emmintrin.h

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

}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_comilt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdlt ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_comile_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdle ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_comigt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdgt ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_comige_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdge ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_comineq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdneq ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomieq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdeq ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomilt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdlt ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomile_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdle ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomigt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdgt ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomige_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdge ((__v2df)__A, (__v2df)__B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_ucomineq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdneq ((__v2df)__A, (__v2df)__B);
}

/* Create a vector of Qi, where i is the element number.  */

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set_epi64x (long long __q1, long long __q0)
{
  return __extension__ (__m128i)(__v2di){ __q0, __q1 };
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set_epi64 (__m64 __q1,  __m64 __q0)
{
  return _mm_set_epi64x ((long long)__q1, (long long)__q0);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set_epi32 (int __q3, int __q2, int __q1, int __q0)
{
  return __extension__ (__m128i)(__v4si){ __q0, __q1, __q2, __q3 };
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set_epi16 (short __q7, short __q6, short __q5, short __q4,
	       short __q3, short __q2, short __q1, short __q0)
{
  return __extension__ (__m128i)(__v8hi){
    __q0, __q1, __q2, __q3, __q4, __q5, __q6, __q7 };
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_set_epi8 (char __q15, char __q14, char __q13, char __q12,
	      char __q11, char __q10, char __q09, char __q08,
	      char __q07, char __q06, char __q05, char __q04,
	      char __q03, char __q02, char __q01, char __q00)
{
  return __extension__ (__m128i)(__v16qi){
    __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07,
    __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15
  };
}

/* APPLE LOCAL begin 4220129 */
/* functions moved to end of file */
/* APPLE LOCAL end 4220129 */

/* Create a vector of Qi, where i is the element number.
   The parameter order is reversed from the _mm_set_epi* functions.  */

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_setr_epi64 (__m64 __q0, __m64 __q1)
{
  return _mm_set_epi64 (__q1, __q0);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_setr_epi32 (int __q0, int __q1, int __q2, int __q3)
{
  return _mm_set_epi32 (__q3, __q2, __q1, __q0);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_setr_epi16 (short __q0, short __q1, short __q2, short __q3,
	        short __q4, short __q5, short __q6, short __q7)
{
  return _mm_set_epi16 (__q7, __q6, __q5, __q4, __q3, __q2, __q1, __q0);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_setr_epi8 (char __q00, char __q01, char __q02, char __q03,
	       char __q04, char __q05, char __q06, char __q07,
	       char __q08, char __q09, char __q10, char __q11,
	       char __q12, char __q13, char __q14, char __q15)
{
  return _mm_set_epi8 (__q15, __q14, __q13, __q12, __q11, __q10, __q09, __q08,
		       __q07, __q06, __q05, __q04, __q03, __q02, __q01, __q00);
}

/* Create a vector with element 0 as *P and the rest zero.  */

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_load_si128 (__m128i const *__P)
{
  return *__P;
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_loadu_si128 (__m128i const *__P)
{
  return (__m128i) __builtin_ia32_loaddqu ((char const *)__P);
}

/* APPLE LOCAL begin 4099020 */
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_loadl_epi64 (__m128i const *__P)
{
  return  (__m128i)__builtin_ia32_loadlv4si ((__v2si *)__P);
}
/* APPLE LOCAL end 4099020 */

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_store_si128 (__m128i *__P, __m128i __B)
{
  *__P = __B;
}

static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_storeu_si128 (__m128i *__P, __m128i __B)
{
  __builtin_ia32_storedqu ((char *)__P, (__v16qi)__B);
}

/* APPLE LOCAL begin 4099020 */
static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_storel_epi64 (__m128i *__P, __m128i __B)
{
  __builtin_ia32_storelv4si ((__v2si *)__P, __B);
}
/* APPLE LOCAL end 4099020 */

static __inline __m64 __attribute__((__always_inline__, __nodebug__))
_mm_movepi64_pi64 (__m128i __B)
{
  return (__m64) __builtin_ia32_vec_ext_v2di ((__v2di)__B, 0);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_movpi64_epi64 (__m64 __A)
{
  return _mm_set_epi64 ((__m64)0LL, __A);
}

/* APPLE LOCAL begin 4099020 */
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_move_epi64 (__m128i __A)
{
  return (__m128i)__builtin_ia32_movqv4si ((__v4si)__A) ;
}
/* APPLE LOCAL end 4099020 */

/* Create a vector of zeros.  */
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_setzero_si128 (void)
{
  return __extension__ (__m128i)(__v4si){ 0, 0, 0, 0 };
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtepi32_pd (__m128i __A)
{
  return (__m128d)__builtin_ia32_cvtdq2pd ((__v4si) __A);
}

static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cvtepi32_ps (__m128i __A)
{
  return (__m128)__builtin_ia32_cvtdq2ps ((__v4si) __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvtpd_epi32 (__m128d __A)
{
  return (__m128i)__builtin_ia32_cvtpd2dq ((__v2df) __A);
}

static __inline __m64 __attribute__((__always_inline__, __nodebug__))
_mm_cvtpd_pi32 (__m128d __A)
{
  return (__m64)__builtin_ia32_cvtpd2pi ((__v2df) __A);
}

static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cvtpd_ps (__m128d __A)
{
  return (__m128)__builtin_ia32_cvtpd2ps ((__v2df) __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvttpd_epi32 (__m128d __A)
{
  return (__m128i)__builtin_ia32_cvttpd2dq ((__v2df) __A);
}

static __inline __m64 __attribute__((__always_inline__, __nodebug__))
_mm_cvttpd_pi32 (__m128d __A)
{
  return (__m64)__builtin_ia32_cvttpd2pi ((__v2df) __A);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtpi32_pd (__m64 __A)
{
  return (__m128d)__builtin_ia32_cvtpi2pd ((__v2si) __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvtps_epi32 (__m128 __A)
{
  return (__m128i)__builtin_ia32_cvtps2dq ((__v4sf) __A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_cvttps_epi32 (__m128 __A)
{
  return (__m128i)__builtin_ia32_cvttps2dq ((__v4sf) __A);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtps_pd (__m128 __A)
{
  return (__m128d)__builtin_ia32_cvtps2pd ((__v4sf) __A);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_cvtsd_si32 (__m128d __A)
{
  return __builtin_ia32_cvtsd2si ((__v2df) __A);
}

#ifdef __x86_64__
/* Intel intrinsic.  */
static __inline long long __attribute__((__always_inline__, __nodebug__))
_mm_cvtsd_si64 (__m128d __A)
{
  return __builtin_ia32_cvtsd2si64 ((__v2df) __A);
}

/* Microsoft intrinsic.  */
static __inline long long __attribute__((__always_inline__, __nodebug__))
_mm_cvtsd_si64x (__m128d __A)
{
  return __builtin_ia32_cvtsd2si64 ((__v2df) __A);
}
#endif

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_cvttsd_si32 (__m128d __A)
{
  return __builtin_ia32_cvttsd2si ((__v2df) __A);
}

#ifdef __x86_64__
/* Intel intrinsic.  */
static __inline long long __attribute__((__always_inline__, __nodebug__))
_mm_cvttsd_si64 (__m128d __A)
{
  return __builtin_ia32_cvttsd2si64 ((__v2df) __A);
}

/* Microsoft intrinsic.  */
static __inline long long __attribute__((__always_inline__, __nodebug__))
_mm_cvttsd_si64x (__m128d __A)
{
  return __builtin_ia32_cvttsd2si64 ((__v2df) __A);
}
#endif

static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cvtsd_ss (__m128 __A, __m128d __B)
{
  return (__m128)__builtin_ia32_cvtsd2ss ((__v4sf) __A, (__v2df) __B);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi32_sd (__m128d __A, int __B)
{
  return (__m128d)__builtin_ia32_cvtsi2sd ((__v2df) __A, __B);
}

#ifdef __x86_64__
/* Intel intrinsic.  */
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi64_sd (__m128d __A, long long __B)
{
  return (__m128d)__builtin_ia32_cvtsi642sd ((__v2df) __A, __B);
}

/* Microsoft intrinsic.  */
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtsi64x_sd (__m128d __A, long long __B)
{
  return (__m128d)__builtin_ia32_cvtsi642sd ((__v2df) __A, __B);
}
#endif

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cvtss_sd (__m128d __A, __m128 __B)
{
  return (__m128d)__builtin_ia32_cvtss2sd ((__v2df) __A, (__v4sf)__B);
}

#define _mm_shuffle_pd(__A, __B, __C) ((__m128d)__builtin_ia32_shufpd ((__v2df)__A, (__v2df)__B, (__C)))

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_unpackhi_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_unpckhpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_unpacklo_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_unpcklpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_loadh_pd (__m128d __A, double const *__B)
{
  return (__m128d)__builtin_ia32_loadhpd ((__v2df)__A, __B);
}

static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_loadl_pd (__m128d __A, double const *__B)
{
  return (__m128d)__builtin_ia32_loadlpd ((__v2df)__A, __B);
}

static __inline int __attribute__((__always_inline__, __nodebug__))
_mm_movemask_pd (__m128d __A)
{
  return __builtin_ia32_movmskpd ((__v2df)__A);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_packs_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packsswb128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_packs_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packssdw128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_packus_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packuswb128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpackhi_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhbw128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpackhi_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhwd128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpackhi_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhdq128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpackhi_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhqdq128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpacklo_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklbw128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpacklo_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklwd128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpacklo_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckldq128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_unpacklo_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklqdq128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_add_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_add_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_add_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddd128 ((__v4si)__A, (__v4si)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_add_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddq128 ((__v2di)__A, (__v2di)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_adds_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddsb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_adds_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddsw128 ((__v8hi)__A, (__v8hi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_adds_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddusb128 ((__v16qi)__A, (__v16qi)__B);
}

static __inline __m128i __attribute__((__always_inline__, __nodebug__))