rgb, flags, alpha);
        outptr += rgbstride * 8;
      }
      if (trailing_cols)
      {
        uint8x8x3_t ycbcr;
        ycbcr.val[0] = vdup_n_u8(0);
        ycbcr.val[1] = vdup_n_u8(0);
        ycbcr.val[2] = vdup_n_u8(0);
        if (trailing_cols & 4)
        {
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 0); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 0); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 0); ++j; } while (0);
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 1); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 1); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 1); ++j; } while (0);
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 2); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 2); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 2); ++j; } while (0);
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 3); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 3); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 3); ++j; } while (0);
        }
        if (trailing_cols & 2)
        {
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 4); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 4); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 4); ++j; } while (0);
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 5); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 5); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 5); ++j; } while (0);
        }
        if (trailing_cols & 1)
        {
          do { ycbcr.val[0] = vld1_lane_u8((uint8_t*) &yrow[j], ycbcr.val[0], 6); ycbcr.val[1] = vld1_lane_u8((uint8_t*) &cbrow[j], ycbcr.val[1], 6); ycbcr.val[2] = vld1_lane_u8((uint8_t*) &crrow[j], ycbcr.val[2], 6); ++j; } while (0);
        }
        uint8x8x3_t rgb = ycbcr2rgb(ycbcr, c, v128);
        rgb = vswp_rgb(rgb, flags);
        if (flags & 4)
          ++outptr;
        if (trailing_cols & 4)
        {
          do { vst3_lane_u8(outptr, rgb, 0); outptr += rgbstride; } while (0);
          do { vst3_lane_u8(outptr, rgb, 1); outptr += rgbstride; } while (0);
          do { vst3_lane_u8(outptr, rgb, 2); outptr += rgbstride; } while (0);
          do { vst3_lane_u8(outptr, rgb, 3); outptr += rgbstride; } while (0);
        }
        if (trailing_cols & 2)
        {
          do { vst3_lane_u8(outptr, rgb, 4); outptr += rgbstride; } while (0);
          do { vst3_lane_u8(outptr, rgb, 5); outptr += rgbstride; } while (0);
        }
        if (trailing_cols & 1)
        {
          do { vst3_lane_u8(outptr, rgb, 6); outptr += rgbstride; } while (0);
        }
      }
    }
    return;
  }

  void (*neonfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);

  switch(cinfo->out_color_space)
  {
    case JCS_EXT_RGB:
      neonfct=jsimd_ycc_extrgb_convert_neon;
      break;
    case JCS_EXT_RGBX:
      neonfct=jsimd_ycc_extrgbx_convert_neon;
      break;
    case JCS_EXT_BGR:
      neonfct=jsimd_ycc_extbgr_convert_neon;
      break;
    case JCS_EXT_BGRX:
      neonfct=jsimd_ycc_extbgrx_convert_neon;
      break;
    case JCS_EXT_XBGR:
      neonfct=jsimd_ycc_extxbgr_convert_neon;
      break;
    case JCS_EXT_XRGB:
      neonfct=jsimd_ycc_extxrgb_convert_neon;
      break;
  default:
      neonfct=jsimd_ycc_extrgb_convert_neon;
      break;
  }

  neonfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
}

int
jsimd_can_h2v2_downsample (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_h2v1_downsample (void)
{
  init_simd();

  return 0;
}

void
jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
                       JSAMPARRAY input_data, JSAMPARRAY output_data)
{
}

void
jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
                       JSAMPARRAY input_data, JSAMPARRAY output_data)
{
}

int
jsimd_can_h2v2_upsample (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_h2v1_upsample (void)
{
  init_simd();

  return 0;
}

void
jsimd_h2v2_upsample (j_decompress_ptr cinfo,
                     jpeg_component_info * compptr,
                     JSAMPARRAY input_data,
                     JSAMPARRAY * output_data_ptr)
{
}

void
jsimd_h2v1_upsample (j_decompress_ptr cinfo,
                     jpeg_component_info * compptr,
                     JSAMPARRAY input_data,
                     JSAMPARRAY * output_data_ptr)
{
}

int
jsimd_can_h2v2_fancy_upsample (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_h2v1_fancy_upsample (void)
{
  init_simd();

  return 0;
}

void
jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
                           jpeg_component_info * compptr,
                           JSAMPARRAY input_data,
                           JSAMPARRAY * output_data_ptr)
{
}

void
jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
                           jpeg_component_info * compptr,
                           JSAMPARRAY input_data,
                           JSAMPARRAY * output_data_ptr)
{
}

int
jsimd_can_h2v2_merged_upsample (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_h2v1_merged_upsample (void)
{
  init_simd();

  return 0;
}

void
jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
                            JSAMPIMAGE input_buf,
                            JDIMENSION in_row_group_ctr,
                            JSAMPARRAY output_buf)
{
}

void
jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
                            JSAMPIMAGE input_buf,
                            JDIMENSION in_row_group_ctr,
                            JSAMPARRAY output_buf)
{
}

int
jsimd_can_convsamp (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 0;
    return 2;
  }

  return 0;
}

int
jsimd_can_convsamp_float (void)
{
  init_simd();

  return 0;
}

void
jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
                DCTELEM * workspace)
{
  if (!(simd_support & 0x10))
    return;

  int8x16_t v128 = vdupq_n_s8(-128);
  int8x16x2_t samples;
  uint16x8x2_t samples_as_u16;
  int16x8x4_t coef;
  int16x8x4_t* wp = (int16x8x4_t*) workspace;

  samples.val[0] = vcombine_s8(
    vld1_s8((const int8_t*)(sample_data[0] + start_col)),
    vld1_s8((const int8_t*)(sample_data[1] + start_col)));
  samples.val[1] = vcombine_s8(
    vld1_s8((const int8_t*)(sample_data[2] + start_col)),
    vld1_s8((const int8_t*)(sample_data[3] + start_col)));
  samples.val[0] = vaddq_s8(samples.val[0], v128);
  samples.val[1] = vaddq_s8(samples.val[1], v128);
  samples_as_u16 = vuzpq_u16(vreinterpretq_u16_s8(samples.val[0]), vreinterpretq_u16_s8(samples.val[1]));
  samples = vuzpq_s8(vreinterpretq_s8_u16(samples_as_u16.val[0]), vreinterpretq_s8_u16(samples_as_u16.val[1]));

  coef.val[0] = vshll_n_s8(vget_low_s8(samples.val[0]), 3);
  coef.val[1] = vshll_n_s8(vget_low_s8(samples.val[1]), 3);
  coef.val[2] = vshll_n_s8(vget_high_s8(samples.val[0]), 3);
  coef.val[3] = vshll_n_s8(vget_high_s8(samples.val[1]), 3);
  vst4q_s16((int16_t*)(wp++), coef);

  samples.val[0] = vcombine_s8(
    vld1_s8((const int8_t*)(sample_data[4] + start_col)),
    vld1_s8((const int8_t*)(sample_data[5] + start_col)));
  samples.val[1] = vcombine_s8(
    vld1_s8((const int8_t*)(sample_data[6] + start_col)),
    vld1_s8((const int8_t*)(sample_data[7] + start_col)));
  samples.val[0] = vaddq_s8(samples.val[0], v128);
  samples.val[1] = vaddq_s8(samples.val[1], v128);
  samples_as_u16 = vuzpq_u16(vreinterpretq_u16_s8(samples.val[0]), vreinterpretq_u16_s8(samples.val[1]));
  samples = vuzpq_s8(vreinterpretq_s8_u16(samples_as_u16.val[0]), vreinterpretq_s8_u16(samples_as_u16.val[1]));

  coef.val[0] = vshll_n_s8(vget_low_s8(samples.val[0]), 3);
  coef.val[1] = vshll_n_s8(vget_low_s8(samples.val[1]), 3);
  coef.val[2] = vshll_n_s8(vget_high_s8(samples.val[0]), 3);
  coef.val[3] = vshll_n_s8(vget_high_s8(samples.val[1]), 3);
  vst4q_s16((int16_t*)(wp++), coef);
# 1067 "jsimd_arm.c"
}

void
jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
                      float * workspace)
{
}

int
jsimd_can_fdct_islow (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_fdct_ifast (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if (sizeof(IFAST_MULT_TYPE) != 2)
    return 0;
  if (2 != 2)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 0;
    return 2;
  }

  return 0;
}

int
jsimd_can_fdct_float (void)
{
  init_simd();

  return 0;
}

void
jsimd_fdct_islow (DCTELEM * data)
{
}

void
jsimd_fdct_ifast (DCTELEM * data)
{
  if (!(simd_support & 0x10))
    return;

  int16_8x8_t coef;

  {
    const int16x8x4_t* coefp = (const int16x8x4_t*) data;
    coef = load_transpose_8x8(coefp);
  }

  {
    static int16_t fdct_constants[4] __attribute__ ((aligned (8)))
      = { 13573, 25080, 2700, 20091 };
    int16x4_t c = vld1_s16(fdct_constants);
    coef = transpose_8x8(fdct_helper_8x8(c, coef));

    coef.half[0].val[0] = vrshrq_n_s16(coef.half[0].val[0], 4);
    coef.half[0].val[1] = vrshrq_n_s16(coef.half[0].val[1], 4);
    coef.half[0].val[2] = vrshrq_n_s16(coef.half[0].val[2], 4);
    coef.half[0].val[3] = vrshrq_n_s16(coef.half[0].val[3], 4);
    coef.half[1].val[0] = vrshrq_n_s16(coef.half[1].val[0], 4);
    coef.half[1].val[1] = vrshrq_n_s16(coef.half[1].val[1], 4);
    coef.half[1].val[2] = vrshrq_n_s16(coef.half[1].val[2], 4);
    coef.half[1].val[3] = vrshrq_n_s16(coef.half[1].val[3], 4);

    coef = fdct_helper_8x8(c, coef);
  }

  {
    int16x8_t* coeflp = (int16x8_t*) data;
# 1178 "jsimd_arm.c"
    vst1q_s16((int16_t*)(coeflp++), coef.half[0].val[0]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[0].val[1]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[0].val[2]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[0].val[3]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[1].val[0]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[1].val[1]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[1].val[2]);
    vst1q_s16((int16_t*)(coeflp++), coef.half[1].val[3]);
  }
}

void
jsimd_fdct_float (float * data)
{
}

int
jsimd_can_quantize (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 0;
    return 2;
  }

  return 0;
}

int
jsimd_can_quantize_float (void)
{
  init_simd();

  return 0;
}

void
jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
                DCTELEM * workspace)
{
  if (!(simd_support & 0x10))
    return;

  int16x8x4_t* wp = (int16x8x4_t*) workspace;
  int16x8x4_t* coefp = (int16x8x4_t*) coef_block;
  int16x8x4_t* recipp = (int16x8x4_t*) &(divisors[64 * 0]);
  int16x8x4_t* corrp = (int16x8x4_t*) &(divisors[64 * 1]);
  int16x8x4_t* shiftp = (int16x8x4_t*) &(divisors[64 * 3]);
  int16x8x4_t coef, recip, corr, shift;

  coef = vld4q_s16((int16_t*)(wp++));
  recip = vld4q_s16((int16_t*)(recipp++));
  corr = vld4q_s16((int16_t*)(corrp++));
  shift = vld4q_s16((int16_t*)(shiftp++));
  coef.val[0] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[0], corr.val[0]), recip.val[0]), shift.val[0]);
  coef.val[1] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[1], corr.val[1]), recip.val[1]), shift.val[1]);
  coef.val[2] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[2], corr.val[2]), recip.val[2]), shift.val[2]);
  coef.val[3] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[3], corr.val[3]), recip.val[3]), shift.val[3]);
  vst4q_s16((int16_t*)(coefp++), coef);

  coef = vld4q_s16((int16_t*)(wp++));
  recip = vld4q_s16((int16_t*)(recipp++));
  corr = vld4q_s16((int16_t*)(corrp++));
  shift = vld4q_s16((int16_t*)(shiftp++));
  coef.val[0] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[0], corr.val[0]), recip.val[0]), shift.val[0]);
  coef.val[1] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[1], corr.val[1]), recip.val[1]), shift.val[1]);
  coef.val[2] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[2], corr.val[2]), recip.val[2]), shift.val[2]);
  coef.val[3] = vshlq_s16(vqdmulhq_s16(vqaddq_s16(coef.val[3], corr.val[3]), recip.val[3]), shift.val[3]);
  vst4q_s16((int16_t*)(coefp++), coef);
# 1334 "jsimd_arm.c"
}

void
jsimd_quantize_float (JCOEFPTR coef_block, float * divisors,
                      float * workspace)
{
}

int
jsimd_can_idct_2x2 (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if (sizeof(ISLOW_MULT_TYPE) != 2)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 1;
    return 2;
  }

  return 0;
}

int
jsimd_can_idct_4x4 (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if (sizeof(ISLOW_MULT_TYPE) != 2)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 1;
    return 2;
  }

  return 0;
}

void
jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                JCOEFPTR coef_block, JSAMPARRAY output_buf,
                JDIMENSION output_col)
{
  if (!(simd_support & 0x10))
    return;

  if ((simd_support & 0x20))
  {
    jsimd_idct_2x2_neon(compptr->dct_table, coef_block, output_buf, output_col);
    return;
  }

  int16_8x8_t coef;
  int16x8_t zero8 = vdupq_n_s16(0);

  {
    const int16x8x4_t* coefp = (const int16x8x4_t*) coef_block;
    const int16x8x4_t* dctp = (const int16x8x4_t*) compptr->dct_table;
    coef = dq_transpose_8x8(coefp, dctp);
  }

  {
    static int16_t idct_constants[4] __attribute__ ((aligned (8)))
      = { 13573, 27779, 2700, 20091 };
    int16x4_t c = vld1_s16(idct_constants);
    coef.half[0].val[2] = zero8;
    coef.half[1].val[0] = zero8;
    coef.half[1].val[2] = zero8;
    coef = transpose_8x8(idct_helper_8x8(c, coef));
    coef.half[0].val[2] = zero8;
    coef.half[1].val[0] = zero8;
    coef.half[1].val[2] = zero8;
    coef = idct_helper_8x8(c, coef);
  }

  {
    int32x4_t half0 = vaddq_s32(
      vpaddlq_s16(vqaddq_s16(coef.half[0].val[0], coef.half[0].val[1])),
      vpaddlq_s16(vqaddq_s16(coef.half[0].val[2], coef.half[0].val[3])));
    int32x4_t half1 = vaddq_s32(
      vpaddlq_s16(vqaddq_s16(coef.half[1].val[0], coef.half[1].val[1])),
      vpaddlq_s16(vqaddq_s16(coef.half[1].val[2], coef.half[1].val[3])));
    int16x4_t all = vqrshrn_n_s32(vcombine_s32(
      vpadd_s32(vget_low_s32(half0), vget_high_s32(half0)),
      vpadd_s32(vget_low_s32(half1), vget_high_s32(half1))), 7 + 4);

    int16x4_t zero4 = vdup_n_s16(0);
    uint8x8_t result = vreinterpret_u8_s8(vqmovn_s16(vcombine_s16(all, zero4)));
    uint8x8_t v128 = vdup_n_u8(128);
    result = vadd_u8(result, v128);
# 1457 "jsimd_arm.c"
    uint8_t buf[16];
    uint8_t* bufp = (uint8_t*) (((int) &buf[8]) & ~7);
    vst1_u8(bufp, result);
    *((uint16_t*)(output_buf[0] + output_col)) = *((uint16_t*)&bufp[0]);
    *((uint16_t*)(output_buf[1] + output_col)) = *((uint16_t*)&bufp[2]);
  }
}

void
jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                JCOEFPTR coef_block, JSAMPARRAY output_buf,
                JDIMENSION output_col)
{
  if (!(simd_support & 0x10))
    return;

  if ((simd_support & 0x20))
  {
    jsimd_idct_4x4_neon(compptr->dct_table, coef_block, output_buf, output_col);
    return;
  }

  int16_8x8_t coef;
  int16x8_t zero8 = vdupq_n_s16(0);

  {
    const int16x8x4_t* coefp = (const int16x8x4_t*) coef_block;
    const int16x8x4_t* dctp = (const int16x8x4_t*) compptr->dct_table;
    coef = dq_transpose_8x8(coefp, dctp);
  }

  {
    static int16_t idct_constants[4] __attribute__ ((aligned (8)))
      = { 13573, 27779, 2700, 20091 };
    int16x4_t c = vld1_s16(idct_constants);
    coef.half[1].val[0] = zero8;
    coef = transpose_8x8(idct_helper_8x8(c, coef));
    coef.half[1].val[0] = zero8;
    coef = idct_helper_8x8(c, coef);
  }

  {
    uint8x16_t result;
    result = vreinterpretq_u8_s8(vcombine_s8(
      vqmovn_s16(vcombine_s16(
        vqrshrn_n_s32(vpaddlq_s16(vqaddq_s16(coef.half[0].val[0], coef.half[0].val[1])), 5 + 4),
        vqrshrn_n_s32(vpaddlq_s16(vqaddq_s16(coef.half[0].val[2], coef.half[0].val[3])), 5 + 4))),
      vqmovn_s16(vcombine_s16(
        vqrshrn_n_s32(vpaddlq_s16(vqaddq_s16(coef.half[1].val[0], coef.half[1].val[1])), 5 + 4),
        vqrshrn_n_s32(vpaddlq_s16(vqaddq_s16(coef.half[1].val[2], coef.half[1].val[3])), 5 + 4)))));

    uint8x16_t v128 = vdupq_n_u8(128);
    result = vaddq_u8(result, v128);
# 1534 "jsimd_arm.c"
    uint8_t buf[32];
    uint8_t* bufp = (uint8_t*) (((int) &buf[16]) & ~15);
    vst1q_u8(bufp, result);
    *((uint32_t*)(output_buf[0] + output_col)) = *((uint32_t*)&bufp[0]);
    *((uint32_t*)(output_buf[1] + output_col)) = *((uint32_t*)&bufp[4]);
    *((uint32_t*)(output_buf[2] + output_col)) = *((uint32_t*)&bufp[8]);
    *((uint32_t*)(output_buf[3] + output_col)) = *((uint32_t*)&bufp[12]);
  }
}

int
jsimd_can_idct_islow (void)
{
  init_simd();

  return 0;
}

int
jsimd_can_idct_ifast (void)
{
  init_simd();


  if (8 != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (8 != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if (sizeof(IFAST_MULT_TYPE) != 2)
    return 0;
  if (2 != 2)
    return 0;

  if ((simd_support & 0x10)) {
    if ((simd_support & 0x20))
      return 1;
    return 2;
  }

  return 0;
}

int
jsimd_can_idct_float (void)
{
  init_simd();

  return 0;
}

void
jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                JCOEFPTR coef_block, JSAMPARRAY output_buf,
                JDIMENSION output_col)
{
}

void
jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                JCOEFPTR coef_block, JSAMPARRAY output_buf,
                JDIMENSION output_col)
{
  if (!(simd_support & 0x10))
    return;

  if ((simd_support & 0x20))
  {
    jsimd_idct_ifast_neon(compptr->dct_table, coef_block, output_buf, output_col);
    return;
  }

  int16_8x8_t coef;

  {
    const int16x8x4_t* coefp = (const int16x8x4_t*) coef_block;
    const int16x8x4_t* dctp = (const int16x8x4_t*) compptr->dct_table;
    coef = dq_transpose_8x8(coefp, dctp);
  }

  {
    static int16_t idct_constants[4] __attribute__ ((aligned (8)))
      = { 13573, 27779, 2700, 20091 };
    int16x4_t c = vld1_s16(idct_constants);
    coef = idct_helper_8x8(c, transpose_8x8(idct_helper_8x8(c, coef)));
  }

  {
    uint8x16x4_t result;
    result.val[0] = vreinterpretq_u8_s8(vcombine_s8(
      vqrshrn_n_s16(coef.half[0].val[0], 3 + 4),
      vqrshrn_n_s16(coef.half[0].val[1], 3 + 4)));
    result.val[1] = vreinterpretq_u8_s8(vcombine_s8(
      vqrshrn_n_s16(coef.half[0].val[2], 3 + 4),
      vqrshrn_n_s16(coef.half[0].val[3], 3 + 4)));
    result.val[2] = vreinterpretq_u8_s8(vcombine_s8(
      vqrshrn_n_s16(coef.half[1].val[0], 3 + 4),
      vqrshrn_n_s16(coef.half[1].val[1], 3 + 4)));
    result.val[3] = vreinterpretq_u8_s8(vcombine_s8(
      vqrshrn_n_s16(coef.half[1].val[2], 3 + 4),
      vqrshrn_n_s16(coef.half[1].val[3], 3 + 4)));

    uint8x16_t v128 = vdupq_n_u8(128);
    result.val[0] = vaddq_u8(result.val[0], v128);
    result.val[1] = vaddq_u8(result.val[1], v128);
    result.val[2] = vaddq_u8(result.val[2], v128);
    result.val[3] = vaddq_u8(result.val[3], v128);
# 1672 "jsimd_arm.c"
    vst1_u8((uint8_t*)(output_buf[0] + output_col), vget_low_u8(result.val[0]));
    vst1_u8((uint8_t*)(output_buf[1] + output_col), vget_high_u8(result.val[0]));
    vst1_u8((uint8_t*)(output_buf[2] + output_col), vget_low_u8(result.val[1]));
    vst1_u8((uint8_t*)(output_buf[3] + output_col), vget_high_u8(result.val[1]));
    vst1_u8((uint8_t*)(output_buf[4] + output_col), vget_low_u8(result.val[2]));
    vst1_u8((uint8_t*)(output_buf[5] + output_col), vget_high_u8(result.val[2]));
    vst1_u8((uint8_t*)(output_buf[6] + output_col), vget_low_u8(result.val[3]));
    vst1_u8((uint8_t*)(output_buf[7] + output_col), vget_high_u8(result.val[3]));
  }
}

void
jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                JCOEFPTR coef_block, JSAMPARRAY output_buf,
                JDIMENSION output_col)
{
}