fvec.h

C语言库函数的原型,有用的拿去

{
        F32vec4 mask = _mm_cmpeq_ps(a,b);
        return( (mask & c) | F32vec4((_mm_andnot_ps(mask,d))));
}
*/

#define Fvec32s4_SELECT(op) \
inline F32vec4 select_##op (const F32vec4 &a, const F32vec4 &b, const F32vec4 &c, const F32vec4 &d)        \
{                                                                                                                               \
        F32vec4 mask = _mm_cmp##op##_ps(a,b);                                           \
        return( (mask & c) | F32vec4((_mm_andnot_ps(mask,d)))); \
}
Fvec32s4_SELECT(eq)                     /* generates select_eq(a,b) */
Fvec32s4_SELECT(lt)                     /* generates select_lt(a,b) */
Fvec32s4_SELECT(le)                     /* generates select_le(a,b) */
Fvec32s4_SELECT(gt)                     /* generates select_gt(a,b) */
Fvec32s4_SELECT(ge)                     /* generates select_ge(a,b) */
Fvec32s4_SELECT(neq)            /* generates select_neq(a,b) */
Fvec32s4_SELECT(nlt)            /* generates select_nlt(a,b) */
Fvec32s4_SELECT(nle)            /* generates select_nle(a,b) */
Fvec32s4_SELECT(ngt)            /* generates select_ngt(a,b) */
Fvec32s4_SELECT(nge)            /* generates select_nge(a,b) */
#undef Fvec32s4_SELECT


/* Streaming SIMD Extensions Integer Intrinsics */

/* Max and Min */
inline Is16vec4 simd_max(const Is16vec4 &a, const Is16vec4 &b)          { return _m_pmaxsw(a,b);}
inline Is16vec4 simd_min(const Is16vec4 &a, const Is16vec4 &b)          { return _m_pminsw(a,b);}
inline Iu8vec8 simd_max(const Iu8vec8 &a, const Iu8vec8 &b)                     { return _m_pmaxub(a,b);}
inline Iu8vec8 simd_min(const Iu8vec8 &a, const Iu8vec8 &b)                     { return _m_pminub(a,b);}

/* Average */
inline Iu16vec4 simd_avg(const Iu16vec4 &a, const Iu16vec4 &b)          { return _m_pavgw(a,b); }
inline Iu8vec8 simd_avg(const Iu8vec8 &a, const Iu8vec8 &b)                     { return _m_pavgb(a,b); }

/* Move ByteMask To Int: returns mask formed from most sig bits of each vec of a */
inline int move_mask(const I8vec8 &a)                                                           { return _m_pmovmskb(a);}

/* Packed Multiply High Unsigned */
inline Iu16vec4 mul_high(const Iu16vec4 &a, const Iu16vec4 &b)          { return _m_pmulhuw(a,b); }

/* Byte Mask Write: Write bytes if most significant bit in each corresponding byte is set */
inline void mask_move(const I8vec8 &a, const I8vec8 &b, char *addr)     { _m_maskmovq(a, b, addr); }

/* Data Motion: Store Non Temporal */
inline void store_nta(__m64 *p, M64 &a) { _mm_stream_pi(p,a); }

/* Conversions between ivec <-> fvec */

/* Convert first element of F32vec4 to int with truncation */
inline int F32vec4ToInt(const F32vec4 &a)
{

        return _mm_cvtt_ss2si(a);

}

/* Convert two lower SP FP values of a to Is32vec2 with truncation */
inline Is32vec2 F32vec4ToIs32vec2 (const F32vec4 &a)
{

        __m64 result;
        result = _mm_cvtt_ps2pi(a);
        return Is32vec2(result);

}

/* Convert the 32-bit int i to an SP FP value; the upper three SP FP values are passed through from a. */
inline F32vec4 IntToF32vec4(const F32vec4 &a, int i)
{

        __m128 result;
        result = _mm_cvt_si2ss(a,i);
        return F32vec4(result);

}

/* Convert the two 32-bit integer values in b to two SP FP values; the upper two SP FP values are passed from a. */
inline F32vec4 Is32vec2ToF32vec4(const F32vec4 &a, const Is32vec2 &b)
{

        __m128 result;
        result = _mm_cvt_pi2ps(a,b);
        return F32vec4(result);
}

class F32vec1
{
protected:
         __m128 vec;
public:

        /* Constructors: 1 float */
        F32vec1() {}

        F32vec1(int i)          { vec = _mm_cvt_si2ss(vec,i);};

        /* Initialize each of 4 SP FPs with same float */
        explicit F32vec1(float f)       { vec = _mm_set_ss(f); }

        /* Initialize each of 4 SP FPs with same float */
        explicit F32vec1(double d)      { vec = _mm_set_ss((float) d); }

        /* initialize with __m128 data type */
        F32vec1(__m128 m)       { vec = m; }

        /* Conversion functions */
        operator  __m128() const        { return vec; }         /* Convert to float */

        /* Logical Operators */
        friend F32vec1 operator &(const F32vec1 &a, const F32vec1 &b) { return _mm_and_ps(a,b); }
        friend F32vec1 operator |(const F32vec1 &a, const F32vec1 &b) { return _mm_or_ps(a,b); }
        friend F32vec1 operator ^(const F32vec1 &a, const F32vec1 &b) { return _mm_xor_ps(a,b); }

        /* Arithmetic Operators */
        friend F32vec1 operator +(const F32vec1 &a, const F32vec1 &b) { return _mm_add_ss(a,b); }
        friend F32vec1 operator -(const F32vec1 &a, const F32vec1 &b) { return _mm_sub_ss(a,b); }
        friend F32vec1 operator *(const F32vec1 &a, const F32vec1 &b) { return _mm_mul_ss(a,b); }
        friend F32vec1 operator /(const F32vec1 &a, const F32vec1 &b) { return _mm_div_ss(a,b); }

        F32vec1& operator +=(F32vec1 &a) { return *this = _mm_add_ss(vec,a); }
        F32vec1& operator -=(F32vec1 &a) { return *this = _mm_sub_ss(vec,a); }
        F32vec1& operator *=(F32vec1 &a) { return *this = _mm_mul_ss(vec,a); }
        F32vec1& operator /=(F32vec1 &a) { return *this = _mm_div_ss(vec,a); }
        F32vec1& operator &=(F32vec1 &a) { return *this = _mm_and_ps(vec,a); }
        F32vec1& operator |=(F32vec1 &a) { return *this = _mm_or_ps(vec,a); }
        F32vec1& operator ^=(F32vec1 &a) { return *this = _mm_xor_ps(vec,a); }


        /* Square Root */
        friend F32vec1 sqrt(const F32vec1 &a)           { return _mm_sqrt_ss(a); }
        /* Reciprocal */
        friend F32vec1 rcp(const F32vec1 &a)            { return _mm_rcp_ss(a); }
        /* Reciprocal Square Root */
        friend F32vec1 rsqrt(const F32vec1 &a)          { return _mm_rsqrt_ss(a); }

        /* NewtonRaphson Reciprocal
           [2 * rcpss(x) - (x * rcpss(x) * rcpss(x))] */
        friend F32vec1 rcp_nr(const F32vec1 &a)
        {
                F32vec1 Ra0 = _mm_rcp_ss(a);
                return _mm_sub_ss(_mm_add_ss(Ra0, Ra0), _mm_mul_ss(_mm_mul_ss(Ra0, a), Ra0));
        }

        /*      NewtonRaphson Reciprocal Square Root
                0.5 * rsqrtss * (3 - x * rsqrtss(x) * rsqrtss(x)) */
#pragma warning(push)
#pragma warning(disable : 4640)
        friend F32vec1 rsqrt_nr(const F32vec1 &a)
        {
                static const F32vec1 fvecf0pt5(0.5f);
                static const F32vec1 fvecf3pt0(3.0f);
                F32vec1 Ra0 = _mm_rsqrt_ss(a);
                return (fvecf0pt5 * Ra0) * (fvecf3pt0 - (a * Ra0) * Ra0);
        }
#pragma warning(pop)

        /* Compares: Mask is returned  */
        /* Macros expand to all compare intrinsics.  Example:
        friend F32vec1 cmpeq(const F32vec1 &a, const F32vec1 &b)
        { return _mm_cmpeq_ss(a,b);} */
        #define Fvec32s1_COMP(op) \
        friend F32vec1 cmp##op (const F32vec1 &a, const F32vec1 &b) { return _mm_cmp##op##_ss(a,b); }
                Fvec32s1_COMP(eq)                                       /* expanded to cmpeq(a,b) */
                Fvec32s1_COMP(lt)                                       /* expanded to cmplt(a,b) */
                Fvec32s1_COMP(le)                                       /* expanded to cmple(a,b) */
                Fvec32s1_COMP(gt)                                       /* expanded to cmpgt(a,b) */
                Fvec32s1_COMP(ge)                                       /* expanded to cmpge(a,b) */
                Fvec32s1_COMP(neq)                                      /* expanded to cmpneq(a,b) */
                Fvec32s1_COMP(nlt)                                      /* expanded to cmpnlt(a,b) */
                Fvec32s1_COMP(nle)                                      /* expanded to cmpnle(a,b) */
                Fvec32s1_COMP(ngt)                                      /* expanded to cmpngt(a,b) */
                Fvec32s1_COMP(nge)                                      /* expanded to cmpnge(a,b) */
        #undef Fvec32s1_COMP

        /* Min and Max */
        friend F32vec1 simd_min(const F32vec1 &a, const F32vec1 &b) { return _mm_min_ss(a,b); }
        friend F32vec1 simd_max(const F32vec1 &a, const F32vec1 &b) { return _mm_max_ss(a,b); }

        /* Debug Features */
#if defined (_ENABLE_VEC_DEBUG)
        /* Output */
        friend std::ostream & operator<<(std::ostream & os, const F32vec1 &a)
        {
        /* To use: cout << "Elements of F32vec1 fvec are: " << fvec; */
          float *fp = (float*)&a;
                os << "float:" << *fp;
                return os;
        }
#endif  /* defined (_ENABLE_VEC_DEBUG) */

};

                                                /* Conditional Selects:*/
/*(a OP b)? c : d; where OP is any compare operator
Macros expand to conditional selects which use all compare intrinsics.
Example:
friend F32vec1 select_eq(const F32vec1 &a, const F32vec1 &b, const F32vec1 &c, const F32vec1 &d)
{
        F32vec1 mask = _mm_cmpeq_ss(a,b);
        return( (mask & c) | F32vec1((_mm_andnot_ps(mask,d))));
}
*/

#define Fvec32s1_SELECT(op) \
inline F32vec1 select_##op (const F32vec1 &a, const F32vec1 &b, const F32vec1 &c, const F32vec1 &d)        \
{                                                                                                          \
        F32vec1 mask = _mm_cmp##op##_ss(a,b);                                                              \
        return( (mask & c) | F32vec1((_mm_andnot_ps(mask,d))));                                            \
}
Fvec32s1_SELECT(eq)                     /* generates select_eq(a,b) */
Fvec32s1_SELECT(lt)                     /* generates select_lt(a,b) */
Fvec32s1_SELECT(le)                     /* generates select_le(a,b) */
Fvec32s1_SELECT(gt)                     /* generates select_gt(a,b) */
Fvec32s1_SELECT(ge)                     /* generates select_ge(a,b) */
Fvec32s1_SELECT(neq)            /* generates select_neq(a,b) */
Fvec32s1_SELECT(nlt)            /* generates select_nlt(a,b) */
Fvec32s1_SELECT(nle)            /* generates select_nle(a,b) */
Fvec32s1_SELECT(ngt)            /* generates select_ngt(a,b) */
Fvec32s1_SELECT(nge)            /* generates select_nge(a,b) */
#undef Fvec32s1_SELECT

/* Conversions between ivec <-> fvec */

/* Convert F32vec1 to int */
inline int F32vec1ToInt(const F32vec1 &a)
{
        return _mm_cvtt_ss2si(a);
}



#pragma pack(pop) /* 16-B aligned */

#ifdef _MSC_VER
#pragma pack(pop)
#endif  /* _MSC_VER */

#endif  /* defined (_M_CEE_PURE) */

#endif  /* RC_INVOKED */
#endif  /* _FVEC_H_INCLUDED */