dnxhdenc.c

ffmpeg移植到symbian的全部源代码

/*
 * VC3/DNxHD encoder
 * Copyright (c) 2007 Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
 *
 * VC-3 encoder funded by the British Broadcasting Corporation
 *
 * 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
 */

//#define DEBUG
#define RC_VARIANCE 1 // use variance or ssd for fast rc

#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
#include "dnxhddata.h"

typedef struct {
    uint16_t mb;
    int value;
} RCCMPEntry;

typedef struct {
    int ssd;
    int bits;
} RCEntry;

int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);

typedef struct DNXHDEncContext {
    MpegEncContext m; ///< Used for quantization dsp functions

    AVFrame frame;
    int cid;
    const CIDEntry *cid_table;
    uint8_t *msip; ///< Macroblock Scan Indexes Payload
    uint32_t *slice_size;

    struct DNXHDEncContext *thread[MAX_THREADS];

    unsigned dct_y_offset;
    unsigned dct_uv_offset;
    int interlaced;
    int cur_field;

    DECLARE_ALIGNED_16(DCTELEM, blocks[8][64]);

    int      (*qmatrix_c)     [64];
    int      (*qmatrix_l)     [64];
    uint16_t (*qmatrix_l16)[2][64];
    uint16_t (*qmatrix_c16)[2][64];

    unsigned frame_bits;
    uint8_t *src[3];

    uint32_t *vlc_codes;
    uint8_t  *vlc_bits;
    uint16_t *run_codes;
    uint8_t  *run_bits;

    /** Rate control */
    unsigned slice_bits;
    unsigned qscale;
    unsigned lambda;

    unsigned thread_size;

    uint16_t *mb_bits;
    uint8_t  *mb_qscale;

    RCCMPEntry *mb_cmp;
    RCEntry   (*mb_rc)[8160];
} DNXHDEncContext;

#define LAMBDA_FRAC_BITS 10

static int dnxhd_init_vlc(DNXHDEncContext *ctx)
{
    int i, j, level, run;
    int max_level = 1<<(ctx->cid_table->bit_depth+2);

    CHECKED_ALLOCZ(ctx->vlc_codes, max_level*4*sizeof(*ctx->vlc_codes));
    CHECKED_ALLOCZ(ctx->vlc_bits,  max_level*4*sizeof(*ctx->vlc_bits));
    CHECKED_ALLOCZ(ctx->run_codes, 63*2);
    CHECKED_ALLOCZ(ctx->run_bits,    63);

    ctx->vlc_codes += max_level*2;
    ctx->vlc_bits  += max_level*2;
    for (level = -max_level; level < max_level; level++) {
        for (run = 0; run < 2; run++) {
            int index = (level<<1)|run;
            int sign, offset = 0, alevel = level;

            MASK_ABS(sign, alevel);
            if (alevel > 64) {
                offset = (alevel-1)>>6;
                alevel -= offset<<6;
            }
            for (j = 0; j < 257; j++) {
                if (ctx->cid_table->ac_level[j] == alevel &&
                    (!offset || (ctx->cid_table->ac_index_flag[j] && offset)) &&
                    (!run    || (ctx->cid_table->ac_run_flag  [j] && run))) {
                    assert(!ctx->vlc_codes[index]);
                    if (alevel) {
                        ctx->vlc_codes[index] = (ctx->cid_table->ac_codes[j]<<1)|(sign&1);
                        ctx->vlc_bits [index] = ctx->cid_table->ac_bits[j]+1;
                    } else {
                        ctx->vlc_codes[index] = ctx->cid_table->ac_codes[j];
                        ctx->vlc_bits [index] = ctx->cid_table->ac_bits [j];
                    }
                    break;
                }
            }
            assert(!alevel || j < 257);
            if (offset) {
                ctx->vlc_codes[index] = (ctx->vlc_codes[index]<<ctx->cid_table->index_bits)|offset;
                ctx->vlc_bits [index]+= ctx->cid_table->index_bits;
            }
        }
    }
    for (i = 0; i < 62; i++) {
        int run = ctx->cid_table->run[i];
        assert(run < 63);
        ctx->run_codes[run] = ctx->cid_table->run_codes[i];
        ctx->run_bits [run] = ctx->cid_table->run_bits[i];
    }
    return 0;
 fail:
    return -1;
}

static int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
{
    // init first elem to 1 to avoid div by 0 in convert_matrix
    uint16_t weight_matrix[64] = {1,}; // convert_matrix needs uint16_t*
    int qscale, i;

    CHECKED_ALLOCZ(ctx->qmatrix_l,   (ctx->m.avctx->qmax+1) * 64 * sizeof(int));
    CHECKED_ALLOCZ(ctx->qmatrix_c,   (ctx->m.avctx->qmax+1) * 64 * sizeof(int));
    CHECKED_ALLOCZ(ctx->qmatrix_l16, (ctx->m.avctx->qmax+1) * 64 * 2 * sizeof(uint16_t));
    CHECKED_ALLOCZ(ctx->qmatrix_c16, (ctx->m.avctx->qmax+1) * 64 * 2 * sizeof(uint16_t));

    for (i = 1; i < 64; i++) {
        int j = ctx->m.dsp.idct_permutation[ff_zigzag_direct[i]];
        weight_matrix[j] = ctx->cid_table->luma_weight[i];
    }
    ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_l, ctx->qmatrix_l16, weight_matrix,
                      ctx->m.intra_quant_bias, 1, ctx->m.avctx->qmax, 1);
    for (i = 1; i < 64; i++) {
        int j = ctx->m.dsp.idct_permutation[ff_zigzag_direct[i]];
        weight_matrix[j] = ctx->cid_table->chroma_weight[i];
    }
    ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_c, ctx->qmatrix_c16, weight_matrix,
                      ctx->m.intra_quant_bias, 1, ctx->m.avctx->qmax, 1);
    for (qscale = 1; qscale <= ctx->m.avctx->qmax; qscale++) {
        for (i = 0; i < 64; i++) {
            ctx->qmatrix_l  [qscale]   [i] <<= 2; ctx->qmatrix_c  [qscale]   [i] <<= 2;
            ctx->qmatrix_l16[qscale][0][i] <<= 2; ctx->qmatrix_l16[qscale][1][i] <<= 2;
            ctx->qmatrix_c16[qscale][0][i] <<= 2; ctx->qmatrix_c16[qscale][1][i] <<= 2;
        }
    }
    return 0;
 fail:
    return -1;
}

static int dnxhd_init_rc(DNXHDEncContext *ctx)
{
    CHECKED_ALLOCZ(ctx->mb_rc, 8160*ctx->m.avctx->qmax*sizeof(RCEntry));
    if (ctx->m.avctx->mb_decision != FF_MB_DECISION_RD)
        CHECKED_ALLOCZ(ctx->mb_cmp, ctx->m.mb_num*sizeof(RCCMPEntry));

    ctx->frame_bits = (ctx->cid_table->coding_unit_size - 640 - 4) * 8;
    ctx->qscale = 1;
    ctx->lambda = 2<<LAMBDA_FRAC_BITS; // qscale 2
    return 0;
 fail:
    return -1;
}

static int dnxhd_encode_init(AVCodecContext *avctx)
{
    DNXHDEncContext *ctx = avctx->priv_data;
    int i, index;

    ctx->cid = ff_dnxhd_find_cid(avctx);
    if (!ctx->cid || avctx->pix_fmt != PIX_FMT_YUV422P) {
        av_log(avctx, AV_LOG_ERROR, "video parameters incompatible with DNxHD\n");
        return -1;
    }
    av_log(avctx, AV_LOG_DEBUG, "cid %d\n", ctx->cid);

    index = ff_dnxhd_get_cid_table(ctx->cid);
    ctx->cid_table = &ff_dnxhd_cid_table[index];

    ctx->m.avctx = avctx;
    ctx->m.mb_intra = 1;
    ctx->m.h263_aic = 1;

    dsputil_init(&ctx->m.dsp, avctx);
    ff_dct_common_init(&ctx->m);
    if (!ctx->m.dct_quantize)
        ctx->m.dct_quantize = dct_quantize_c;

    ctx->m.mb_height = (avctx->height + 15) / 16;
    ctx->m.mb_width  = (avctx->width  + 15) / 16;

    if (avctx->flags & CODEC_FLAG_INTERLACED_DCT) {
        ctx->interlaced = 1;
        ctx->m.mb_height /= 2;
    }

    ctx->m.mb_num = ctx->m.mb_height * ctx->m.mb_width;

    if (avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
        ctx->m.intra_quant_bias = avctx->intra_quant_bias;
    if (dnxhd_init_qmat(ctx, ctx->m.intra_quant_bias, 0) < 0) // XXX tune lbias/cbias
        return -1;

    if (dnxhd_init_vlc(ctx) < 0)
        return -1;
    if (dnxhd_init_rc(ctx) < 0)
        return -1;

    CHECKED_ALLOCZ(ctx->slice_size, ctx->m.mb_height*sizeof(uint32_t));
    CHECKED_ALLOCZ(ctx->mb_bits,    ctx->m.mb_num   *sizeof(uint16_t));
    CHECKED_ALLOCZ(ctx->mb_qscale,  ctx->m.mb_num   *sizeof(uint8_t));

    ctx->frame.key_frame = 1;
    ctx->frame.pict_type = FF_I_TYPE;
    ctx->m.avctx->coded_frame = &ctx->frame;

    if (avctx->thread_count > MAX_THREADS || (avctx->thread_count > ctx->m.mb_height)) {
        av_log(avctx, AV_LOG_ERROR, "too many threads\n");
        return -1;
    }

    ctx->thread[0] = ctx;
    for (i = 1; i < avctx->thread_count; i++) {
        ctx->thread[i] =  av_malloc(sizeof(DNXHDEncContext));
        memcpy(ctx->thread[i], ctx, sizeof(DNXHDEncContext));
    }

    for (i = 0; i < avctx->thread_count; i++) {
        ctx->thread[i]->m.start_mb_y = (ctx->m.mb_height*(i  ) + avctx->thread_count/2) / avctx->thread_count;
        ctx->thread[i]->m.end_mb_y   = (ctx->m.mb_height*(i+1) + avctx->thread_count/2) / avctx->thread_count;
    }

    return 0;
 fail: //for CHECKED_ALLOCZ
    return -1;
}

static int dnxhd_write_header(AVCodecContext *avctx, uint8_t *buf)
{
    DNXHDEncContext *ctx = avctx->priv_data;
    const uint8_t header_prefix[5] = { 0x00,0x00,0x02,0x80,0x01 };

    memcpy(buf, header_prefix, 5);
    buf[5] = ctx->interlaced ? ctx->cur_field+2 : 0x01;
    buf[6] = 0x80; // crc flag off
    buf[7] = 0xa0; // reserved
    AV_WB16(buf + 0x18, avctx->height); // ALPF
    AV_WB16(buf + 0x1a, avctx->width);  // SPL
    AV_WB16(buf + 0x1d, avctx->height); // NAL

    buf[0x21] = 0x38; // FIXME 8 bit per comp
    buf[0x22] = 0x88 + (ctx->frame.interlaced_frame<<2);
    AV_WB32(buf + 0x28, ctx->cid); // CID
    buf[0x2c] = ctx->interlaced ? 0 : 0x80;

    buf[0x5f] = 0x01; // UDL

    buf[0x167] = 0x02; // reserved
    AV_WB16(buf + 0x16a, ctx->m.mb_height * 4 + 4); // MSIPS
    buf[0x16d] = ctx->m.mb_height; // Ns
    buf[0x16f] = 0x10; // reserved

    ctx->msip = buf + 0x170;
    return 0;
}

static av_always_inline void dnxhd_encode_dc(DNXHDEncContext *ctx, int diff)
{
    int nbits;
    if (diff < 0) {
        nbits = av_log2_16bit(-2*diff);
        diff--;
    } else {
        nbits = av_log2_16bit(2*diff);
    }
    put_bits(&ctx->m.pb, ctx->cid_table->dc_bits[nbits] + nbits,
             (ctx->cid_table->dc_codes[nbits]<<nbits) + (diff & ((1 << nbits) - 1)));
}

static av_always_inline void dnxhd_encode_block(DNXHDEncContext *ctx, DCTELEM *block, int last_index, int n)
{
    int last_non_zero = 0;
    int slevel, i, j;

    dnxhd_encode_dc(ctx, block[0] - ctx->m.last_dc[n]);
    ctx->m.last_dc[n] = block[0];

    for (i = 1; i <= last_index; i++) {
        j = ctx->m.intra_scantable.permutated[i];
        slevel = block[j];
        if (slevel) {
            int run_level = i - last_non_zero - 1;
            int rlevel = (slevel<<1)|!!run_level;
            put_bits(&ctx->m.pb, ctx->vlc_bits[rlevel], ctx->vlc_codes[rlevel]);
            if (run_level)
                put_bits(&ctx->m.pb, ctx->run_bits[run_level], ctx->run_codes[run_level]);
            last_non_zero = i;
        }
    }
    put_bits(&ctx->m.pb, ctx->vlc_bits[0], ctx->vlc_codes[0]); // EOB
}

static av_always_inline void dnxhd_unquantize_c(DNXHDEncContext *ctx, DCTELEM *block, int n, int qscale, int last_index)
{
    const uint8_t *weight_matrix;
    int level;
    int i;

    weight_matrix = (n&2) ? ctx->cid_table->chroma_weight : ctx->cid_table->luma_weight;

    for (i = 1; i <= last_index; i++) {
        int j = ctx->m.intra_scantable.permutated[i];
        level = block[j];
        if (level) {
            if (level < 0) {
                level = (1-2*level) * qscale * weight_matrix[i];
                if (weight_matrix[i] != 32)
                    level += 32;
                level >>= 6;
                level = -level;
            } else {
                level = (2*level+1) * qscale * weight_matrix[i];
                if (weight_matrix[i] != 32)
                    level += 32;
                level >>= 6;
            }
            block[j] = level;
        }
    }
}

static av_always_inline int dnxhd_ssd_block(DCTELEM *qblock, DCTELEM *block)
{
    int score = 0;
    int i;
    for (i = 0; i < 64; i++)
        score += (block[i]-qblock[i])*(block[i]-qblock[i]);
    return score;
}

static av_always_inline int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, DCTELEM *block, int last_index)
{
    int last_non_zero = 0;
    int bits = 0;
    int i, j, level;
    for (i = 1; i <= last_index; i++) {
        j = ctx->m.intra_scantable.permutated[i];
        level = block[j];
        if (level) {
            int run_level = i - last_non_zero - 1;
            bits += ctx->vlc_bits[(level<<1)|!!run_level]+ctx->run_bits[run_level];
            last_non_zero = i;
        }
    }
    return bits;
}

static av_always_inline void dnxhd_get_pixels_4x8(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
{
    int i;
    for (i = 0; i < 4; i++) {
        block[0] = pixels[0];
        block[1] = pixels[1];
        block[2] = pixels[2];
        block[3] = pixels[3];
        block[4] = pixels[4];
        block[5] = pixels[5];
        block[6] = pixels[6];
        block[7] = pixels[7];
        pixels += line_size;
        block += 8;
    }
    memcpy(block   , block- 8, sizeof(*block)*8);
    memcpy(block+ 8, block-16, sizeof(*block)*8);
    memcpy(block+16, block-24, sizeof(*block)*8);
    memcpy(block+24, block-32, sizeof(*block)*8);
}

static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
{
    const uint8_t *ptr_y = ctx->thread[0]->src[0] + ((mb_y << 4) * ctx->m.linesize)   + (mb_x << 4);
    const uint8_t *ptr_u = ctx->thread[0]->src[1] + ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << 3);
    const uint8_t *ptr_v = ctx->thread[0]->src[2] + ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << 3);
    DSPContext *dsp = &ctx->m.dsp;

    dsp->get_pixels(ctx->blocks[0], ptr_y    , ctx->m.linesize);
    dsp->get_pixels(ctx->blocks[1], ptr_y + 8, ctx->m.linesize);
    dsp->get_pixels(ctx->blocks[2], ptr_u    , ctx->m.uvlinesize);
    dsp->get_pixels(ctx->blocks[3], ptr_v    , ctx->m.uvlinesize);

    if (mb_y+1 == ctx->m.mb_height && ctx->m.avctx->height == 1080) {
        if (ctx->interlaced) {
            dnxhd_get_pixels_4x8(ctx->blocks[4], ptr_y + ctx->dct_y_offset    , ctx->m.linesize);
            dnxhd_get_pixels_4x8(ctx->blocks[5], ptr_y + ctx->dct_y_offset + 8, ctx->m.linesize);
            dnxhd_get_pixels_4x8(ctx->blocks[6], ptr_u + ctx->dct_uv_offset   , ctx->m.uvlinesize);
            dnxhd_get_pixels_4x8(ctx->blocks[7], ptr_v + ctx->dct_uv_offset   , ctx->m.uvlinesize);
        } else
            memset(ctx->blocks[4], 0, 4*64*sizeof(DCTELEM));
    } else {
        dsp->get_pixels(ctx->blocks[4], ptr_y + ctx->dct_y_offset    , ctx->m.linesize);
        dsp->get_pixels(ctx->blocks[5], ptr_y + ctx->dct_y_offset + 8, ctx->m.linesize);
        dsp->get_pixels(ctx->blocks[6], ptr_u + ctx->dct_uv_offset   , ctx->m.uvlinesize);
        dsp->get_pixels(ctx->blocks[7], ptr_v + ctx->dct_uv_offset   , ctx->m.uvlinesize);
    }
}

static av_always_inline int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
{
    if (i&2) {
        ctx->m.q_intra_matrix16 = ctx->qmatrix_c16;