#ifdef YES_AVX
void matvec_YMM(double* a, double* x, double* y, int n, int lb)
{
    int i, j;
    __m256d rx0, ra0, ra1, ra2, ra3, ry0, ry1, ry2, ry3;
    double *ptr_x, *ptr_a;
    __declspec(align(16)) double buf0[4], buf1[4], buf2[4], buf3[4];
    memset((void *)y, 0, n * sizeof(double));
    ptr_a = a;
    for (i = 0; i < n; i += 4)
    {
        ry0 = ry1 = ry2 = ry3 = _mm256_setzero_pd();
        ptr_x = x;


        for (j = 0; j < n; j += 16)
        {
            _mm_prefetch((const char *)(ptr_x + 16), _MM_HINT_T0);
            _mm_prefetch((const char *)(ptr_x + 24), _MM_HINT_T0);

            _mm_prefetch((const char *)(ptr_a + 16), _MM_HINT_NTA);
            _mm_prefetch((const char *)(ptr_a + 24), _MM_HINT_NTA);

                    _mm_prefetch((const char *)(ptr_a +n+ 16), _MM_HINT_NTA);
            _mm_prefetch((const char *)(ptr_a +n+ 24), _MM_HINT_NTA);

                    _mm_prefetch((const char *)(ptr_a +2*n+ 16), _MM_HINT_NTA);
            _mm_prefetch((const char *)(ptr_a +2*n+ 24), _MM_HINT_NTA);

                    _mm_prefetch((const char *)(ptr_a +3*n+ 16), _MM_HINT_NTA);
            _mm_prefetch((const char *)(ptr_a +3*n+ 24), _MM_HINT_NTA);

        //--------------------------0
            rx0 = _mm256_load_pd(ptr_x);
            ra0 = _mm256_load_pd(ptr_a);
            ra1 = _mm256_load_pd(ptr_a + n);
            ra2 = _mm256_load_pd(ptr_a + 2 * n);
            ra3 = _mm256_load_pd(ptr_a + 3 * n);

            ra0 = _mm256_mul_pd(ra0, rx0);
            ra1 = _mm256_mul_pd(ra1, rx0);
            ra2 = _mm256_mul_pd(ra2, rx0);
            ra3 = _mm256_mul_pd(ra3, rx0);

            ry0 = _mm256_add_pd(ry0, ra0);
            ry1 = _mm256_add_pd(ry1, ra1);
            ry2 = _mm256_add_pd(ry2, ra2);
            ry3 = _mm256_add_pd(ry3, ra3);

            //-------256----------------1
            rx0 = _mm256_load_pd(ptr_x + 4);
            ra0 = _mm256_load_pd(ptr_a + 4);
            ra1 = _mm256_load_pd(ptr_a + n + 4);
            ra2 = _mm256_load_pd(ptr_a + 2 * n + 4);
            ra3 = _mm256_load_pd(ptr_a + 3 * n + 4);

            ra0 = _mm256_mul_pd(ra0, rx0);
            ra1 = _mm256_mul_pd(ra1, rx0);
            ra2 = _mm256_mul_pd(ra2, rx0);
            ra3 = _mm256_mul_pd(ra3, rx0);

            ry0 = _mm256_add_pd(ry0, ra0);
            ry1 = _mm256_add_pd(ry1, ra1);
            ry2 = _mm256_add_pd(ry2, ra2);
            ry3 = _mm256_add_pd(ry3, ra3);

            //-------256----------------2
            rx0 = _mm256_load_pd(ptr_x + 8);
            ra0 = _mm256_load_pd(ptr_a + 8);
            ra1 = _mm256_load_pd(ptr_a + n + 8);
            ra2 = _mm256_load_pd(ptr_a + 2 * n + 8);
            ra3 = _mm256_load_pd(ptr_a + 3 * n + 8);

            ra0 = _mm256_mul_pd(ra0, rx0);
            ra1 = _mm256_mul_pd(ra1, rx0);
            ra2 = _mm256_mul_pd(ra2, rx0);
            ra3 = _mm256_mul_pd(ra3, rx0);

            ry0 = _mm256_add_pd(ry0, ra0);
            ry1 = _mm256_add_pd(ry1, ra1);
            ry2 = _mm256_add_pd(ry2, ra2);
            ry3 = _mm256_add_pd(ry3, ra3);

            //-------256----------------3
            rx0 = _mm256_load_pd(ptr_x + 12);
            ra0 = _mm256_load_pd(ptr_a + 12);
            ra1 = _mm256_load_pd(ptr_a + n + 12);
            ra2 = _mm256_load_pd(ptr_a + 2 * n + 12);
            ra3 = _mm256_load_pd(ptr_a + 3 * n + 12);

            ra0 = _mm256_mul_pd(ra0, rx0);
            ra1 = _mm256_mul_pd(ra1, rx0);
            ra2 = _mm256_mul_pd(ra2, rx0);
            ra3 = _mm256_mul_pd(ra3, rx0);

            ry0 = _mm256_add_pd(ry0, ra0);
            ry1 = _mm256_add_pd(ry1, ra1);
            ry2 = _mm256_add_pd(ry2, ra2);
            ry3 = _mm256_add_pd(ry3, ra3);

            ptr_a += 16;
            ptr_x += 16;
        }

        ptr_a += 3 * n;

        _mm256_store_pd(buf0, ry0);
        _mm256_store_pd(buf1, ry1);
        _mm256_store_pd(buf2, ry2);
        _mm256_store_pd(buf3, ry3);

        y[i] = buf0[0] + buf0[1] + buf0[2] + buf0[3];
        y[i + 1] = buf1[0] + buf1[1] + buf1[2] + buf1[3];
        y[i + 2] = buf2[0] + buf2[1] + buf2[2] + buf2[3];
        y[i + 3] = buf3[0] + buf3[1] + buf3[2] + buf3[3];
    }
}
#endif