emmintrin.h

Mac OS X 10.4.9 for x86 Source Code gcc 实现源代码

_mm_adds_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddusw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sub_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sub_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sub_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubd128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sub_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubq128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_subs_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubsb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_subs_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubsw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_subs_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubusb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_subs_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubusw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_madd_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaddwd128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_mulhi_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmulhw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_mullo_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmullw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m64 __attribute__((__always_inline__, __nodebug__))
_mm_mul_su32 (__m64 __A, __m64 __B)
{
  return (__m64)__builtin_ia32_pmuludq ((__v2si)__A, (__v2si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_mul_epu32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmuludq128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_slli_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psllwi128 ((__v8hi)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_slli_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_pslldi128 ((__v4si)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_slli_epi64 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psllqi128 ((__v2di)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srai_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrawi128 ((__v8hi)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srai_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psradi128 ((__v4si)__A, __B);
}

#if 0
static __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_si128 (__m128i __A, const int __B)
{
  return ((__m128i)__builtin_ia32_psrldqi128 (__A, __B))
}

static __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_si128 (__m128i __A, const int __B)
{
  return ((__m128i)__builtin_ia32_pslldqi128 (__A, __B))
}
#else
#define _mm_srli_si128(__A, __B) \
  ((__m128i)__builtin_ia32_psrldqi128 (__A, (__B) * 8))
#define _mm_slli_si128(__A, __B) \
  ((__m128i)__builtin_ia32_pslldqi128 (__A, (__B) * 8))
#endif

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrlwi128 ((__v8hi)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrldi128 ((__v4si)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_epi64 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrlqi128 ((__v2di)__A, __B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sll_epi16 (__m128i __A, __m128i __B)
{
  return _mm_slli_epi16 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sll_epi32 (__m128i __A, __m128i __B)
{
  return _mm_slli_epi32 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sll_epi64 (__m128i __A, __m128i __B)
{
  return _mm_slli_epi64 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sra_epi16 (__m128i __A, __m128i __B)
{
  return _mm_srai_epi16 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sra_epi32 (__m128i __A, __m128i __B)
{
  return _mm_srai_epi32 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srl_epi16 (__m128i __A, __m128i __B)
{
  return _mm_srli_epi16 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srl_epi32 (__m128i __A, __m128i __B)
{
  return _mm_srli_epi32 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srl_epi64 (__m128i __A, __m128i __B)
{
  return _mm_srli_epi64 (__A, _mm_cvtsi128_si32 (__B));
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_and_si128 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pand128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_andnot_si128 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pandn128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_or_si128 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_por128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_xor_si128 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pxor128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpeqb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpeqw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpeqd128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtb128 ((__v16qi)__B, (__v16qi)__A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtw128 ((__v8hi)__B, (__v8hi)__A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtd128 ((__v4si)__B, (__v4si)__A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pcmpgtd128 ((__v4si)__A, (__v4si)__B);
}

#if 0
static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_extract_epi16 (__m128i const __A, int const __N)
{
  return __builtin_ia32_vec_ext_v8hi ((__v8hi)__A, __N);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_insert_epi16 (__m128i const __A, int const __D, int const __N)
{
  return (__m128i) __builtin_ia32_vec_set_v8hi ((__v8hi)__A, __D, __N);
}
#else
#define _mm_extract_epi16(A, N) \
  ((int) __builtin_ia32_vec_ext_v8hi ((__v8hi)(A), (N)))
#define _mm_insert_epi16(A, D, N) \
  ((__m128i) __builtin_ia32_vec_set_v8hi ((__v8hi)(A), (D), (N)))
#endif

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_max_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaxsw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_max_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaxub128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_min_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pminsw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_min_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pminub128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_movemask_epi8 (__m128i __A)
{
  return __builtin_ia32_pmovmskb128 ((__v16qi)__A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_mulhi_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmulhuw128 ((__v8hi)__A, (__v8hi)__B);
}

#define _mm_shufflehi_epi16(__A, __B) ((__m128i)__builtin_ia32_pshufhw ((__v8hi)__A, __B))
#define _mm_shufflelo_epi16(__A, __B) ((__m128i)__builtin_ia32_pshuflw ((__v8hi)__A, __B))
#define _mm_shuffle_epi32(__A, __B) ((__m128i)__builtin_ia32_pshufd ((__v4si)__A, __B))

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_maskmoveu_si128 (__m128i __A, __m128i __B, char *__C)
{
  __builtin_ia32_maskmovdqu ((__v16qi)__A, (__v16qi)__B, __C);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_avg_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pavgb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_avg_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pavgw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_sad_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psadbw128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_stream_si32 (int *__A, int __B)
{
  __builtin_ia32_movnti (__A, __B);
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_stream_si128 (__m128i *__A, __m128i __B)
{
  __builtin_ia32_movntdq ((__v2di *)__A, (__v2di)__B);
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_stream_pd (double *__A, __m128d __B)
{
  __builtin_ia32_movntpd (__A, (__v2df)__B);
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_clflush (void const *__A)
{
  __builtin_ia32_clflush (__A);
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_lfence (void)
{
  __builtin_ia32_lfence ();
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_mfence (void)
{
  __builtin_ia32_mfence ();
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi32_si128 (int __A)
{
  return _mm_set_epi32 (0, 0, 0, __A);
}

#ifdef __x86_64__
/* Intel intrinsic.  */
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi64_si128 (long long __A)
{
  return _mm_set_epi64x (0, __A);
}

/* Microsoft intrinsic.  */
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi64x_si128 (long long __A)
{
  return _mm_set_epi64x (0, __A);
}
#endif

/* Casts between various SP, DP, INT vector types.  Note that these do no
   conversion of values, they just change the type.  */
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_castpd_ps(__m128d __A)
{
  return (__m128) __A;
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_castpd_si128(__m128d __A)
{
  return (__m128i) __A;
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_castps_pd(__m128 __A)
{
  return (__m128d) __A;
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_castps_si128(__m128 __A)
{
  return (__m128i) __A;
}

static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_castsi128_ps(__m128i __A)
{
  return (__m128) __A;
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_castsi128_pd(__m128i __A)
{
  return (__m128d) __A;
}
/* APPLE LOCAL end radar 4152603 */

/* APPLE LOCAL begin 4220129 */
/* Set all of the elements of the vector to A.  */

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set1_epi64x (long long __A)
{
  return _mm_set_epi64x (__A, __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set1_epi64 (__m64 __A)
{
  return _mm_set_epi64 (__A, __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set1_epi32 (int __A)
{
  return _mm_set_epi32 (__A, __A, __A, __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set1_epi16 (short __A)
{
  return _mm_set_epi16 (__A, __A, __A, __A, __A, __A, __A, __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set1_epi8 (char __A)
{
    __m128i temp, temp2, temp3, temp4;
    temp = _mm_cvtsi32_si128 ((int)__A);
    temp2 = _mm_unpacklo_epi8 (temp, temp);
    temp3 = _mm_unpacklo_epi8 (temp2, temp2);
    temp4 = _mm_shuffle_epi32 (temp3, 0);
    return temp4;
}
/* APPLE LOCAL end 4220129 */
#endif /* __SSE2__  */

#endif /* _EMMINTRIN_H_INCLUDED */