umc_h264_sad.cpp

audio-video-codecs.rar语音编解码器

    _p_3 = _mm_loadl_epi64((__m128i*)(&diff[7][4]));
    _p_0 = _mm_unpacklo_epi16(_p_0, _p_1);
    _p_2 = _mm_unpacklo_epi16(_p_2, _p_3);
    _p_1 = _mm_unpackhi_epi32(_p_0, _p_2);
    _p_0 = _mm_unpacklo_epi32(_p_0, _p_2);
    _mm_storel_epi64((__m128i*)&tmp[4][4], _p_0);
    _mm_storel_epi64((__m128i*)&tmp[6][4], _p_1);
    _p_0 = _mm_srli_si128(_p_0, 8);
    _p_1 = _mm_srli_si128(_p_1, 8);
    _mm_storel_epi64((__m128i*)&tmp[5][4], _p_0);
    _mm_storel_epi64((__m128i*)&tmp[7][4], _p_1);

    _p_0 = _mm_load_si128((__m128i*)(tmp[0]));
    _p_4 = _mm_sub_epi16(_p_0, *(__m128i*)(tmp[4]));
    _p_0 = _mm_add_epi16(_p_0, *(__m128i*)(tmp[4]));
    _p_1 = _mm_load_si128((__m128i*)(tmp[1]));
    _p_5 = _mm_sub_epi16(_p_1, *(__m128i*)(tmp[5]));
    _p_1 = _mm_add_epi16(_p_1, *(__m128i*)(tmp[5]));
    _p_2 = _mm_load_si128((__m128i*)(tmp[2]));
    _p_6 = _mm_sub_epi16(_p_2, *(__m128i*)(tmp[6]));
    _p_2 = _mm_add_epi16(_p_2, *(__m128i*)(tmp[6]));
    _p_3 = _mm_load_si128((__m128i*)(tmp[3]));
    _p_7 = _mm_sub_epi16(_p_3, *(__m128i*)(tmp[7]));
    _p_3 = _mm_add_epi16(_p_3, *(__m128i*)(tmp[7]));

    _b_2 = _mm_sub_epi16(_p_0, _p_2);
    _p_0 = _mm_add_epi16(_p_0, _p_2);
    _b_3 = _mm_sub_epi16(_p_1, _p_3);
    _p_1 = _mm_add_epi16(_p_1, _p_3);
    _b_6 = _mm_sub_epi16(_p_4, _p_6);
    _p_4 = _mm_add_epi16(_p_4, _p_6);
    _b_7 = _mm_sub_epi16(_p_5, _p_7);
    _p_5 = _mm_add_epi16(_p_5, _p_7);

    _p_s = _mm_sub_epi16(_p_0, _p_1);
    _p_t = _mm_srai_epi16(_p_s, 15);
    _p_s = _mm_xor_si128(_p_s, _p_t);
    _p_s = _mm_sub_epi16(_p_s, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_1);
    _p_t = _mm_srai_epi16(_p_0, 15);
    _p_0 = _mm_xor_si128(_p_0, _p_t);
    _p_0 = _mm_sub_epi16(_p_0, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_s);

    _p_s = _mm_sub_epi16(_b_2, _b_3);
    _p_t = _mm_srai_epi16(_p_s, 15);
    _p_s = _mm_xor_si128(_p_s, _p_t);
    _p_s = _mm_sub_epi16(_p_s, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_s);
    _b_2 = _mm_add_epi16(_b_2, _b_3);
    _p_t = _mm_srai_epi16(_b_2, 15);
    _b_2 = _mm_xor_si128(_b_2, _p_t);
    _b_2 = _mm_sub_epi16(_b_2, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _b_2);

    _p_s = _mm_sub_epi16(_p_4, _p_5);
    _p_t = _mm_srai_epi16(_p_s, 15);
    _p_s = _mm_xor_si128(_p_s, _p_t);
    _p_s = _mm_sub_epi16(_p_s, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_s);
    _p_4 = _mm_add_epi16(_p_4, _p_5);
    _p_t = _mm_srai_epi16(_p_4, 15);
    _p_4 = _mm_xor_si128(_p_4, _p_t);
    _p_4 = _mm_sub_epi16(_p_4, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_4);

    _p_s = _mm_sub_epi16(_b_6, _b_7);
    _p_t = _mm_srai_epi16(_p_s, 15);
    _p_s = _mm_xor_si128(_p_s, _p_t);
    _p_s = _mm_sub_epi16(_p_s, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _p_s);
    _b_6 = _mm_add_epi16(_b_6, _b_7);
    _p_t = _mm_srai_epi16(_b_6, 15);
    _b_6 = _mm_xor_si128(_b_6, _p_t);
    _b_6 = _mm_sub_epi16(_b_6, _p_t);
    _p_0 = _mm_add_epi16(_p_0, _b_6);

    _p_t = _mm_srli_si128(_p_0, 8);
    _p_0 = _mm_add_epi16(_p_0, _p_t);
    _p_t = _mm_srli_si128(_p_0, 4);
    _p_0 = _mm_add_epi16(_p_0, _p_t);
    s = _mm_cvtsi128_si32(_p_0);
    satd += (s >> 16) + (Ipp16s)s;
#endif
    return satd>> 2;
}


Ipp32u SAT8x8D(const Ipp16u *pSrc1, Ipp32s src1Step, const Ipp16u *pSrc2, Ipp32s src2Step)
{
    __ALIGN16 Ipp32s tmp[8][8];
    __ALIGN16 Ipp16s diff[8][8];
    Ipp32s i;
    Ipp32u satd = 0;

    ippiSub8x8_16u16s_C1R(pSrc1, src1Step, pSrc2, src2Step, &diff[0][0], 32);
    for (i = 0; i < 8; i++) {
        Ipp32s t0 = diff[i][0] + diff[i][4];
        Ipp32s t4 = diff[i][0] - diff[i][4];
        Ipp32s t1 = diff[i][1] + diff[i][5];
        Ipp32s t5 = diff[i][1] - diff[i][5];
        Ipp32s t2 = diff[i][2] + diff[i][6];
        Ipp32s t6 = diff[i][2] - diff[i][6];
        Ipp32s t3 = diff[i][3] + diff[i][7];
        Ipp32s t7 = diff[i][3] - diff[i][7];
        Ipp32s s0 = t0 + t2;
        Ipp32s s2 = t0 - t2;
        Ipp32s s1 = t1 + t3;
        Ipp32s s3 = t1 - t3;
        Ipp32s s4 = t4 + t6;
        Ipp32s s6 = t4 - t6;
        Ipp32s s5 = t5 + t7;
        Ipp32s s7 = t5 - t7;
        tmp[i][0] = s0 + s1;
        tmp[i][1] = s0 - s1;
        tmp[i][2] = s2 + s3;
        tmp[i][3] = s2 - s3;
        tmp[i][4] = s4 + s5;
        tmp[i][5] = s4 - s5;
        tmp[i][6] = s6 + s7;
        tmp[i][7] = s6 - s7;
    }
    for (i = 0; i < 8; i++) {
        Ipp32s t0 = tmp[0][i] + tmp[4][i];
        Ipp32s t4 = tmp[0][i] - tmp[4][i];
        Ipp32s t1 = tmp[1][i] + tmp[5][i];
        Ipp32s t5 = tmp[1][i] - tmp[5][i];
        Ipp32s t2 = tmp[2][i] + tmp[6][i];
        Ipp32s t6 = tmp[2][i] - tmp[6][i];
        Ipp32s t3 = tmp[3][i] + tmp[7][i];
        Ipp32s t7 = tmp[3][i] - tmp[7][i];
        Ipp32s s0 = t0 + t2;
        Ipp32s s2 = t0 - t2;
        Ipp32s s1 = t1 + t3;
        Ipp32s s3 = t1 - t3;
        Ipp32s s4 = t4 + t6;
        Ipp32s s6 = t4 - t6;
        Ipp32s s5 = t5 + t7;
        Ipp32s s7 = t5 - t7;
        satd += ABS(s0 + s1);
        satd += ABS(s0 - s1);
        satd += ABS(s2 + s3);
        satd += ABS(s2 - s3);
        satd += ABS(s4 + s5);
        satd += ABS(s4 - s5);
        satd += ABS(s6 + s7);
        satd += ABS(s6 - s7);
    }
    return satd >> 2;
}


Ipp32u SATD_8u_C1R(const Ipp8u *pSrc1, Ipp32s src1Step, const Ipp8u *pSrc2, Ipp32s src2Step, Ipp32s width, Ipp32s height)
{
#ifndef H264_SATD_OPT
    __ALIGN16 Ipp16s tmpBuff[4][4];
    __ALIGN16 Ipp16s diffBuff[4][4];
#endif
    Ipp32s x, y;
    Ipp32u satd = 0;

    for( y = 0; y < height; y += 4 ) {
        for( x = 0; x < width; x += 4 )  {
#ifndef H264_SATD_OPT
            Ipp32s b;
            ippiSub4x4_8u16s_C1R(pSrc1 + x, src1Step, pSrc2 + x, src2Step, &diffBuff[0][0], 8);
            for (b = 0; b < 4; b ++) {
                Ipp32s a01, a23, b01, b23;

                a01 = diffBuff[b][0] + diffBuff[b][1];
                a23 = diffBuff[b][2] + diffBuff[b][3];
                b01 = diffBuff[b][0] - diffBuff[b][1];
                b23 = diffBuff[b][2] - diffBuff[b][3];
                tmpBuff[b][0] = a01 + a23;
                tmpBuff[b][1] = a01 - a23;
                tmpBuff[b][2] = b01 - b23;
                tmpBuff[b][3] = b01 + b23;
            }
            for (b = 0; b < 4; b ++) {
                Ipp32s a01, a23, b01, b23;

                a01 = tmpBuff[0][b] + tmpBuff[1][b];
                a23 = tmpBuff[2][b] + tmpBuff[3][b];
                b01 = tmpBuff[0][b] - tmpBuff[1][b];
                b23 = tmpBuff[2][b] - tmpBuff[3][b];
                satd += ABS(a01 + a23) + ABS(a01 - a23) + ABS(b01 - b23) + ABS(b01 + b23);
            }
#else
            __ALIGN16 __m128i  _p_0, _p_1, _p_2, _p_3, _p_4, _p_5, _p_7, _p_zero;
            const Ipp8u *pS1, *pS2;
            Ipp32s  s;

            pS1 = pSrc1 + x;
            pS2 = pSrc2 + x;
            _p_zero = _mm_setzero_si128();
            _p_0 = _mm_cvtsi32_si128(*(int*)(pS1));
            _p_4 = _mm_cvtsi32_si128(*(int*)(pS2));
            _p_1 = _mm_cvtsi32_si128(*(int*)(pS1+src1Step));
            _p_5 = _mm_cvtsi32_si128(*(int*)(pS2+src2Step));
            _p_0 = _mm_unpacklo_epi8(_p_0, _p_zero);
            _p_4 = _mm_unpacklo_epi8(_p_4, _p_zero);
            _p_1 = _mm_unpacklo_epi8(_p_1, _p_zero);
            _p_5 = _mm_unpacklo_epi8(_p_5, _p_zero);
            _p_0 = _mm_sub_epi16(_p_0, _p_4);
            _p_1 = _mm_sub_epi16(_p_1, _p_5);
            pS1 += 2 * src1Step;
            pS2 += 2 * src2Step;
            _p_2 = _mm_cvtsi32_si128(*(int*)(pS1));
            _p_4 = _mm_cvtsi32_si128(*(int*)(pS2));
            _p_3 = _mm_cvtsi32_si128(*(int*)(pS1+src1Step));
            _p_5 = _mm_cvtsi32_si128(*(int*)(pS2+src2Step));
            _p_2 = _mm_unpacklo_epi8(_p_2, _p_zero);
            _p_4 = _mm_unpacklo_epi8(_p_4, _p_zero);
            _p_3 = _mm_unpacklo_epi8(_p_3, _p_zero);
            _p_5 = _mm_unpacklo_epi8(_p_5, _p_zero);
            _p_2 = _mm_sub_epi16(_p_2, _p_4);
            _p_3 = _mm_sub_epi16(_p_3, _p_5);
            _p_5 = _mm_subs_epi16(_p_0, _p_1);
            _p_0 = _mm_adds_epi16(_p_0, _p_1);
            _p_7 = _mm_subs_epi16(_p_2, _p_3);
            _p_2 = _mm_adds_epi16(_p_2, _p_3);
            _p_1 = _mm_subs_epi16(_p_0, _p_2);
            _p_0 = _mm_adds_epi16(_p_0, _p_2);
            _p_3 = _mm_adds_epi16(_p_5, _p_7);
            _p_5 = _mm_subs_epi16(_p_5, _p_7);
            _p_0 = _mm_unpacklo_epi16(_p_0, _p_1);
            _p_5 = _mm_unpacklo_epi16(_p_5, _p_3);
            _p_7 = _mm_unpackhi_epi32(_p_0, _p_5);
            _p_0 = _mm_unpacklo_epi32(_p_0, _p_5);
            _p_1 = _mm_srli_si128(_p_0, 8);
            _p_3 = _mm_srli_si128(_p_7, 8);
            _p_5 = _mm_subs_epi16(_p_0, _p_1);
            _p_0 = _mm_adds_epi16(_p_0, _p_1);
            _p_2 = _mm_subs_epi16(_p_7, _p_3);
            _p_7 = _mm_adds_epi16(_p_7, _p_3);
            _p_1 = _mm_subs_epi16(_p_0, _p_7);
            _p_0 = _mm_adds_epi16(_p_0, _p_7);
            _p_3 = _mm_adds_epi16(_p_5, _p_2);
            _p_5 = _mm_subs_epi16(_p_5, _p_2);
            _p_0 = _mm_unpacklo_epi16(_p_0, _p_1);
            _p_5 = _mm_unpacklo_epi16(_p_5, _p_3);
            _p_2 = _mm_unpackhi_epi32(_p_0, _p_5);
            _p_0 = _mm_unpacklo_epi32(_p_0, _p_5);
            _p_3 = _mm_srai_epi16(_p_2, 15);
            _p_1 = _mm_srai_epi16(_p_0, 15);
            _p_2 = _mm_xor_si128(_p_2, _p_3);
            _p_0 = _mm_xor_si128(_p_0, _p_1);
            _p_2 = _mm_sub_epi16(_p_2, _p_3);
            _p_0 = _mm_sub_epi16(_p_0, _p_1);
            _p_0 = _mm_add_epi16(_p_0, _p_2);
            _p_2 = _mm_srli_si128(_p_0, 8);
            _p_0 = _mm_add_epi16(_p_0, _p_2);
            _p_2 = _mm_srli_si128(_p_0, 4);
            _p_0 = _mm_add_epi16(_p_0, _p_2);
            s = _mm_cvtsi128_si32(_p_0);
            satd += (s >> 16) + (Ipp16s)s;
#endif
        }
        pSrc1 += 4 * src1Step;
        pSrc2 += 4 * src2Step;
    }
    return satd >> 1;
}

Ipp32u SATD_16u_C1R(const Ipp16u *pSrc1, Ipp32s src1Step, const Ipp16u *pSrc2, Ipp32s src2Step, Ipp32s width, Ipp32s height)
{
    __ALIGN16 Ipp32s tmpBuff[4][4];
    __ALIGN16 Ipp16s diffBuff[4][4];
    Ipp32s x, y;
    Ipp32u satd = 0;

    src1Step >>= 1;
    src2Step >>= 1;
    for( y = 0; y < height; y += 4 ) {
        for( x = 0; x < width; x += 4 )  {
            Ipp32s b;
            ippiSub4x4_16u16s_C1R(pSrc1 + x, src1Step, pSrc2 + x, src2Step, &diffBuff[0][0], 16);
            for (b = 0; b < 4; b ++) {
                Ipp32s a01, a23, b01, b23;

                a01 = diffBuff[b][0] + diffBuff[b][1];
                a23 = diffBuff[b][2] + diffBuff[b][3];
                b01 = diffBuff[b][0] - diffBuff[b][1];
                b23 = diffBuff[b][2] - diffBuff[b][3];
                tmpBuff[b][0] = a01 + a23;
                tmpBuff[b][1] = a01 - a23;
                tmpBuff[b][2] = b01 - b23;
                tmpBuff[b][3] = b01 + b23;
            }
            for (b = 0; b < 4; b ++) {
                Ipp32s a01, a23, b01, b23;

                a01 = tmpBuff[0][b] + tmpBuff[1][b];
                a23 = tmpBuff[2][b] + tmpBuff[3][b];
                b01 = tmpBuff[0][b] - tmpBuff[1][b];
                b23 = tmpBuff[2][b] - tmpBuff[3][b];
                satd += ABS(a01 + a23) + ABS(a01 - a23) + ABS(b01 - b23) + ABS(b01 + b23);
            }
        }
        pSrc1 += 4 * src1Step;
        pSrc2 += 4 * src2Step;
    }
    return satd >> 1;
}


} //namespace UMC_H264_ENCODER