void ff_hevc_put_hevc_qpel_v4_8_sse (
                                    int16_t *dst, ptrdiff_t dststride,
                                    uint8_t *_src, ptrdiff_t _srcstride,
                                    int width, int height, int mx, int my) {
    int x, y;
    int shift = 8 - 8;
    const int8_t *filter_v = ff_hevc_qpel_filters[my - 1];
   __m128i x1, x2, x3, x4, r1, r2, r3, r4, c1, c2, c3, c4, t1, t2;
    const __m128i c0    = _mm_setzero_si128();
    uint8_t  *src       = (uint8_t*) _src;
    ptrdiff_t srcstride = _srcstride;
    c1 = _mm_unpacklo_epi8(_mm_set1_epi8(filter_v[0]), _mm_set1_epi8(filter_v[1]));
    c2 = _mm_unpacklo_epi8(_mm_set1_epi8(filter_v[2]), _mm_set1_epi8(filter_v[3]));
    c3 = _mm_unpacklo_epi8(_mm_set1_epi8(filter_v[4]), _mm_set1_epi8(filter_v[5]));
    c4 = _mm_unpacklo_epi8(_mm_set1_epi8(filter_v[6]), _mm_set1_epi8(filter_v[7]));
    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x += 4) {
            x1 = _mm_loadl_epi64((__m128i *) &src[x - 3 * srcstride]);
    x2 = _mm_loadl_epi64((__m128i *) &src[x - 2 * srcstride]);
    x3 = _mm_loadl_epi64((__m128i *) &src[x - 1 * srcstride]);
    x4 = _mm_loadl_epi64((__m128i *) &src[x                ]);
    x1 = _mm_unpacklo_epi8(x1, x2);
    x2 = _mm_unpacklo_epi8(x3, x4);
            x1   = _mm_maddubs_epi16(x1, c1);
    x2   = _mm_maddubs_epi16(x2, c2);
    r1 = _mm_add_epi16(x1, x2);
            x1 = _mm_loadl_epi64((__m128i *) &src[x +     srcstride]);
    x2 = _mm_loadl_epi64((__m128i *) &src[x + 2 * srcstride]);
    x3 = _mm_loadl_epi64((__m128i *) &src[x + 3 * srcstride]);
    x4 = _mm_loadl_epi64((__m128i *) &src[x + 4 * srcstride]);
    x1 = _mm_unpacklo_epi8(x1, x2);
    x2 = _mm_unpacklo_epi8(x3, x4);
            x1   = _mm_maddubs_epi16(x1, c3);
    x2   = _mm_maddubs_epi16(x2, c4);
    r3 = _mm_add_epi16(x1, x2);
            r1 = _mm_add_epi16(r1, r3);
            _mm_storel_epi64((__m128i *) &dst[x], r1);
        }
        src += srcstride;
        dst += dststride;
    }
}