#ifndef _MMINTRIN_H_INCLUDED
#define _MMINTRIN_H_INCLUDED
#ifndef __MMX__
# error "MMX instruction set not enabled"
#else
typedef int __m64 __attribute__ ((__mode__ (__V2SI__)));
typedef int __v2si __attribute__ ((__mode__ (__V2SI__)));
typedef int __v4hi __attribute__ ((__mode__ (__V4HI__)));
typedef int __v8qi __attribute__ ((__mode__ (__V8QI__)));
static __inline void
_mm_empty (void)
{
__builtin_ia32_emms ();
}
static __inline __m64
_mm_cvtsi32_si64 (int __i)
{
long long __tmp = (unsigned int)__i;
return (__m64) __tmp;
}
#ifdef __x86_64__
static __inline __m64
_mm_cvtsi64x_si64 (long long __i)
{
return (__m64) __i;
}
static __inline __m64
_mm_set_pi64x (long long __i)
{
return (__m64) __i;
}
#endif
static __inline int
_mm_cvtsi64_si32 (__m64 __i)
{
long long __tmp = (long long)__i;
return __tmp;
}
#ifdef __x86_64__
static __inline long long
_mm_cvtsi64_si64x (__m64 __i)
{
return (long long)__i;
}
#endif
static __inline __m64
_mm_packs_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_packsswb ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_packs_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_packssdw ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_packs_pu16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_packuswb ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpckhbw ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_unpackhi_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpckhwd ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_unpackhi_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpckhdq ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_unpacklo_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpcklbw ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_unpacklo_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpcklwd ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_unpacklo_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_punpckldq ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_add_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_add_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_add_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_add_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddq ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_adds_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddsb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_adds_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddsw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_adds_pu8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddusb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_adds_pu16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddusw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_sub_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_sub_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_sub_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_sub_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubq ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_subs_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubsb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_subs_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubsw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_subs_pu8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubusb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_subs_pu16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubusw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_madd_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pmaddwd ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_mulhi_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pmulhw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_mullo_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pmullw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_sll_pi16 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
_mm_slli_pi16 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psllw ((__v4hi)__m, __count);
}
static __inline __m64
_mm_sll_pi32 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_pslld ((__v2si)__m, (long long)__count);
}
static __inline __m64
_mm_slli_pi32 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_pslld ((__v2si)__m, __count);
}
static __inline __m64
_mm_sll_si64 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_slli_si64 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_sra_pi16 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
_mm_srai_pi16 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psraw ((__v4hi)__m, __count);
}
static __inline __m64
_mm_sra_pi32 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psrad ((__v2si)__m, (long long)__count);
}
static __inline __m64
_mm_srai_pi32 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psrad ((__v2si)__m, __count);
}
static __inline __m64
_mm_srl_pi16 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
_mm_srli_pi16 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, __count);
}
static __inline __m64
_mm_srl_pi32 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psrld ((__v2si)__m, (long long)__count);
}
static __inline __m64
_mm_srli_pi32 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psrld ((__v2si)__m, __count);
}
static __inline __m64
_mm_srl_si64 (__m64 __m, __m64 __count)
{
return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_srli_si64 (__m64 __m, int __count)
{
return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_and_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pand ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_andnot_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pandn ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_or_si64 (__m64 __m1, __m64 __m2)
{
return (__m64)__builtin_ia32_por ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_xor_si64 (__m64 __m1, __m64 __m2)
{
return (__m64)__builtin_ia32_pxor ((long long)__m1, (long long)__m2);
}
static __inline __m64
_mm_cmpeq_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpeqb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_cmpgt_pi8 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpgtb ((__v8qi)__m1, (__v8qi)__m2);
}
static __inline __m64
_mm_cmpeq_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpeqw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_cmpgt_pi16 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpgtw ((__v4hi)__m1, (__v4hi)__m2);
}
static __inline __m64
_mm_cmpeq_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpeqd ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_cmpgt_pi32 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_pcmpgtd ((__v2si)__m1, (__v2si)__m2);
}
static __inline __m64
_mm_setzero_si64 (void)
{
return (__m64)__builtin_ia32_mmx_zero ();
}
static __inline __m64
_mm_set_pi32 (int __i1, int __i0)
{
union {
__m64 __q;
struct {
unsigned int __i0;
unsigned int __i1;
} __s;
} __u;
__u.__s.__i0 = __i0;
__u.__s.__i1 = __i1;
return __u.__q;
}
static __inline __m64
_mm_set_pi16 (short __w3, short __w2, short __w1, short __w0)
{
unsigned int __i1 = (unsigned short)__w3 << 16 | (unsigned short)__w2;
unsigned int __i0 = (unsigned short)__w1 << 16 | (unsigned short)__w0;
return _mm_set_pi32 (__i1, __i0);
}
static __inline __m64
_mm_set_pi8 (char __b7, char __b6, char __b5, char __b4,
char __b3, char __b2, char __b1, char __b0)
{
unsigned int __i1, __i0;
__i1 = (unsigned char)__b7;
__i1 = __i1 << 8 | (unsigned char)__b6;
__i1 = __i1 << 8 | (unsigned char)__b5;
__i1 = __i1 << 8 | (unsigned char)__b4;
__i0 = (unsigned char)__b3;
__i0 = __i0 << 8 | (unsigned char)__b2;
__i0 = __i0 << 8 | (unsigned char)__b1;
__i0 = __i0 << 8 | (unsigned char)__b0;
return _mm_set_pi32 (__i1, __i0);
}
static __inline __m64
_mm_setr_pi32 (int __i0, int __i1)
{
return _mm_set_pi32 (__i1, __i0);
}
static __inline __m64
_mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3)
{
return _mm_set_pi16 (__w3, __w2, __w1, __w0);
}
static __inline __m64
_mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3,
char __b4, char __b5, char __b6, char __b7)
{
return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0);
}
static __inline __m64
_mm_set1_pi32 (int __i)
{
return _mm_set_pi32 (__i, __i);
}
static __inline __m64
_mm_set1_pi16 (short __w)
{
unsigned int __i = (unsigned short)__w << 16 | (unsigned short)__w;
return _mm_set1_pi32 (__i);
}
static __inline __m64
_mm_set1_pi8 (char __b)
{
unsigned int __w = (unsigned char)__b << 8 | (unsigned char)__b;
unsigned int __i = __w << 16 | __w;
return _mm_set1_pi32 (__i);
}
#endif
#endif