decoder.c

mpeg4 video codec mpeg4 video codec

/*****************************************************************************
 *
 *  XVID MPEG-4 VIDEO CODEC
 *  - Decoder Module -
 *
 *  Copyright(C) 2002      MinChen <[e-mail]>
 *               2002-2004 Peter Ross <[e-mail]>
 *
 *  This program is free software ; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation ; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program 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 General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program ; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 *
 * $Id: decoder.c,v 1.1 2006/02/23 15:13:27 kevin-fu Exp $
 *
 ****************************************************************************/

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#ifdef BFRAMES_DEC_DEBUG
#define BFRAMES_DEC
#endif

#include "xvid.h"
#include "portab.h"
#include "global.h"

#include "decoder.h"
#include "bitstream/bitstream.h"
#include "bitstream/mbcoding.h"

#include "quant/quant.h"
#include "quant/quant_matrix.h"
#include "dct/idct.h"
#include "dct/fdct.h"
#include "utils/mem_transfer.h"
#include "image/interpolate8x8.h"
#include "image/font.h"
#include "image/qpel.h"

#include "bitstream/mbcoding.h"
#include "prediction/mbprediction.h"
#include "utils/timer.h"
#include "utils/emms.h"
#include "motion/motion.h"
#include "motion/gmc.h"

#include "image/image.h"
#include "image/colorspace.h"
#include "image/postprocessing.h"
#include "utils/mem_align.h"

#define DIV2ROUND(n)  (((n)>>1)|((n)&1))
#define DIV2(n)       ((n)>>1)
#define DIVUVMOV(n) (((n) >> 1) + roundtab_79[(n) & 0x3]) //

static int
decoder_resize(DECODER * dec)
{
	/* free existing */
	image_destroy(&dec->cur, dec->edged_width, dec->edged_height);
	image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height);
	image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height);
	image_destroy(&dec->tmp, dec->edged_width, dec->edged_height);
	image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height);

	image_destroy(&dec->gmc, dec->edged_width, dec->edged_height);

  image_null(&dec->cur);
  image_null(&dec->refn[0]);
  image_null(&dec->refn[1]);
  image_null(&dec->tmp);
  image_null(&dec->qtmp);
  image_null(&dec->gmc);


  xvid_free(dec->last_mbs);
  xvid_free(dec->mbs);
  xvid_free(dec->qscale);
  dec->last_mbs = NULL;
  dec->mbs = NULL;
  dec->qscale = NULL;

	/* realloc */
	dec->mb_width = (dec->width + 15) / 16;
	dec->mb_height = (dec->height + 15) / 16;

	dec->edged_width = 16 * dec->mb_width + 2 * EDGE_SIZE;
	dec->edged_height = 16 * dec->mb_height + 2 * EDGE_SIZE;

	if (   image_create(&dec->cur, dec->edged_width, dec->edged_height) 
	    || image_create(&dec->refn[0], dec->edged_width, dec->edged_height)
	    || image_create(&dec->refn[1], dec->edged_width, dec->edged_height) 	/* Support B-frame to reference last 2 frame */
	    || image_create(&dec->tmp, dec->edged_width, dec->edged_height)
	    || image_create(&dec->qtmp, dec->edged_width, dec->edged_height)
      || image_create(&dec->gmc, dec->edged_width, dec->edged_height) )
    goto memory_error;

	dec->mbs =
		xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height,
					CACHE_LINE);
	if (dec->mbs == NULL)
	  goto memory_error;
	memset(dec->mbs, 0, sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height);

	/* For skip MB flag */
	dec->last_mbs =
		xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height,
					CACHE_LINE);
	if (dec->last_mbs == NULL)
	  goto memory_error;
	memset(dec->last_mbs, 0, sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height);

	/* nothing happens if that fails */
	dec->qscale =
		xvid_malloc(sizeof(int) * dec->mb_width * dec->mb_height, CACHE_LINE);
	
	if (dec->qscale)
		memset(dec->qscale, 0, sizeof(int) * dec->mb_width * dec->mb_height);

	return 0;

memory_error:
        /* Most structures were deallocated / nullifieded, so it should be safe */
        /* decoder_destroy(dec) minus the write_timer */
  xvid_free(dec->mbs);
  image_destroy(&dec->cur, dec->edged_width, dec->edged_height);
  image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height);
  image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height);
  image_destroy(&dec->tmp, dec->edged_width, dec->edged_height);
  image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height);

  xvid_free(dec);
  return XVID_ERR_MEMORY;
}


int
decoder_create(xvid_dec_create_t * create)
{
  DECODER *dec;

  if (XVID_VERSION_MAJOR(create->version) != 1) /* v1.x.x */
    return XVID_ERR_VERSION;

  dec = xvid_malloc(sizeof(DECODER), CACHE_LINE);
  if (dec == NULL) {
    return XVID_ERR_MEMORY;
  }

  memset(dec, 0, sizeof(DECODER));

  dec->mpeg_quant_matrices = xvid_malloc(sizeof(uint16_t) * 64 * 8, CACHE_LINE);
  if (dec->mpeg_quant_matrices == NULL) {
    xvid_free(dec);
    return XVID_ERR_MEMORY;
  }

  create->handle = dec;

  dec->width = create->width;
  dec->height = create->height;

  image_null(&dec->cur);
  image_null(&dec->refn[0]);
  image_null(&dec->refn[1]);
  image_null(&dec->tmp);
  image_null(&dec->qtmp);

  /* image based GMC */
  image_null(&dec->gmc);

  dec->mbs = NULL;
  dec->last_mbs = NULL;
  dec->qscale = NULL;

  init_timer();
  init_postproc(&dec->postproc);
  init_mpeg_matrix(dec->mpeg_quant_matrices);

  /* For B-frame support (used to save reference frame's time */
  dec->frames = 0;
  dec->time = dec->time_base = dec->last_time_base = 0;
  dec->low_delay = 0;
  dec->packed_mode = 0;
  dec->time_inc_resolution = 1; /* until VOL header says otherwise */

  dec->bs_version = 0xffff; /* Initialize to very high value -> assume bugfree stream */

  dec->fixed_dimensions = (dec->width > 0 && dec->height > 0);

  if (dec->fixed_dimensions)
    return decoder_resize(dec);
  else
    return 0;
}


int
decoder_destroy(DECODER * dec)
{
  xvid_free(dec->last_mbs);
  xvid_free(dec->mbs);
  xvid_free(dec->qscale);

  /* image based GMC */
  image_destroy(&dec->gmc, dec->edged_width, dec->edged_height);

  image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height);
  image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height);
  image_destroy(&dec->tmp, dec->edged_width, dec->edged_height);
  image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height);
  image_destroy(&dec->cur, dec->edged_width, dec->edged_height);
  xvid_free(dec->mpeg_quant_matrices);
  xvid_free(dec);

  write_timer();
  return 0;
}

static const int32_t dquant_table[4] = {
  -1, -2, 1, 2
};

/* decode an intra macroblock */
static void
decoder_mbintra(DECODER * dec,
        MACROBLOCK * pMB,
        const uint32_t x_pos,
        const uint32_t y_pos,
        const uint32_t acpred_flag,
        const uint32_t cbp,
        Bitstream * bs,
        const uint32_t quant,
        const uint32_t intra_dc_threshold,
        const unsigned int bound)
{

  DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE);
  DECLARE_ALIGNED_MATRIX(data, 6, 64, int16_t, CACHE_LINE);

  uint32_t stride = dec->edged_width;
  uint32_t stride2 = stride / 2;
  uint32_t next_block = stride * 8;
  uint32_t i;
  uint32_t iQuant = pMB->quant;
  uint8_t *pY_Cur, *pU_Cur, *pV_Cur;

  pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4);
  pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3);
  pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3);

  memset(block, 0, 6 * 64 * sizeof(int16_t)); /* clear */

  for (i = 0; i < 6; i++) {
    uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4);
    int16_t predictors[8];
    int start_coeff;

    start_timer();
    predict_acdc(dec->mbs, x_pos, y_pos, dec->mb_width, i, &block[i * 64],
           iQuant, iDcScaler, predictors, bound);
    if (!acpred_flag) {
      pMB->acpred_directions[i] = 0;
    }
    stop_prediction_timer();

    if (quant < intra_dc_threshold) {
      int dc_size;
      int dc_dif;

      dc_size = i < 4 ? get_dc_size_lum(bs) : get_dc_size_chrom(bs);
      dc_dif = dc_size ? get_dc_dif(bs, dc_size) : 0;

      if (dc_size > 8) {
        BitstreamSkip(bs, 1); /* marker */
      }

      block[i * 64 + 0] = dc_dif;
      start_coeff = 1;

      DPRINTF(XVID_DEBUG_COEFF,"block[0] %i\n", dc_dif);
    } else {
      start_coeff = 0;
    }

    start_timer();
    if (cbp & (1 << (5 - i))) /* coded */
    {
      int direction = dec->alternate_vertical_scan ?
        2 : pMB->acpred_directions[i];

      get_intra_block(bs, &block[i * 64], direction, start_coeff);
    }
    stop_coding_timer();

    start_timer();
    add_acdc(pMB, i, &block[i * 64], iDcScaler, predictors, dec->bs_version);
    stop_prediction_timer();

    start_timer();
    if (dec->quant_type == 0) {
      dequant_h263_intra(&data[i * 64], &block[i * 64], iQuant, iDcScaler, dec->mpeg_quant_matrices);
    } else {
      dequant_mpeg_intra(&data[i * 64], &block[i * 64], iQuant, iDcScaler, dec->mpeg_quant_matrices);
    }
    stop_iquant_timer();

    start_timer();
    idct((short * const)&data[i * 64]);
    stop_idct_timer();

  }

  if (dec->interlacing && pMB->field_dct) {
    next_block = stride;
    stride *= 2;
  }

  start_timer();
  transfer_16to8copy(pY_Cur, &data[0 * 64], stride);
  transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride);
  transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride);
  transfer_16to8copy(pY_Cur + 8 + next_block, &data[3 * 64], stride);
  transfer_16to8copy(pU_Cur, &data[4 * 64], stride2);
  transfer_16to8copy(pV_Cur, &data[5 * 64], stride2);
  stop_transfer_timer();
}

static void
decoder_mb_decode(DECODER * dec,
        const uint32_t cbp,
        Bitstream * bs,
        uint8_t * pY_Cur,
        uint8_t * pU_Cur,
        uint8_t * pV_Cur,
        const MACROBLOCK * pMB)
{
  DECLARE_ALIGNED_MATRIX(data, 1, 64, int16_t, CACHE_LINE);

  int stride = dec->edged_width;
  int i;
  const uint32_t iQuant = pMB->quant;
  const int direction = dec->alternate_vertical_scan ? 2 : 0;
  typedef void (*get_inter_block_function_t)(
      Bitstream * bs,
      int16_t * block,
      int direction,
      const int quant,
      const uint16_t *matrix);
  typedef void (*add_residual_function_t)(
      uint8_t *predicted_block,
      const int16_t *residual,
      int stride);

  const get_inter_block_function_t get_inter_block = (dec->quant_type == 0)
    ? (get_inter_block_function_t)get_inter_block_h263
    : (get_inter_block_function_t)get_inter_block_mpeg;

  uint8_t *dst[6];
  int strides[6];


  if (dec->interlacing && pMB->field_dct) {
    dst[0] = pY_Cur;
    dst[1] = pY_Cur + 8;
    dst[2] = pY_Cur + stride;
    dst[3] = dst[2] + 8;
    dst[4] = pU_Cur;
    dst[5] = pV_Cur;
    strides[0] = strides[1] = strides[2] = strides[3] = stride*2;
    strides[4] = stride/2;
    strides[5] = stride/2;
  } else {
    dst[0] = pY_Cur;
    dst[1] = pY_Cur + 8;
    dst[2] = pY_Cur + 8*stride;
    dst[3] = dst[2] + 8;
    dst[4] = pU_Cur;
    dst[5] = pV_Cur;
    strides[0] = strides[1] = strides[2] = strides[3] = stride;
    strides[4] = stride/2;
    strides[5] = stride/2;
  }

  for (i = 0; i < 6; i++) {
    /* Process only coded blocks */
    if (cbp & (1 << (5 - i))) {

      /* Clear the block */
      memset(&data[0], 0, 64*sizeof(int16_t));

      /* Decode coeffs and dequantize on the fly */
      start_timer();
      get_inter_block(bs, &data[0], direction, iQuant, get_inter_matrix(dec->mpeg_quant_matrices));
      stop_coding_timer();

      /* iDCT */
      start_timer();
      idct((short * const)&data[0]);
      stop_idct_timer();

      /* Add this residual to the predicted block */
      start_timer();
      transfer_16to8add(dst[i], &data[0], strides[i]);
      stop_transfer_timer();
    }
  }
}

static void __inline
validate_vector(VECTOR * mv, unsigned int x_pos, unsigned int y_pos, const DECODER * dec)
{
  /* clip a vector to valid range
     prevents crashes if bitstream is broken
  */
  int shift = 5 + dec->quarterpel;
  int xborder_high = (int)(dec->mb_width - x_pos) << shift;
  int xborder_low = (-(int)x_pos-1) << shift;
  int yborder_high = (int)(dec->mb_height - y_pos) << shift;
  int yborder_low = (-(int)y_pos-1) << shift;

#define CHECK_MV(mv) \
  do { \
  if ((mv).x > xborder_high) { \
    DPRINTF(XVID_DEBUG_MV, "mv.x > max -- %d > %d, MB %d, %d", (mv).x, xborder_high, x_pos, y_pos); \
    (mv).x = xborder_high; \
  } else if ((mv).x < xborder_low) { \
    DPRINTF(XVID_DEBUG_MV, "mv.x < min -- %d < %d, MB %d, %d", (mv).x, xborder_low, x_pos, y_pos); \