h264_loopfilter.c 32.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
/*
 * H.26L/H.264/AVC/JVT/14496-10/... loop filter
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file libavcodec/h264_loopfilter.c
 * H.264 / AVC / MPEG4 part10 loop filter.
 * @author Michael Niedermayer <michaelni@gmx.at>
 */

#include "internal.h"
#include "dsputil.h"
#include "avcodec.h"
#include "mpegvideo.h"
#include "h264.h"
#include "mathops.h"
#include "rectangle.h"

//#undef NDEBUG
#include <assert.h>

/* Deblocking filter (p153) */
static const uint8_t alpha_table[52*3] = {
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  4,  4,  5,  6,
     7,  8,  9, 10, 12, 13, 15, 17, 20, 22,
    25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
    80, 90,101,113,127,144,162,182,203,226,
   255,255,
   255,255,255,255,255,255,255,255,255,255,255,255,255,
   255,255,255,255,255,255,255,255,255,255,255,255,255,
   255,255,255,255,255,255,255,255,255,255,255,255,255,
   255,255,255,255,255,255,255,255,255,255,255,255,255,
};
static const uint8_t beta_table[52*3] = {
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
     0,  0,  0,  0,  0,  0,  2,  2,  2,  3,
     3,  3,  3,  4,  4,  4,  6,  6,  7,  7,
     8,  8,  9,  9, 10, 10, 11, 11, 12, 12,
    13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
    18, 18,
    18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
    18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
    18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
    18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
};
static const uint8_t tc0_table[52*3][4] = {
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
    {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 },
    {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
    {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
    {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
    {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
    {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
    {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
};

102
static void av_noinline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
103
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
104
    const int alpha = alpha_table[index_a];
105
    const int beta  = beta_table[qp + h->slice_beta_offset];
106 107 108 109
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
110 111 112 113
        tc[0] = tc0_table[index_a][bS[0]];
        tc[1] = tc0_table[index_a][bS[1]];
        tc[2] = tc0_table[index_a][bS[2]];
        tc[3] = tc0_table[index_a][bS[3]];
114 115 116 117 118
        h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}
119
static void av_noinline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
120
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
121
    const int alpha = alpha_table[index_a];
122
    const int beta  = beta_table[qp + h->slice_beta_offset];
123 124 125 126
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
127 128 129 130
        tc[0] = tc0_table[index_a][bS[0]]+1;
        tc[1] = tc0_table[index_a][bS[1]]+1;
        tc[2] = tc0_table[index_a][bS[2]]+1;
        tc[3] = tc0_table[index_a][bS[3]]+1;
131 132 133 134 135 136
        h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

137
static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
138
    int i;
139
    int index_a = qp + h->slice_alpha_c0_offset;
140 141
    int alpha = alpha_table[index_a];
    int beta  = beta_table[qp + h->slice_beta_offset];
142 143
    for( i = 0; i < 8; i++, pix += stride) {
        const int bS_index = (i >> 1) * bsi;
144 145 146 147 148 149

        if( bS[bS_index] == 0 ) {
            continue;
        }

        if( bS[bS_index] < 4 ) {
150
            const int tc0 = tc0_table[index_a][bS[bS_index]];
151 152 153 154 155 156 157 158 159 160 161 162 163 164
            const int p0 = pix[-1];
            const int p1 = pix[-2];
            const int p2 = pix[-3];
            const int q0 = pix[0];
            const int q1 = pix[1];
            const int q2 = pix[2];

            if( FFABS( p0 - q0 ) < alpha &&
                FFABS( p1 - p0 ) < beta &&
                FFABS( q1 - q0 ) < beta ) {
                int tc = tc0;
                int i_delta;

                if( FFABS( p2 - p0 ) < beta ) {
165
                    if(tc0)
166 167 168 169
                    pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
                    tc++;
                }
                if( FFABS( q2 - q0 ) < beta ) {
170
                    if(tc0)
171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
                    pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
                    tc++;
                }

                i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
                pix[-1] = av_clip_uint8( p0 + i_delta );    /* p0' */
                pix[0]  = av_clip_uint8( q0 - i_delta );    /* q0' */
                tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
            }
        }else{
            const int p0 = pix[-1];
            const int p1 = pix[-2];
            const int p2 = pix[-3];

            const int q0 = pix[0];
            const int q1 = pix[1];
            const int q2 = pix[2];

            if( FFABS( p0 - q0 ) < alpha &&
                FFABS( p1 - p0 ) < beta &&
                FFABS( q1 - q0 ) < beta ) {

                if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
                    if( FFABS( p2 - p0 ) < beta)
                    {
                        const int p3 = pix[-4];
                        /* p0', p1', p2' */
                        pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
                        pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
                        pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
                    } else {
                        /* p0' */
                        pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                    }
                    if( FFABS( q2 - q0 ) < beta)
                    {
                        const int q3 = pix[3];
                        /* q0', q1', q2' */
                        pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
                        pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
                        pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
                    } else {
                        /* q0' */
                        pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                    }
                }else{
                    /* p0', q0' */
                    pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                    pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                }
                tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
            }
        }
    }
}
226
static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
227
    int i;
228
    int index_a = qp + h->slice_alpha_c0_offset;
229 230
    int alpha = alpha_table[index_a];
    int beta  = beta_table[qp + h->slice_beta_offset];
231 232
    for( i = 0; i < 4; i++, pix += stride) {
        const int bS_index = i*bsi;
233 234 235 236 237 238

        if( bS[bS_index] == 0 ) {
            continue;
        }

        if( bS[bS_index] < 4 ) {
239
            const int tc = tc0_table[index_a][bS[bS_index]] + 1;
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
            const int p0 = pix[-1];
            const int p1 = pix[-2];
            const int q0 = pix[0];
            const int q1 = pix[1];

            if( FFABS( p0 - q0 ) < alpha &&
                FFABS( p1 - p0 ) < beta &&
                FFABS( q1 - q0 ) < beta ) {
                const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

                pix[-1] = av_clip_uint8( p0 + i_delta );    /* p0' */
                pix[0]  = av_clip_uint8( q0 - i_delta );    /* q0' */
                tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
            }
        }else{
            const int p0 = pix[-1];
            const int p1 = pix[-2];
            const int q0 = pix[0];
            const int q1 = pix[1];

            if( FFABS( p0 - q0 ) < alpha &&
                FFABS( p1 - p0 ) < beta &&
                FFABS( q1 - q0 ) < beta ) {

                pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
                pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
                tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
            }
        }
    }
}

272
static void av_noinline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
273
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
274
    const int alpha = alpha_table[index_a];
275
    const int beta  = beta_table[qp + h->slice_beta_offset];
276 277 278 279
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
280 281 282 283
        tc[0] = tc0_table[index_a][bS[0]];
        tc[1] = tc0_table[index_a][bS[1]];
        tc[2] = tc0_table[index_a][bS[2]];
        tc[3] = tc0_table[index_a][bS[3]];
284 285 286 287 288 289
        h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}

290
static void av_noinline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
291
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
292
    const int alpha = alpha_table[index_a];
293
    const int beta  = beta_table[qp + h->slice_beta_offset];
294 295 296 297
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
298 299 300 301
        tc[0] = tc0_table[index_a][bS[0]]+1;
        tc[1] = tc0_table[index_a][bS[1]]+1;
        tc[2] = tc0_table[index_a][bS[2]]+1;
        tc[3] = tc0_table[index_a][bS[3]]+1;
302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
        h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
    MpegEncContext * const s = &h->s;
    int mb_y_firstrow = s->picture_structure == PICT_BOTTOM_FIELD;
    int mb_xy, mb_type;
    int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;

    mb_xy = h->mb_xy;

    if(mb_x==0 || mb_y==mb_y_firstrow || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff ||
317 318
       (h->deblocking_filter == 2 && (h->slice_num != h->slice_table[h->top_mb_xy] ||
                                      h->slice_num != h->slice_table[mb_xy - 1]))) {
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
        ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
        return;
    }
    assert(!FRAME_MBAFF);

    mb_type = s->current_picture.mb_type[mb_xy];
    qp = s->current_picture.qscale_table[mb_xy];
    qp0 = s->current_picture.qscale_table[mb_xy-1];
    qp1 = s->current_picture.qscale_table[h->top_mb_xy];
    qpc = get_chroma_qp( h, 0, qp );
    qpc0 = get_chroma_qp( h, 0, qp0 );
    qpc1 = get_chroma_qp( h, 0, qp1 );
    qp0 = (qp + qp0 + 1) >> 1;
    qp1 = (qp + qp1 + 1) >> 1;
    qpc0 = (qpc + qpc0 + 1) >> 1;
    qpc1 = (qpc + qpc1 + 1) >> 1;
335
    qp_thresh = 15+52 - h->slice_alpha_c0_offset;
336 337 338 339 340 341 342 343 344
    if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
       qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
        return;

    if( IS_INTRA(mb_type) ) {
        int16_t bS4[4] = {4,4,4,4};
        int16_t bS3[4] = {3,3,3,3};
        int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
        if( IS_8x8DCT(mb_type) ) {
345 346 347 348
            filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
            filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
            filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
349
        } else {
350 351 352 353 354 355 356 357
            filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
            filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
            filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
            filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
            filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
358
        }
359 360 361 362 363 364 365 366
        filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
        filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
        filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
        filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
        filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
        filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
        filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
        filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
367 368
        return;
    } else {
369
        DECLARE_ALIGNED_8(int16_t, bS)[2][4][4];
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
        uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
        int edges;
        if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
            edges = 4;
            bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
        } else {
            int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :
                             (mb_type & MB_TYPE_16x8) ? 1 : 0;
            int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))
                             && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))
                             ? 3 : 0;
            int step = IS_8x8DCT(mb_type) ? 2 : 1;
            edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
            s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
                                              (h->slice_type_nos == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
        }
        if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )
            bSv[0][0] = 0x0004000400040004ULL;
        if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
            bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;

#define FILTER(hv,dir,edge)\
        if(bSv[dir][edge]) {\
393
            filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
394
            if(!(edge&1)) {\
395 396
                filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
                filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
            }\
        }
        if( edges == 1 ) {
            FILTER(v,0,0);
            FILTER(h,1,0);
        } else if( IS_8x8DCT(mb_type) ) {
            FILTER(v,0,0);
            FILTER(v,0,2);
            FILTER(h,1,0);
            FILTER(h,1,2);
        } else {
            FILTER(v,0,0);
            FILTER(v,0,1);
            FILTER(v,0,2);
            FILTER(v,0,3);
            FILTER(h,1,0);
            FILTER(h,1,1);
            FILTER(h,1,2);
            FILTER(h,1,3);
        }
#undef FILTER
    }
}


static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int dir) {
    MpegEncContext * const s = &h->s;
    int edge;
    const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
    const int mbm_type = s->current_picture.mb_type[mbm_xy];

    // how often to recheck mv-based bS when iterating between edges
429 430 431
    static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
                                              {0,3,1,1,3,3,3,3}};
    const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
432 433
    const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;

434 435
    // how often to recheck mv-based bS when iterating along each edge
    const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
436 437 438
    int start =   h->slice_table[mbm_xy] == 0xFFFF
               || first_vertical_edge_done
               || (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_num);
439 440 441


    if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0
442
        && IS_INTERLACED(mbm_type&~mb_type)
443 444 445 446 447 448 449 450
        ) {
        // This is a special case in the norm where the filtering must
        // be done twice (one each of the field) even if we are in a
        // frame macroblock.
        //
        unsigned int tmp_linesize   = 2 *   linesize;
        unsigned int tmp_uvlinesize = 2 * uvlinesize;
        int mbn_xy = mb_xy - 2 * s->mb_stride;
451
        int j;
452 453

        for(j=0; j<2; j++, mbn_xy += s->mb_stride){
454
            DECLARE_ALIGNED_8(int16_t, bS)[4];
455
            int qp;
456
            if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
457
                *(uint64_t*)bS= 0x0003000300030003ULL;
458
            } else {
459 460
                const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8;
                int i;
461
                for( i = 0; i < 4; i++ ) {
462
                    bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
463 464 465 466 467 468 469
                }
            }
            // Do not use s->qscale as luma quantizer because it has not the same
            // value in IPCM macroblocks.
            qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
            tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
            { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
470 471 472 473 474
            filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
            filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
            filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
475 476 477 478 479 480 481 482 483 484
        }

        start = 1;
    }

    /* Calculate bS */
    for( edge = start; edge < edges; edge++ ) {
        /* mbn_xy: neighbor macroblock */
        const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
        const int mbn_type = s->current_picture.mb_type[mbn_xy];
485
        DECLARE_ALIGNED_8(int16_t, bS)[4];
486 487
        int qp;

488
        if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
489 490
            continue;

491 492
        if( IS_INTRA(mb_type|mbn_type)) {
            *(uint64_t*)bS= 0x0003000300030003ULL;
493
            if (edge == 0) {
494
                if (   (!IS_INTERLACED(mb_type|mbm_type))
495
                    || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
496 497
                )
                    *(uint64_t*)bS= 0x0004000400040004ULL;
498 499 500 501 502 503
            }
        } else {
            int i, l;
            int mv_done;

            if( edge & mask_edge ) {
504
                *(uint64_t*)bS= 0;
505 506 507
                mv_done = 1;
            }
            else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
508
                *(uint64_t*)bS= 0x0001000100010001ULL;
509 510 511 512 513 514 515 516
                mv_done = 1;
            }
            else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
                int b_idx= 8 + 4 + edge * (dir ? 8:1);
                int bn_idx= b_idx - (dir ? 8:1);
                int v = 0;

                for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
517
                    v |= h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
518
                         h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
519 520 521 522 523 524 525
                         FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
                }

                if(h->slice_type_nos == FF_B_TYPE && v){
                    v=0;
                    for( l = 0; !v && l < 2; l++ ) {
                        int ln= 1-l;
526
                        v |= h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
527
                            h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
                            FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
                    }
                }

                bS[0] = bS[1] = bS[2] = bS[3] = v;
                mv_done = 1;
            }
            else
                mv_done = 0;

            for( i = 0; i < 4; i++ ) {
                int x = dir == 0 ? edge : i;
                int y = dir == 0 ? i    : edge;
                int b_idx= 8 + 4 + x + 8*y;
                int bn_idx= b_idx - (dir ? 8:1);

                if( h->non_zero_count_cache[b_idx] |
                    h->non_zero_count_cache[bn_idx] ) {
                    bS[i] = 2;
                }
                else if(!mv_done)
                {
                    bS[i] = 0;
                    for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
552
                        if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
553
                            h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
554 555 556 557 558 559 560 561 562 563
                            FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
                            bS[i] = 1;
                            break;
                        }
                    }

                    if(h->slice_type_nos == FF_B_TYPE && bS[i]){
                        bS[i] = 0;
                        for( l = 0; l < 2; l++ ) {
                            int ln= 1-l;
564
                            if( h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
565
                                h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582
                                FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
                                bS[i] = 1;
                                break;
                            }
                        }
                    }
                }
            }

            if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
                continue;
        }

        /* Filter edge */
        // Do not use s->qscale as luma quantizer because it has not the same
        // value in IPCM macroblocks.
        qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
583
        //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], s->current_picture.qscale_table[mbn_xy]);
584
        tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
Michael Niedermayer's avatar
Michael Niedermayer committed
585
        //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
586
        if( dir == 0 ) {
587
            filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
588
            if( (edge&1) == 0 ) {
589
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
590
                filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, qp, h);
591 592
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
593
                filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, qp, h);
594 595
            }
        } else {
596
            filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
597
            if( (edge&1) == 0 ) {
598
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
599
                filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, qp, h);
600 601
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
602
                filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, qp, h);
603 604 605 606 607 608 609 610 611 612 613 614
            }
        }
    }
}

void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
    MpegEncContext * const s = &h->s;
    const int mb_xy= mb_x + mb_y*s->mb_stride;
    const int mb_type = s->current_picture.mb_type[mb_xy];
    const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
    int first_vertical_edge_done = 0;
    av_unused int dir;
615
    int list;
616 617 618 619 620

    if (FRAME_MBAFF
            // left mb is in picture
            && h->slice_table[mb_xy-1] != 0xFFFF
            // and current and left pair do not have the same interlaced type
621
            && IS_INTERLACED(mb_type^s->current_picture.mb_type[mb_xy-1])
622
            // and left mb is in the same slice if deblocking_filter == 2
623
            && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_num)) {
624 625 626 627 628
        /* First vertical edge is different in MBAFF frames
         * There are 8 different bS to compute and 2 different Qp
         */
        const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
        const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
629
        DECLARE_ALIGNED_8(int16_t, bS)[8];
630 631 632 633 634 635 636 637
        int qp[2];
        int bqp[2];
        int rqp[2];
        int mb_qp, mbn0_qp, mbn1_qp;
        int i;
        first_vertical_edge_done = 1;

        if( IS_INTRA(mb_type) )
638 639
            *(uint64_t*)&bS[0]=
            *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
640 641 642 643 644 645 646 647 648 649
        else {
            for( i = 0; i < 8; i++ ) {
                int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];

                if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
                    bS[i] = 4;
                else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
                         ((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
                            (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
                                                                       :
650
                            h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8]))
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673
                    bS[i] = 2;
                else
                    bS[i] = 1;
            }
        }

        mb_qp = s->current_picture.qscale_table[mb_xy];
        mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
        mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
        qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
        bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
                   get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
        rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
                   get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
        qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
        bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
                   get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
        rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
                   get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;

        /* Filter edge */
        tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);
        { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
674 675 676 677 678 679 680 681 682 683 684 685 686 687 688
        if(MB_FIELD){
            filter_mb_mbaff_edgev ( h, img_y                ,   linesize, bS  , 1, qp [0] );
            filter_mb_mbaff_edgev ( h, img_y  + 8*  linesize,   linesize, bS+4, 1, qp [1] );
            filter_mb_mbaff_edgecv( h, img_cb,                uvlinesize, bS  , 1, bqp[0] );
            filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
            filter_mb_mbaff_edgecv( h, img_cr,                uvlinesize, bS  , 1, rqp[0] );
            filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
        }else{
            filter_mb_mbaff_edgev ( h, img_y              , 2*  linesize, bS  , 2, qp [0] );
            filter_mb_mbaff_edgev ( h, img_y  +   linesize, 2*  linesize, bS+1, 2, qp [1] );
            filter_mb_mbaff_edgecv( h, img_cb,              2*uvlinesize, bS  , 2, bqp[0] );
            filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
            filter_mb_mbaff_edgecv( h, img_cr,              2*uvlinesize, bS  , 2, rqp[0] );
            filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
        }
689 690 691 692 693 694 695 696 697 698
    }

#if CONFIG_SMALL
    for( dir = 0; dir < 2; dir++ )
        filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, dir);
#else
    filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, 0);
    filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
#endif
}