#ifndef _TMMINTRIN_H_INCLUDED
#define _TMMINTRIN_H_INCLUDED
#include "third_party/intel/pmmintrin.internal.h"

#ifndef __SSSE3__
#pragma GCC push_options
#pragma GCC target("ssse3")
#define __DISABLE_SSSE3__
#endif /* __SSSE3__ */

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadd_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phaddw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadd_epi32(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phaddd128((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadds_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phaddsw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadd_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phaddw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadd_pi32(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phaddd((__v2si)__X, (__v2si)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hadds_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phaddsw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsub_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phsubw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsub_epi32(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phsubd128((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsubs_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_phsubsw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsub_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phsubw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsub_pi32(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phsubd((__v2si)__X, (__v2si)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_hsubs_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_phsubsw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_maddubs_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_pmaddubsw128((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_maddubs_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_pmaddubsw((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_mulhrs_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_pmulhrsw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_mulhrs_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_pmulhrsw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_shuffle_epi8(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_pshufb128((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_shuffle_pi8(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_pshufb((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_epi8(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_psignb128((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_epi16(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_psignw128((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_epi32(__m128i __X, __m128i __Y) {
  return (__m128i)__builtin_ia32_psignd128((__v4si)__X, (__v4si)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_pi8(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_psignb((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_pi16(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_psignw((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_sign_pi32(__m64 __X, __m64 __Y) {
  return (__m64)__builtin_ia32_psignd((__v2si)__X, (__v2si)__Y);
}

#ifdef __OPTIMIZE__
extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_alignr_epi8(__m128i __X, __m128i __Y, const int __N) {
  return (__m128i)__builtin_ia32_palignr128((__v2di)__X, (__v2di)__Y, __N * 8);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_alignr_pi8(__m64 __X, __m64 __Y, const int __N) {
  return (__m64)__builtin_ia32_palignr((__v1di)__X, (__v1di)__Y, __N * 8);
}
#else
#define _mm_alignr_epi8(X, Y, N)                            \
  ((__m128i)__builtin_ia32_palignr128((__v2di)(__m128i)(X), \
                                      (__v2di)(__m128i)(Y), (int)(N)*8))
#define _mm_alignr_pi8(X, Y, N)                                          \
  ((__m64)__builtin_ia32_palignr((__v1di)(__m64)(X), (__v1di)(__m64)(Y), \
                                 (int)(N)*8))
#endif

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_epi8(__m128i __X) {
  return (__m128i)__builtin_ia32_pabsb128((__v16qi)__X);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_epi16(__m128i __X) {
  return (__m128i)__builtin_ia32_pabsw128((__v8hi)__X);
}

extern __inline __m128i
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_epi32(__m128i __X) {
  return (__m128i)__builtin_ia32_pabsd128((__v4si)__X);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_pi8(__m64 __X) {
  return (__m64)__builtin_ia32_pabsb((__v8qi)__X);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_pi16(__m64 __X) {
  return (__m64)__builtin_ia32_pabsw((__v4hi)__X);
}

extern __inline __m64
    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    _mm_abs_pi32(__m64 __X) {
  return (__m64)__builtin_ia32_pabsd((__v2si)__X);
}

#ifdef __DISABLE_SSSE3__
#undef __DISABLE_SSSE3__
#pragma GCC pop_options
#endif /* __DISABLE_SSSE3__ */

#endif /* _TMMINTRIN_H_INCLUDED */