xmmintrin.h

gcc-you can use this code to learn something about gcc, and inquire further into linux,

{
  return (__m64) __builtin_ia32_pavgb ((__v8qi)__A, (__v8qi)__B);
}

/* Compute the rounded averages of the unsigned 16-bit values in A and B.  */
static __inline __m64
_mm_avg_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pavgw ((__v4hi)__A, (__v4hi)__B);
}

/* Compute the sum of the absolute differences of the unsigned 8-bit
   values in A and B.  Return the value in the lower 16-bit word; the
   upper words are cleared.  */
static __inline __m64
_mm_sad_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_psadbw ((__v8qi)__A, (__v8qi)__B);
}

/* Loads one cache line from address P to a location "closer" to the
   processor.  The selector I specifies the type of prefetch operation.  */
#if 0
static __inline void
_mm_prefetch (void *__P, enum _mm_hint __I)
{
  __builtin_prefetch (__P, 0, __I);
}
#else
#define _mm_prefetch(P, I) \
  __builtin_prefetch ((P), 0, (I))
#endif

/* Stores the data in A to the address P without polluting the caches.  */
static __inline void
_mm_stream_pi (__m64 *__P, __m64 __A)
{
  __builtin_ia32_movntq ((unsigned long long *)__P, (unsigned long long)__A);
}

/* Likewise.  The address must be 16-byte aligned.  */
static __inline void
_mm_stream_ps (float *__P, __m128 __A)
{
  __builtin_ia32_movntps (__P, (__v4sf)__A);
}

/* Guarantees that every preceeding store is globally visible before
   any subsequent store.  */
static __inline void
_mm_sfence (void)
{
  __builtin_ia32_sfence ();
}

/* The execution of the next instruction is delayed by an implementation
   specific amount of time.  The instruction does not modify the
   architectural state.  */
static __inline void
_mm_pause (void)
{
  __asm__ __volatile__ ("rep; nop" : : );
}

/* Transpose the 4x4 matrix composed of row[0-3].  */
#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3)			\
do {									\
  __v4sf __r0 = (row0), __r1 = (row1), __r2 = (row2), __r3 = (row3);	\
  __v4sf __t0 = __builtin_ia32_shufps (__r0, __r1, 0x44);		\
  __v4sf __t2 = __builtin_ia32_shufps (__r0, __r1, 0xEE);		\
  __v4sf __t1 = __builtin_ia32_shufps (__r2, __r3, 0x44);		\
  __v4sf __t3 = __builtin_ia32_shufps (__r2, __r3, 0xEE);		\
  (row0) = __builtin_ia32_shufps (__t0, __t1, 0x88);			\
  (row1) = __builtin_ia32_shufps (__t0, __t1, 0xDD);			\
  (row2) = __builtin_ia32_shufps (__t2, __t3, 0x88);			\
  (row3) = __builtin_ia32_shufps (__t2, __t3, 0xDD);			\
} while (0)

#ifdef __SSE2__
/* SSE2 */
typedef int __v2df __attribute__ ((mode (V2DF)));
typedef int __v2di __attribute__ ((mode (V2DI)));
typedef int __v4si __attribute__ ((mode (V4SI)));
typedef int __v8hi __attribute__ ((mode (V8HI)));
typedef int __v16qi __attribute__ ((mode (V16QI)));

/* Create a selector for use with the SHUFPD instruction.  */
#define _MM_SHUFFLE2(fp1,fp0) \
 (((fp1) << 1) | (fp0))

#define __m128i __v2di
#define __m128d __v2df

/* Create a vector with element 0 as *P and the rest zero.  */
static __inline __m128d
_mm_load_sd (double const *__P)
{
  return (__m128d) __builtin_ia32_loadsd (__P);
}

/* Create a vector with all two elements equal to *P.  */
static __inline __m128d
_mm_load1_pd (double const *__P)
{
  __v2df __tmp = __builtin_ia32_loadsd (__P);
  return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,0));
}

static __inline __m128d
_mm_load_pd1 (double const *__P)
{
  return _mm_load1_pd (__P);
}

/* Load two DPFP values from P.  The addresd must be 16-byte aligned.  */
static __inline __m128d
_mm_load_pd (double const *__P)
{
  return (__m128d) __builtin_ia32_loadapd (__P);
}

/* Load two DPFP values from P.  The addresd need not be 16-byte aligned.  */
static __inline __m128d
_mm_loadu_pd (double const *__P)
{
  return (__m128d) __builtin_ia32_loadupd (__P);
}

/* Load two DPFP values in reverse order.  The addresd must be aligned.  */
static __inline __m128d
_mm_loadr_pd (double const *__P)
{
  __v2df __tmp = __builtin_ia32_loadapd (__P);
  return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,1));
}

/* Create a vector with element 0 as F and the rest zero.  */
static __inline __m128d
_mm_set_sd (double __F)
{
  return (__m128d) __builtin_ia32_loadsd (&__F);
}

/* Create a vector with all two elements equal to F.  */
static __inline __m128d
_mm_set1_pd (double __F)
{
  __v2df __tmp = __builtin_ia32_loadsd (&__F);
  return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,0));
}

static __inline __m128d
_mm_set_pd1 (double __F)
{
  return _mm_set1_pd (__F);
}

/* Create the vector [Z Y].  */
static __inline __m128d
_mm_set_pd (double __Z, double __Y)
{
  union {
    double __a[2];
    __m128d __v;
  } __u;

  __u.__a[0] = __Y;
  __u.__a[1] = __Z;

  return __u.__v;
}

/* Create the vector [Y Z].  */
static __inline __m128d
_mm_setr_pd (double __Z, double __Y)
{
  return _mm_set_pd (__Y, __Z);
}

/* Create a vector of zeros.  */
static __inline __m128d
_mm_setzero_pd (void)
{
  return (__m128d) __builtin_ia32_setzeropd ();
}

/* Stores the lower DPFP value.  */
static __inline void
_mm_store_sd (double *__P, __m128d __A)
{
  __builtin_ia32_storesd (__P, (__v2df)__A);
}

/* Store the lower DPFP value acrosd two words.  */
static __inline void
_mm_store1_pd (double *__P, __m128d __A)
{
  __v2df __va = (__v2df)__A;
  __v2df __tmp = __builtin_ia32_shufpd (__va, __va, _MM_SHUFFLE2 (0,0));
  __builtin_ia32_storeapd (__P, __tmp);
}

static __inline void
_mm_store_pd1 (double *__P, __m128d __A)
{
  _mm_store1_pd (__P, __A);
}

/* Store two DPFP values.  The addresd must be 16-byte aligned.  */
static __inline void
_mm_store_pd (double *__P, __m128d __A)
{
  __builtin_ia32_storeapd (__P, (__v2df)__A);
}

/* Store two DPFP values.  The addresd need not be 16-byte aligned.  */
static __inline void
_mm_storeu_pd (double *__P, __m128d __A)
{
  __builtin_ia32_storeupd (__P, (__v2df)__A);
}

/* Store two DPFP values in reverse order.  The addresd must be aligned.  */
static __inline void
_mm_storer_pd (double *__P, __m128d __A)
{
  __v2df __va = (__v2df)__A;
  __v2df __tmp = __builtin_ia32_shufpd (__va, __va, _MM_SHUFFLE2 (0,1));
  __builtin_ia32_storeapd (__P, __tmp);
}

/* Sets the low DPFP value of A from the low value of B.  */
static __inline __m128d
_mm_move_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
}


static __inline __m128d
_mm_add_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_addpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_add_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_addsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_sub_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_subpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_sub_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_subsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_mul_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_mulpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_mul_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_mulsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_div_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_divpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_div_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_divsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_sqrt_pd (__m128d __A)
{
  return (__m128d)__builtin_ia32_sqrtpd ((__v2df)__A);
}

/* Return pair {sqrt (A[0), B[1]}.  */
static __inline __m128d
_mm_sqrt_sd (__m128d __A, __m128d __B)
{
  __v2df __tmp = __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
  return (__m128d)__builtin_ia32_sqrtsd ((__v2df)__tmp);
}

static __inline __m128d
_mm_min_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_minpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_min_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_minsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_max_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_maxpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_max_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_maxsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_and_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_andpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_andnot_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_andnpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_or_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_orpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_xor_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_xorpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpeq_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpeqpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmplt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpltpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmple_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmplepd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpgt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpgtpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpge_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpgepd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpneq_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpneqpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpnlt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnltpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpnle_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnlepd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpngt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpngtpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpnge_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpngepd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpord_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpordpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpunord_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpunordpd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpeq_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpeqsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmplt_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpltsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmple_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmplesd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpgt_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
					 (__v2df)
					 __builtin_ia32_cmpltsd ((__v2df) __B,
								 (__v2df)
								 __A));
}

static __inline __m128d
_mm_cmpge_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
					 (__v2df)
					 __builtin_ia32_cmplesd ((__v2df) __B,
								 (__v2df)
								 __A));
}

static __inline __m128d
_mm_cmpneq_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpneqsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpnlt_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnltsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpnle_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnlesd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpngt_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
					 (__v2df)
					 __builtin_ia32_cmpnltsd ((__v2df) __B,
								  (__v2df)
								  __A));
}

static __inline __m128d
_mm_cmpnge_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
					 (__v2df)
					 __builtin_ia32_cmpnlesd ((__v2df) __B,
								  (__v2df)
								  __A));
}

static __inline __m128d
_mm_cmpord_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpordsd ((__v2df)__A, (__v2df)__B);
}

static __inline __m128d
_mm_cmpunord_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpunordsd ((__v2df)__A, (__v2df)__B);
}

static __inline int
_mm_comieq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdeq ((__v2df)__A, (__v2df)__B);
}