SIMD class

gistfile1.cpp

PRE_ALIGN(128) struct __vec8_d { 
    __vec8_d() { }

    FORCEINLINE __vec8_d(const double v0) {
        u.v0 = _mm_set_pd(v0, v0);
        u.v1 = _mm_set_pd(v0, v0);
        u.v2 = _mm_set_pd(v0, v0);
        u.v3 = _mm_set_pd(v0, v0);
    }

    FORCEINLINE __vec8_d(const double v0,  const double v1,  const double v2,  const double v3, 
              const double v4,  const double v5,  const double v6,  const double v7) {
        u.v0 = _mm_set_pd(v0, v1);
        u.v1 = _mm_set_pd(v2, v3);      
        u.v2 = _mm_set_pd(v4, v5);      
        u.v3 = _mm_set_pd(v6, v7);      
    }

    FORCEINLINE __vec8_d(const __m128d w0, const __m128d w1, const __m128d w2, const __m128d w3) {
        u.v0 = w0;
        u.v1 = w1;
        u.v2 = w2;
        u.v3 = w3;
    }

    FORCEINLINE __vec8_d(const __vec8_d& rhs) {
        u.v0 = rhs.u.v0;
        u.v1 = rhs.u.v1;
        u.v2 = rhs.u.v2;
        u.v3 = rhs.u.v3;
    }

    FORCEINLINE __vec8_d operator=(const __vec8_d& rhs){
        u.v0 = rhs.u.v0;
        u.v1 = rhs.u.v1;
        u.v2 = rhs.u.v2;
        u.v3 = rhs.u.v3;
        return (*this);
    }

    FORCEINLINE __vec8_d operator+(const __vec8_d& rhs) const {
        return __vec8_d(_mm_add_pd(u.v0, rhs.u.v0),
                         _mm_add_pd(u.v1, rhs.u.v1),
                         _mm_add_pd(u.v2, rhs.u.v2),
                         _mm_add_pd(u.v3, rhs.u.v3));
    }

    FORCEINLINE __vec8_d operator-(const __vec8_d& rhs) const {
        return __vec8_d(_mm_sub_pd(u.v0, rhs.u.v0),
                         _mm_sub_pd(u.v1, rhs.u.v1),
                         _mm_sub_pd(u.v2, rhs.u.v2),
                         _mm_sub_pd(u.v3, rhs.u.v3));
    }

    FORCEINLINE __vec8_d operator*(const __vec8_d& rhs) const {
        return __vec8_d(_mm_mul_pd(u.v0, rhs.u.v0),
                         _mm_mul_pd(u.v1, rhs.u.v1),
                         _mm_mul_pd(u.v2, rhs.u.v2),
                         _mm_mul_pd(u.v3, rhs.u.v3));
    }

    union {
        double v[8];
        struct { __m128d v0, v1, v2, v3; };
    } u;

} POST_ALIGN(128);