#ifndef _PMMINTRIN_H_INCLUDED
#define _PMMINTRIN_H_INCLUDED
#ifdef __SSE3__
#include <xmmintrin.h>
#include <emmintrin.h>
#define _MM_DENORMALS_ZERO_MASK 0x0040
#define _MM_DENORMALS_ZERO_ON 0x0040
#define _MM_DENORMALS_ZERO_OFF 0x0000
#define _MM_SET_DENORMALS_ZERO_MODE(mode) \
_mm_setcsr ((_mm_getcsr () & ~_MM_DENORMALS_ZERO_MASK) | (mode))
#define _MM_GET_DENORMALS_ZERO_MODE() \
(_mm_getcsr() & _MM_DENORMALS_ZERO_MASK)
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_addsub_ps (__m128 __X, __m128 __Y)
{
return (__m128) __builtin_ia32_addsubps ((__v4sf)__X, (__v4sf)__Y);
}
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_hadd_ps (__m128 __X, __m128 __Y)
{
return (__m128) __builtin_ia32_haddps ((__v4sf)__X, (__v4sf)__Y);
}
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_hsub_ps (__m128 __X, __m128 __Y)
{
return (__m128) __builtin_ia32_hsubps ((__v4sf)__X, (__v4sf)__Y);
}
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_movehdup_ps (__m128 __X)
{
return (__m128) __builtin_ia32_movshdup ((__v4sf)__X);
}
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
_mm_moveldup_ps (__m128 __X)
{
return (__m128) __builtin_ia32_movsldup ((__v4sf)__X);
}
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_addsub_pd (__m128d __X, __m128d __Y)
{
return (__m128d) __builtin_ia32_addsubpd ((__v2df)__X, (__v2df)__Y);
}
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_hadd_pd (__m128d __X, __m128d __Y)
{
return (__m128d) __builtin_ia32_haddpd ((__v2df)__X, (__v2df)__Y);
}
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_hsub_pd (__m128d __X, __m128d __Y)
{
return (__m128d) __builtin_ia32_hsubpd ((__v2df)__X, (__v2df)__Y);
}
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_loaddup_pd (double const *__P)
{
return _mm_load1_pd (__P);
}
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
_mm_movedup_pd (__m128d __X)
{
return _mm_shuffle_pd (__X, __X, _MM_SHUFFLE2 (0,0));
}
static __inline __m128i __attribute__((__always_inline__, __nodebug__))
_mm_lddqu_si128 (__m128i const *__P)
{
return (__m128i) __builtin_ia32_lddqu ((char const *)__P);
}
#if 0
static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_monitor (void const * __P, unsigned int __E, unsigned int __H)
{
__builtin_ia32_monitor (__P, __E, __H);
}
static __inline void __attribute__((__always_inline__, __nodebug__))
_mm_mwait (unsigned int __E, unsigned int __H)
{
__builtin_ia32_mwait (__E, __H);
}
#else
#define _mm_monitor(P, E, H) __builtin_ia32_monitor ((P), (E), (H))
#define _mm_mwait(E, H) __builtin_ia32_mwait ((E), (H))
#endif
#endif
#endif