xmmintrin.h

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

{
  return (__m64) __builtin_ia32_cvttps2pi ((__v4sf) __A);
}

/* Convert B to a SPFP value and insert it as element zero in A.  */
static __inline __m128
_mm_cvtsi32_ss (__m128 __A, int __B)
{
  return (__m128) __builtin_ia32_cvtsi2ss ((__v4sf) __A, __B);
}

#ifdef __x86_64__
/* Convert B to a SPFP value and insert it as element zero in A.  */
static __inline __m128
_mm_cvtsi64x_ss (__m128 __A, long long __B)
{
  return (__m128) __builtin_ia32_cvtsi642ss ((__v4sf) __A, __B);
}
#endif

/* Convert the two 32-bit values in B to SPFP form and insert them
   as the two lower elements in A.  */
static __inline __m128
_mm_cvtpi32_ps (__m128 __A, __m64 __B)
{
  return (__m128) __builtin_ia32_cvtpi2ps ((__v4sf) __A, (__v2si)__B);
}

/* Convert the four signed 16-bit values in A to SPFP form.  */
static __inline __m128
_mm_cvtpi16_ps (__m64 __A)
{
  __v4hi __sign;
  __v2si __hisi, __losi;
  __v4sf __r;

  /* This comparison against zero gives us a mask that can be used to
     fill in the missing sign bits in the unpack operations below, so
     that we get signed values after unpacking.  */
  __sign = (__v4hi) __builtin_ia32_mmx_zero ();
  __sign = __builtin_ia32_pcmpgtw (__sign, (__v4hi)__A);

  /* Convert the four words to doublewords.  */
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __sign);
  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __sign);

  /* Convert the doublewords to floating point two at a time.  */
  __r = (__v4sf) __builtin_ia32_setzerops ();
  __r = __builtin_ia32_cvtpi2ps (__r, __hisi);
  __r = __builtin_ia32_movlhps (__r, __r);
  __r = __builtin_ia32_cvtpi2ps (__r, __losi);

  return (__m128) __r;
}

/* Convert the four unsigned 16-bit values in A to SPFP form.  */
static __inline __m128
_mm_cvtpu16_ps (__m64 __A)
{
  __v4hi __zero = (__v4hi) __builtin_ia32_mmx_zero ();
  __v2si __hisi, __losi;
  __v4sf __r;

  /* Convert the four words to doublewords.  */
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __zero);
  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __zero);

  /* Convert the doublewords to floating point two at a time.  */
  __r = (__v4sf) __builtin_ia32_setzerops ();
  __r = __builtin_ia32_cvtpi2ps (__r, __hisi);
  __r = __builtin_ia32_movlhps (__r, __r);
  __r = __builtin_ia32_cvtpi2ps (__r, __losi);

  return (__m128) __r;
}

/* Convert the low four signed 8-bit values in A to SPFP form.  */
static __inline __m128
_mm_cvtpi8_ps (__m64 __A)
{
  __v8qi __sign;

  /* This comparison against zero gives us a mask that can be used to
     fill in the missing sign bits in the unpack operations below, so
     that we get signed values after unpacking.  */
  __sign = (__v8qi) __builtin_ia32_mmx_zero ();
  __sign = __builtin_ia32_pcmpgtb (__sign, (__v8qi)__A);

  /* Convert the four low bytes to words.  */
  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __sign);

  return _mm_cvtpi16_ps(__A);
}

/* Convert the low four unsigned 8-bit values in A to SPFP form.  */
static __inline __m128
_mm_cvtpu8_ps(__m64 __A)
{
  __v8qi __zero = (__v8qi) __builtin_ia32_mmx_zero ();
  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __zero);
  return _mm_cvtpu16_ps(__A);
}

/* Convert the four signed 32-bit values in A and B to SPFP form.  */
static __inline __m128
_mm_cvtpi32x2_ps(__m64 __A, __m64 __B)
{
  __v4sf __zero = (__v4sf) __builtin_ia32_setzerops ();
  __v4sf __sfa = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__A);
  __v4sf __sfb = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__B);
  return (__m128) __builtin_ia32_movlhps (__sfa, __sfb);
}

/* Convert the four SPFP values in A to four signed 16-bit integers.  */
static __inline __m64
_mm_cvtps_pi16(__m128 __A)
{
  __v4sf __hisf = (__v4sf)__A;
  __v4sf __losf = __builtin_ia32_movhlps (__hisf, __hisf);
  __v2si __hisi = __builtin_ia32_cvtps2pi (__hisf);
  __v2si __losi = __builtin_ia32_cvtps2pi (__losf);
  return (__m64) __builtin_ia32_packssdw (__hisi, __losi);
}

/* Convert the four SPFP values in A to four signed 8-bit integers.  */
static __inline __m64
_mm_cvtps_pi8(__m128 __A)
{
  __v4hi __tmp = (__v4hi) _mm_cvtps_pi16 (__A);
  __v4hi __zero = (__v4hi) __builtin_ia32_mmx_zero ();
  return (__m64) __builtin_ia32_packsswb (__tmp, __zero);
}

/* Selects four specific SPFP values from A and B based on MASK.  */
#if 0
static __inline __m128
_mm_shuffle_ps (__m128 __A, __m128 __B, int __mask)
{
  return (__m128) __builtin_ia32_shufps ((__v4sf)__A, (__v4sf)__B, __mask);
}
#else
#define _mm_shuffle_ps(A, B, MASK) \
 ((__m128) __builtin_ia32_shufps ((__v4sf)(A), (__v4sf)(B), (MASK)))
#endif


/* Selects and interleaves the upper two SPFP values from A and B.  */
static __inline __m128
_mm_unpackhi_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpckhps ((__v4sf)__A, (__v4sf)__B);
}

/* Selects and interleaves the lower two SPFP values from A and B.  */
static __inline __m128
_mm_unpacklo_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpcklps ((__v4sf)__A, (__v4sf)__B);
}

/* Sets the upper two SPFP values with 64-bits of data loaded from P;
   the lower two values are passed through from A.  */
static __inline __m128
_mm_loadh_pi (__m128 __A, __m64 const *__P)
{
  return (__m128) __builtin_ia32_loadhps ((__v4sf)__A, (__v2si *)__P);
}

/* Stores the upper two SPFP values of A into P.  */
static __inline void
_mm_storeh_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storehps ((__v2si *)__P, (__v4sf)__A);
}

/* Moves the upper two values of B into the lower two values of A.  */
static __inline __m128
_mm_movehl_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movhlps ((__v4sf)__A, (__v4sf)__B);
}

/* Moves the lower two values of B into the upper two values of A.  */
static __inline __m128
_mm_movelh_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movlhps ((__v4sf)__A, (__v4sf)__B);
}

/* Sets the lower two SPFP values with 64-bits of data loaded from P;
   the upper two values are passed through from A.  */
static __inline __m128
_mm_loadl_pi (__m128 __A, __m64 const *__P)
{
  return (__m128) __builtin_ia32_loadlps ((__v4sf)__A, (__v2si *)__P);
}

/* Stores the lower two SPFP values of A into P.  */
static __inline void
_mm_storel_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storelps ((__v2si *)__P, (__v4sf)__A);
}

/* Creates a 4-bit mask from the most significant bits of the SPFP values.  */
static __inline int
_mm_movemask_ps (__m128 __A)
{
  return __builtin_ia32_movmskps ((__v4sf)__A);
}

/* Return the contents of the control register.  */
static __inline unsigned int
_mm_getcsr (void)
{
  return __builtin_ia32_stmxcsr ();
}

/* Read exception bits from the control register.  */
static __inline unsigned int
_MM_GET_EXCEPTION_STATE (void)
{
  return _mm_getcsr() & _MM_EXCEPT_MASK;
}

static __inline unsigned int
_MM_GET_EXCEPTION_MASK (void)
{
  return _mm_getcsr() & _MM_MASK_MASK;
}

static __inline unsigned int
_MM_GET_ROUNDING_MODE (void)
{
  return _mm_getcsr() & _MM_ROUND_MASK;
}

static __inline unsigned int
_MM_GET_FLUSH_ZERO_MODE (void)
{
  return _mm_getcsr() & _MM_FLUSH_ZERO_MASK;
}

/* Set the control register to I.  */
static __inline void
_mm_setcsr (unsigned int __I)
{
  __builtin_ia32_ldmxcsr (__I);
}

/* Set exception bits in the control register.  */
static __inline void
_MM_SET_EXCEPTION_STATE(unsigned int __mask)
{
  _mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | __mask);
}

static __inline void
_MM_SET_EXCEPTION_MASK (unsigned int __mask)
{
  _mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | __mask);
}

static __inline void
_MM_SET_ROUNDING_MODE (unsigned int __mode)
{
  _mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | __mode);
}

static __inline void
_MM_SET_FLUSH_ZERO_MODE (unsigned int __mode)
{
  _mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | __mode);
}

/* Create a vector with element 0 as *P and the rest zero.  */
static __inline __m128
_mm_load_ss (float const *__P)
{
  return (__m128) __builtin_ia32_loadss (__P);
}

/* Create a vector with all four elements equal to *P.  */
static __inline __m128
_mm_load1_ps (float const *__P)
{
  __v4sf __tmp = __builtin_ia32_loadss (__P);
  return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,0,0,0));
}

static __inline __m128
_mm_load_ps1 (float const *__P)
{
  return _mm_load1_ps (__P);
}

/* Load four SPFP values from P.  The address must be 16-byte aligned.  */
static __inline __m128
_mm_load_ps (float const *__P)
{
  return (__m128) __builtin_ia32_loadaps (__P);
}

/* Load four SPFP values from P.  The address need not be 16-byte aligned.  */
static __inline __m128
_mm_loadu_ps (float const *__P)
{
  return (__m128) __builtin_ia32_loadups (__P);
}

/* Load four SPFP values in reverse order.  The address must be aligned.  */
static __inline __m128
_mm_loadr_ps (float const *__P)
{
  __v4sf __tmp = __builtin_ia32_loadaps (__P);
  return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,1,2,3));
}

/* Create a vector with element 0 as F and the rest zero.  */
static __inline __m128
_mm_set_ss (float __F)
{
  return (__m128) __builtin_ia32_loadss (&__F);
}

/* Create a vector with all four elements equal to F.  */
static __inline __m128
_mm_set1_ps (float __F)
{
  __v4sf __tmp = __builtin_ia32_loadss (&__F);
  return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,0,0,0));
}

static __inline __m128
_mm_set_ps1 (float __F)
{
  return _mm_set1_ps (__F);
}

/* Create the vector [Z Y X W].  */
static __inline __m128
_mm_set_ps (float __Z, float __Y, float __X, float __W)
{
  union {
    float __a[4];
    __m128 __v;
  } __u;

  __u.__a[0] = __W;
  __u.__a[1] = __X;
  __u.__a[2] = __Y;
  __u.__a[3] = __Z;

  return __u.__v;
}

/* Create the vector [W X Y Z].  */
static __inline __m128
_mm_setr_ps (float __Z, float __Y, float __X, float __W)
{
  return _mm_set_ps (__W, __X, __Y, __Z);
}

/* Create a vector of zeros.  */
static __inline __m128
_mm_setzero_ps (void)
{
  return (__m128) __builtin_ia32_setzerops ();
}

/* Stores the lower SPFP value.  */
static __inline void
_mm_store_ss (float *__P, __m128 __A)
{
  __builtin_ia32_storess (__P, (__v4sf)__A);
}

/* Store the lower SPFP value across four words.  */
static __inline void
_mm_store1_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,0,0,0));
  __builtin_ia32_storeaps (__P, __tmp);
}

static __inline void
_mm_store_ps1 (float *__P, __m128 __A)
{
  _mm_store1_ps (__P, __A);
}

/* Store four SPFP values.  The address must be 16-byte aligned.  */
static __inline void
_mm_store_ps (float *__P, __m128 __A)
{
  __builtin_ia32_storeaps (__P, (__v4sf)__A);
}

/* Store four SPFP values.  The address need not be 16-byte aligned.  */
static __inline void
_mm_storeu_ps (float *__P, __m128 __A)
{
  __builtin_ia32_storeups (__P, (__v4sf)__A);
}

/* Store four SPFP values in reverse order.  The address must be aligned.  */
static __inline void
_mm_storer_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,1,2,3));
  __builtin_ia32_storeaps (__P, __tmp);
}

/* Sets the low SPFP value of A from the low value of B.  */
static __inline __m128
_mm_move_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf)__A, (__v4sf)__B);
}

/* Extracts one of the four words of A.  The selector N must be immediate.  */
#if 0
static __inline int
_mm_extract_pi16 (__m64 __A, int __N)
{
  return __builtin_ia32_pextrw ((__v4hi)__A, __N);
}
#else
#define _mm_extract_pi16(A, N) \
  __builtin_ia32_pextrw ((__v4hi)(A), (N))
#endif

/* Inserts word D into one of four words of A.  The selector N must be
   immediate.  */
#if 0
static __inline __m64
_mm_insert_pi16 (__m64 __A, int __D, int __N)
{
  return (__m64)__builtin_ia32_pinsrw ((__v4hi)__A, __D, __N);
}
#else
#define _mm_insert_pi16(A, D, N) \
  ((__m64) __builtin_ia32_pinsrw ((__v4hi)(A), (D), (N)))
#endif

/* Compute the element-wise maximum of signed 16-bit values.  */
static __inline __m64
_mm_max_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxsw ((__v4hi)__A, (__v4hi)__B);
}

/* Compute the element-wise maximum of unsigned 8-bit values.  */
static __inline __m64
_mm_max_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxub ((__v8qi)__A, (__v8qi)__B);
}

/* Compute the element-wise minimum of signed 16-bit values.  */
static __inline __m64
_mm_min_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminsw ((__v4hi)__A, (__v4hi)__B);
}

/* Compute the element-wise minimum of unsigned 8-bit values.  */
static __inline __m64
_mm_min_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminub ((__v8qi)__A, (__v8qi)__B);
}

/* Create an 8-bit mask of the signs of 8-bit values.  */
static __inline int
_mm_movemask_pi8 (__m64 __A)
{
  return __builtin_ia32_pmovmskb ((__v8qi)__A);
}

/* Multiply four unsigned 16-bit values in A by four unsigned 16-bit values
   in B and produce the high 16 bits of the 32-bit results.  */
static __inline __m64
_mm_mulhi_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmulhuw ((__v4hi)__A, (__v4hi)__B);
}

/* Return a combination of the four 16-bit values in A.  The selector
   must be an immediate.  */
#if 0
static __inline __m64
_mm_shuffle_pi16 (__m64 __A, int __N)
{
  return (__m64) __builtin_ia32_pshufw ((__v4hi)__A, __N);
}
#else
#define _mm_shuffle_pi16(A, N) \
  ((__m64) __builtin_ia32_pshufw ((__v4hi)(A), (N)))
#endif

/* Conditionally store byte elements of A into P.  The high bit of each
   byte in the selector N determines whether the corresponding byte from
   A is stored.  */
static __inline void
_mm_maskmove_si64 (__m64 __A, __m64 __N, char *__P)
{
  __builtin_ia32_maskmovq ((__v8qi)__A, (__v8qi)__N, __P);
}

/* Compute the rounded averages of the unsigned 8-bit values in A and B.  */
static __inline __m64
_mm_avg_pu8 (__m64 __A, __m64 __B)