annotate vp6.c @ 11034:fd5921186064 libavcodec

Make the fast loop filter path work with unavailable left MBs. This prevents the issue with having to switch between slow and fast code paths in each row. 0.5% faster loopfilter for cathedral

/**
 * @file libavcodec/vp6.c
 * VP6 compatible video decoder
 *
 * Copyright (C) 2006  Aurelien Jacobs <[e-mail]>
 *
 * The VP6F decoder accepts an optional 1 byte extradata. It is composed of:
 *  - upper 4bits: difference between encoded width and visible width
 *  - lower 4bits: difference between encoded height and visible height
 *
 * 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
 */

#include <stdlib.h>

#include "avcodec.h"
#include "dsputil.h"
#include "get_bits.h"
#include "huffman.h"

#include "vp56.h"
#include "vp56data.h"
#include "vp6data.h"


static void vp6_parse_coeff(VP56Context *s);
static void vp6_parse_coeff_huffman(VP56Context *s);

static int vp6_parse_header(VP56Context *s, const uint8_t *buf, int buf_size,
                            int *golden_frame)
{
    VP56RangeCoder *c = &s->c;
    int parse_filter_info = 0;
    int coeff_offset = 0;
    int vrt_shift = 0;
    int sub_version;
    int rows, cols;
    int res = 1;
    int separated_coeff = buf[0] & 1;

    s->framep[VP56_FRAME_CURRENT]->key_frame = !(buf[0] & 0x80);
    vp56_init_dequant(s, (buf[0] >> 1) & 0x3F);

    if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
        sub_version = buf[1] >> 3;
        if (sub_version > 8)
            return 0;
        s->filter_header = buf[1] & 0x06;
        if (buf[1] & 1) {
            av_log(s->avctx, AV_LOG_ERROR, "interlacing not supported\n");
            return 0;
        }
        if (separated_coeff || !s->filter_header) {
            coeff_offset = AV_RB16(buf+2) - 2;
            buf += 2;
            buf_size -= 2;
        }

        rows = buf[2];  /* number of stored macroblock rows */
        cols = buf[3];  /* number of stored macroblock cols */
        /* buf[4] is number of displayed macroblock rows */
        /* buf[5] is number of displayed macroblock cols */

        if (!s->macroblocks || /* first frame */
            16*cols != s->avctx->coded_width ||
            16*rows != s->avctx->coded_height) {
            avcodec_set_dimensions(s->avctx, 16*cols, 16*rows);
            if (s->avctx->extradata_size == 1) {
                s->avctx->width  -= s->avctx->extradata[0] >> 4;
                s->avctx->height -= s->avctx->extradata[0] & 0x0F;
            }
            res = 2;
        }

        vp56_init_range_decoder(c, buf+6, buf_size-6);
        vp56_rac_gets(c, 2);

        parse_filter_info = s->filter_header;
        if (sub_version < 8)
            vrt_shift = 5;
        s->sub_version = sub_version;
    } else {
        if (!s->sub_version)
            return 0;

        if (separated_coeff || !s->filter_header) {
            coeff_offset = AV_RB16(buf+1) - 2;
            buf += 2;
            buf_size -= 2;
        }
        vp56_init_range_decoder(c, buf+1, buf_size-1);

        *golden_frame = vp56_rac_get(c);
        if (s->filter_header) {
            s->deblock_filtering = vp56_rac_get(c);
            if (s->deblock_filtering)
                vp56_rac_get(c);
            if (s->sub_version > 7)
                parse_filter_info = vp56_rac_get(c);
        }
    }

    if (parse_filter_info) {
        if (vp56_rac_get(c)) {
            s->filter_mode = 2;
            s->sample_variance_threshold = vp56_rac_gets(c, 5) << vrt_shift;
            s->max_vector_length = 2 << vp56_rac_gets(c, 3);
        } else if (vp56_rac_get(c)) {
            s->filter_mode = 1;
        } else {
            s->filter_mode = 0;
        }
        if (s->sub_version > 7)
            s->filter_selection = vp56_rac_gets(c, 4);
        else
            s->filter_selection = 16;
    }

    s->use_huffman = vp56_rac_get(c);

    s->parse_coeff = vp6_parse_coeff;
    if (coeff_offset) {
        buf      += coeff_offset;
        buf_size -= coeff_offset;
        if (buf_size < 0)
            return 0;
        if (s->use_huffman) {
            s->parse_coeff = vp6_parse_coeff_huffman;
            init_get_bits(&s->gb, buf, buf_size<<3);
        } else {
            vp56_init_range_decoder(&s->cc, buf, buf_size);
            s->ccp = &s->cc;
        }
    } else {
        s->ccp = &s->c;
    }

    return res;
}

static void vp6_coeff_order_table_init(VP56Context *s)
{
    int i, pos, idx = 1;

    s->modelp->coeff_index_to_pos[0] = 0;
    for (i=0; i<16; i++)
        for (pos=1; pos<64; pos++)
            if (s->modelp->coeff_reorder[pos] == i)
                s->modelp->coeff_index_to_pos[idx++] = pos;
}

static void vp6_default_models_init(VP56Context *s)
{
    VP56Model *model = s->modelp;

    model->vector_dct[0] = 0xA2;
    model->vector_dct[1] = 0xA4;
    model->vector_sig[0] = 0x80;
    model->vector_sig[1] = 0x80;

    memcpy(model->mb_types_stats, vp56_def_mb_types_stats, sizeof(model->mb_types_stats));
    memcpy(model->vector_fdv, vp6_def_fdv_vector_model, sizeof(model->vector_fdv));
    memcpy(model->vector_pdv, vp6_def_pdv_vector_model, sizeof(model->vector_pdv));
    memcpy(model->coeff_runv, vp6_def_runv_coeff_model, sizeof(model->coeff_runv));
    memcpy(model->coeff_reorder, vp6_def_coeff_reorder, sizeof(model->coeff_reorder));

    vp6_coeff_order_table_init(s);
}

static void vp6_parse_vector_models(VP56Context *s)
{
    VP56RangeCoder *c = &s->c;
    VP56Model *model = s->modelp;
    int comp, node;

    for (comp=0; comp<2; comp++) {
        if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][0]))
            model->vector_dct[comp] = vp56_rac_gets_nn(c, 7);
        if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][1]))
            model->vector_sig[comp] = vp56_rac_gets_nn(c, 7);
    }

    for (comp=0; comp<2; comp++)
        for (node=0; node<7; node++)
            if (vp56_rac_get_prob(c, vp6_pdv_pct[comp][node]))
                model->vector_pdv[comp][node] = vp56_rac_gets_nn(c, 7);

    for (comp=0; comp<2; comp++)
        for (node=0; node<8; node++)
            if (vp56_rac_get_prob(c, vp6_fdv_pct[comp][node]))
                model->vector_fdv[comp][node] = vp56_rac_gets_nn(c, 7);
}

/* nodes must ascend by count, but with descending symbol order */
static int vp6_huff_cmp(const void *va, const void *vb)
{
    const Node *a = va, *b = vb;
    return (a->count - b->count)*16 + (b->sym - a->sym);
}

static void vp6_build_huff_tree(VP56Context *s, uint8_t coeff_model[],
                                const uint8_t *map, unsigned size, VLC *vlc)
{
    Node nodes[2*size], *tmp = &nodes[size];
    int a, b, i;

    /* first compute probabilities from model */
    tmp[0].count = 256;
    for (i=0; i<size-1; i++) {
        a = tmp[i].count *        coeff_model[i]  >> 8;
        b = tmp[i].count * (255 - coeff_model[i]) >> 8;
        nodes[map[2*i  ]].count = a + !a;
        nodes[map[2*i+1]].count = b + !b;
    }

    /* then build the huffman tree accodring to probabilities */
    ff_huff_build_tree(s->avctx, vlc, size, nodes, vp6_huff_cmp,
                       FF_HUFFMAN_FLAG_HNODE_FIRST);
}

static void vp6_parse_coeff_models(VP56Context *s)
{
    VP56RangeCoder *c = &s->c;
    VP56Model *model = s->modelp;
    int def_prob[11];
    int node, cg, ctx, pos;
    int ct;    /* code type */
    int pt;    /* plane type (0 for Y, 1 for U or V) */

    memset(def_prob, 0x80, sizeof(def_prob));

    for (pt=0; pt<2; pt++)
        for (node=0; node<11; node++)
            if (vp56_rac_get_prob(c, vp6_dccv_pct[pt][node])) {
                def_prob[node] = vp56_rac_gets_nn(c, 7);
                model->coeff_dccv[pt][node] = def_prob[node];
            } else if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
                model->coeff_dccv[pt][node] = def_prob[node];
            }

    if (vp56_rac_get(c)) {
        for (pos=1; pos<64; pos++)
            if (vp56_rac_get_prob(c, vp6_coeff_reorder_pct[pos]))
                model->coeff_reorder[pos] = vp56_rac_gets(c, 4);
        vp6_coeff_order_table_init(s);
    }

    for (cg=0; cg<2; cg++)
        for (node=0; node<14; node++)
            if (vp56_rac_get_prob(c, vp6_runv_pct[cg][node]))
                model->coeff_runv[cg][node] = vp56_rac_gets_nn(c, 7);

    for (ct=0; ct<3; ct++)
        for (pt=0; pt<2; pt++)
            for (cg=0; cg<6; cg++)
                for (node=0; node<11; node++)
                    if (vp56_rac_get_prob(c, vp6_ract_pct[ct][pt][cg][node])) {
                        def_prob[node] = vp56_rac_gets_nn(c, 7);
                        model->coeff_ract[pt][ct][cg][node] = def_prob[node];
                    } else if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
                        model->coeff_ract[pt][ct][cg][node] = def_prob[node];
                    }

    if (s->use_huffman) {
        for (pt=0; pt<2; pt++) {
            vp6_build_huff_tree(s, model->coeff_dccv[pt],
                                vp6_huff_coeff_map, 12, &s->dccv_vlc[pt]);
            vp6_build_huff_tree(s, model->coeff_runv[pt],
                                vp6_huff_run_map, 9, &s->runv_vlc[pt]);
            for (ct=0; ct<3; ct++)
                for (cg = 0; cg < 6; cg++)
                    vp6_build_huff_tree(s, model->coeff_ract[pt][ct][cg],
                                        vp6_huff_coeff_map, 12,
                                        &s->ract_vlc[pt][ct][cg]);
        }
        memset(s->nb_null, 0, sizeof(s->nb_null));
    } else {
    /* coeff_dcct is a linear combination of coeff_dccv */
    for (pt=0; pt<2; pt++)
        for (ctx=0; ctx<3; ctx++)
            for (node=0; node<5; node++)
                model->coeff_dcct[pt][ctx][node] = av_clip(((model->coeff_dccv[pt][node] * vp6_dccv_lc[ctx][node][0] + 128) >> 8) + vp6_dccv_lc[ctx][node][1], 1, 255);
    }
}

static void vp6_parse_vector_adjustment(VP56Context *s, VP56mv *vect)
{
    VP56RangeCoder *c = &s->c;
    VP56Model *model = s->modelp;
    int comp;

    *vect = (VP56mv) {0,0};
    if (s->vector_candidate_pos < 2)
        *vect = s->vector_candidate[0];

    for (comp=0; comp<2; comp++) {
        int i, delta = 0;

        if (vp56_rac_get_prob(c, model->vector_dct[comp])) {
            static const uint8_t prob_order[] = {0, 1, 2, 7, 6, 5, 4};
            for (i=0; i<sizeof(prob_order); i++) {
                int j = prob_order[i];
                delta |= vp56_rac_get_prob(c, model->vector_fdv[comp][j])<<j;
            }
            if (delta & 0xF0)
                delta |= vp56_rac_get_prob(c, model->vector_fdv[comp][3])<<3;
            else
                delta |= 8;
        } else {
            delta = vp56_rac_get_tree(c, vp56_pva_tree,
                                      model->vector_pdv[comp]);
        }

        if (delta && vp56_rac_get_prob(c, model->vector_sig[comp]))
            delta = -delta;

        if (!comp)
            vect->x += delta;
        else
            vect->y += delta;
    }
}

/**
 * Read number of consecutive blocks with null DC or AC.
 * This value is < 74.
 */
static unsigned vp6_get_nb_null(VP56Context *s)
{
    unsigned val = get_bits(&s->gb, 2);
    if (val == 2)
        val += get_bits(&s->gb, 2);
    else if (val == 3) {
        val = get_bits1(&s->gb) << 2;
        val = 6+val + get_bits(&s->gb, 2+val);
    }
    return val;
}

static void vp6_parse_coeff_huffman(VP56Context *s)
{
    VP56Model *model = s->modelp;
    uint8_t *permute = s->scantable.permutated;
    VLC *vlc_coeff;
    int coeff, sign, coeff_idx;
    int b, cg, idx;
    int pt = 0;    /* plane type (0 for Y, 1 for U or V) */

    for (b=0; b<6; b++) {
        int ct = 0;    /* code type */
        if (b > 3) pt = 1;
        vlc_coeff = &s->dccv_vlc[pt];

        for (coeff_idx=0; coeff_idx<64; ) {
            int run = 1;
            if (coeff_idx<2 && s->nb_null[coeff_idx][pt]) {
                s->nb_null[coeff_idx][pt]--;
                if (coeff_idx)
                    break;
            } else {
                if (get_bits_count(&s->gb) >= s->gb.size_in_bits)
                    return;
                coeff = get_vlc2(&s->gb, vlc_coeff->table, 9, 3);
                if (coeff == 0) {
                    if (coeff_idx) {
                        int pt = (coeff_idx >= 6);
                        run += get_vlc2(&s->gb, s->runv_vlc[pt].table, 9, 3);
                        if (run >= 9)
                            run += get_bits(&s->gb, 6);
                    } else
                        s->nb_null[0][pt] = vp6_get_nb_null(s);
                    ct = 0;
                } else if (coeff == 11) {  /* end of block */
                    if (coeff_idx == 1)    /* first AC coeff ? */
                        s->nb_null[1][pt] = vp6_get_nb_null(s);
                    break;
                } else {
                    int coeff2 = vp56_coeff_bias[coeff];
                    if (coeff > 4)
                        coeff2 += get_bits(&s->gb, coeff <= 9 ? coeff - 4 : 11);
                    ct = 1 + (coeff2 > 1);
                    sign = get_bits1(&s->gb);
                    coeff2 = (coeff2 ^ -sign) + sign;
                    if (coeff_idx)
                        coeff2 *= s->dequant_ac;
                    idx = model->coeff_index_to_pos[coeff_idx];
                    s->block_coeff[b][permute[idx]] = coeff2;
                }
            }
            coeff_idx+=run;
            cg = FFMIN(vp6_coeff_groups[coeff_idx], 3);
            vlc_coeff = &s->ract_vlc[pt][ct][cg];
        }
    }
}

static void vp6_parse_coeff(VP56Context *s)
{
    VP56RangeCoder *c = s->ccp;
    VP56Model *model = s->modelp;
    uint8_t *permute = s->scantable.permutated;
    uint8_t *model1, *model2, *model3;
    int coeff, sign, coeff_idx;
    int b, i, cg, idx, ctx;
    int pt = 0;    /* plane type (0 for Y, 1 for U or V) */

    for (b=0; b<6; b++) {
        int ct = 1;    /* code type */
        int run = 1;

        if (b > 3) pt = 1;

        ctx = s->left_block[vp56_b6to4[b]].not_null_dc
              + s->above_blocks[s->above_block_idx[b]].not_null_dc;
        model1 = model->coeff_dccv[pt];
        model2 = model->coeff_dcct[pt][ctx];

        for (coeff_idx=0; coeff_idx<64; ) {
            if ((coeff_idx>1 && ct==0) || vp56_rac_get_prob(c, model2[0])) {
                /* parse a coeff */
                if (vp56_rac_get_prob(c, model2[2])) {
                    if (vp56_rac_get_prob(c, model2[3])) {
                        idx = vp56_rac_get_tree(c, vp56_pc_tree, model1);
                        coeff = vp56_coeff_bias[idx+5];
                        for (i=vp56_coeff_bit_length[idx]; i>=0; i--)
                            coeff += vp56_rac_get_prob(c, vp56_coeff_parse_table[idx][i]) << i;
                    } else {
                        if (vp56_rac_get_prob(c, model2[4]))
                            coeff = 3 + vp56_rac_get_prob(c, model1[5]);
                        else
                            coeff = 2;
                    }
                    ct = 2;
                } else {
                    ct = 1;
                    coeff = 1;
                }
                sign = vp56_rac_get(c);
                coeff = (coeff ^ -sign) + sign;
                if (coeff_idx)
                    coeff *= s->dequant_ac;
                idx = model->coeff_index_to_pos[coeff_idx];
                s->block_coeff[b][permute[idx]] = coeff;
                run = 1;
            } else {
                /* parse a run */
                ct = 0;
                if (coeff_idx > 0) {
                    if (!vp56_rac_get_prob(c, model2[1]))
                        break;

                    model3 = model->coeff_runv[coeff_idx >= 6];
                    run = vp56_rac_get_tree(c, vp6_pcr_tree, model3);
                    if (!run)
                        for (run=9, i=0; i<6; i++)
                            run += vp56_rac_get_prob(c, model3[i+8]) << i;
                }
            }

            cg = vp6_coeff_groups[coeff_idx+=run];
            model1 = model2 = model->coeff_ract[pt][ct][cg];
        }

        s->left_block[vp56_b6to4[b]].not_null_dc =
        s->above_blocks[s->above_block_idx[b]].not_null_dc = !!s->block_coeff[b][0];
    }
}

static int vp6_adjust(int v, int t)
{
    int V = v, s = v >> 31;
    V ^= s;
    V -= s;
    if (V-t-1 >= (unsigned)(t-1))
        return v;
    V = 2*t - V;
    V += s;
    V ^= s;
    return V;
}

static int vp6_block_variance(uint8_t *src, int stride)
{
    int sum = 0, square_sum = 0;
    int y, x;

    for (y=0; y<8; y+=2) {
        for (x=0; x<8; x+=2) {
            sum += src[x];
            square_sum += src[x]*src[x];
        }
        src += 2*stride;
    }
    return (16*square_sum - sum*sum) >> 8;
}

static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
                           int delta, const int16_t *weights)
{
    int x, y;

    for (y=0; y<8; y++) {
        for (x=0; x<8; x++) {
            dst[x] = av_clip_uint8((  src[x-delta  ] * weights[0]
                                 + src[x        ] * weights[1]
                                 + src[x+delta  ] * weights[2]
                                 + src[x+2*delta] * weights[3] + 64) >> 7);
        }
        src += stride;
        dst += stride;
    }
}

static void vp6_filter_diag2(VP56Context *s, uint8_t *dst, uint8_t *src,
                             int stride, int h_weight, int v_weight)
{
    uint8_t *tmp = s->edge_emu_buffer+16;
    s->dsp.put_h264_chroma_pixels_tab[0](tmp, src, stride, 9, h_weight, 0);
    s->dsp.put_h264_chroma_pixels_tab[0](dst, tmp, stride, 8, 0, v_weight);
}

static void vp6_filter(VP56Context *s, uint8_t *dst, uint8_t *src,
                       int offset1, int offset2, int stride,
                       VP56mv mv, int mask, int select, int luma)
{
    int filter4 = 0;
    int x8 = mv.x & mask;
    int y8 = mv.y & mask;

    if (luma) {
        x8 *= 2;
        y8 *= 2;
        filter4 = s->filter_mode;
        if (filter4 == 2) {
            if (s->max_vector_length &&
                (FFABS(mv.x) > s->max_vector_length ||
                 FFABS(mv.y) > s->max_vector_length)) {
                filter4 = 0;
            } else if (s->sample_variance_threshold
                       && (vp6_block_variance(src+offset1, stride)
                           < s->sample_variance_threshold)) {
                filter4 = 0;
            }
        }
    }

    if ((y8 && (offset2-offset1)*s->flip<0) || (!y8 && offset1 > offset2)) {
        offset1 = offset2;
    }

    if (filter4) {
        if (!y8) {                      /* left or right combine */
            vp6_filter_hv4(dst, src+offset1, stride, 1,
                           vp6_block_copy_filter[select][x8]);
        } else if (!x8) {               /* above or below combine */
            vp6_filter_hv4(dst, src+offset1, stride, stride,
                           vp6_block_copy_filter[select][y8]);
        } else {
            s->dsp.vp6_filter_diag4(dst, src+offset1+((mv.x^mv.y)>>31), stride,
                             vp6_block_copy_filter[select][x8],
                             vp6_block_copy_filter[select][y8]);
        }
    } else {
        if (!x8 || !y8) {
            s->dsp.put_h264_chroma_pixels_tab[0](dst, src+offset1, stride, 8, x8, y8);
        } else {
            vp6_filter_diag2(s, dst, src+offset1 + ((mv.x^mv.y)>>31), stride, x8, y8);
        }
    }
}

static av_cold int vp6_decode_init(AVCodecContext *avctx)
{
    VP56Context *s = avctx->priv_data;

    vp56_init(avctx, avctx->codec->id == CODEC_ID_VP6,
                     avctx->codec->id == CODEC_ID_VP6A);
    s->vp56_coord_div = vp6_coord_div;
    s->parse_vector_adjustment = vp6_parse_vector_adjustment;
    s->adjust = vp6_adjust;
    s->filter = vp6_filter;
    s->default_models_init = vp6_default_models_init;
    s->parse_vector_models = vp6_parse_vector_models;
    s->parse_coeff_models = vp6_parse_coeff_models;
    s->parse_header = vp6_parse_header;

    return 0;
}

AVCodec vp6_decoder = {
    "vp6",
    CODEC_TYPE_VIDEO,
    CODEC_ID_VP6,
    sizeof(VP56Context),
    vp6_decode_init,
    NULL,
    vp56_free,
    vp56_decode_frame,
    CODEC_CAP_DR1,
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6"),
};

/* flash version, not flipped upside-down */
AVCodec vp6f_decoder = {
    "vp6f",
    CODEC_TYPE_VIDEO,
    CODEC_ID_VP6F,
    sizeof(VP56Context),
    vp6_decode_init,
    NULL,
    vp56_free,
    vp56_decode_frame,
    CODEC_CAP_DR1,
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6 (Flash version)"),
};

/* flash version, not flipped upside-down, with alpha channel */
AVCodec vp6a_decoder = {
    "vp6a",
    CODEC_TYPE_VIDEO,
    CODEC_ID_VP6A,
    sizeof(VP56Context),
    vp6_decode_init,
    NULL,
    vp56_free,
    vp56_decode_frame,
    CODEC_CAP_DR1,
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6 (Flash version, with alpha channel)"),
};