layer3.c

基于mips架构的ATI-XILLEON 226的mp3解码程序

/*
 * NAME:	III_imdct_l()
 * DESCRIPTION:	perform IMDCT and windowing for long blocks
 */
static
void III_imdct_l(mad_fixed_t const X[18], mad_fixed_t z[36],
		 unsigned int block_type)
{
  unsigned int i;

  /* IMDCT */

  imdct36(X, z);

  /* windowing */

  switch (block_type) {
  case 0:  /* normal window */
# if defined(ASO_INTERLEAVE1)
    {
      register mad_fixed_t tmp1, tmp2;

      tmp1 = window_l[0];
      tmp2 = window_l[1];

      for (i = 0; i < 34; i += 2) {
	z[i + 0] = mad_f_mul(z[i + 0], tmp1);
	tmp1 = window_l[i + 2];
	z[i + 1] = mad_f_mul(z[i + 1], tmp2);
	tmp2 = window_l[i + 3];
      }

      z[34] = mad_f_mul(z[34], tmp1);
      z[35] = mad_f_mul(z[35], tmp2);
    }
# elif defined(ASO_INTERLEAVE2)
    {
      register mad_fixed_t tmp1, tmp2;

      tmp1 = z[0];
      tmp2 = window_l[0];

      for (i = 0; i < 35; ++i) {
	z[i] = mad_f_mul(tmp1, tmp2);
	tmp1 = z[i + 1];
	tmp2 = window_l[i + 1];
      }

      z[35] = mad_f_mul(tmp1, tmp2);
    }
# elif 1
    for (i = 0; i < 36; i += 4) {
      z[i + 0] = mad_f_mul(z[i + 0], window_l[i + 0]);
      z[i + 1] = mad_f_mul(z[i + 1], window_l[i + 1]);
      z[i + 2] = mad_f_mul(z[i + 2], window_l[i + 2]);
      z[i + 3] = mad_f_mul(z[i + 3], window_l[i + 3]);
    }
# else
    for (i =  0; i < 36; ++i) z[i] = mad_f_mul(z[i], window_l[i]);
# endif
    break;

  case 1:  /* start block */
    for (i =  0; i < 18; ++i) z[i] = mad_f_mul(z[i], window_l[i]);
    /*  (i = 18; i < 24; ++i) z[i] unchanged */
    for (i = 24; i < 30; ++i) z[i] = mad_f_mul(z[i], window_s[i - 18]);
    for (i = 30; i < 36; ++i) z[i] = 0;
    break;

  case 3:  /* stop block */
    for (i =  0; i <  6; ++i) z[i] = 0;
    for (i =  6; i < 12; ++i) z[i] = mad_f_mul(z[i], window_s[i - 6]);
    /*  (i = 12; i < 18; ++i) z[i] unchanged */
    for (i = 18; i < 36; ++i) z[i] = mad_f_mul(z[i], window_l[i]);
    break;
  }
}
# endif  /* ASO_IMDCT */

/*
 * NAME:	III_imdct_s()
 * DESCRIPTION:	perform IMDCT and windowing for short blocks
 */
static
void III_imdct_s(mad_fixed_t const X[18], mad_fixed_t z[36])
{
  mad_fixed_t y[36], *yptr;
  mad_fixed_t const *wptr;
  int w, i;
  register mad_fixed64hi_t hi;
  register mad_fixed64lo_t lo;

  /* IMDCT */

  yptr = &y[0];

  for (w = 0; w < 3; ++w) {
    register mad_fixed_t const (*s)[6];

    s = imdct_s;

    for (i = 0; i < 3; ++i) {
      MAD_F_ML0(hi, lo, X[0], (*s)[0]);
      MAD_F_MLA(hi, lo, X[1], (*s)[1]);
      MAD_F_MLA(hi, lo, X[2], (*s)[2]);
      MAD_F_MLA(hi, lo, X[3], (*s)[3]);
      MAD_F_MLA(hi, lo, X[4], (*s)[4]);
      MAD_F_MLA(hi, lo, X[5], (*s)[5]);

      yptr[i + 0] = MAD_F_MLZ(hi, lo);
      yptr[5 - i] = -yptr[i + 0];

      ++s;

      MAD_F_ML0(hi, lo, X[0], (*s)[0]);
      MAD_F_MLA(hi, lo, X[1], (*s)[1]);
      MAD_F_MLA(hi, lo, X[2], (*s)[2]);
      MAD_F_MLA(hi, lo, X[3], (*s)[3]);
      MAD_F_MLA(hi, lo, X[4], (*s)[4]);
      MAD_F_MLA(hi, lo, X[5], (*s)[5]);

      yptr[ i + 6] = MAD_F_MLZ(hi, lo);
      yptr[11 - i] = yptr[i + 6];

      ++s;
    }

    yptr += 12;
    X    += 6;
  }

  /* windowing, overlapping and concatenation */

  yptr = &y[0];
  wptr = &window_s[0];

  for (i = 0; i < 6; ++i) {
    z[i +  0] = 0;
    z[i +  6] = mad_f_mul(yptr[ 0 + 0], wptr[0]);

    MAD_F_ML0(hi, lo, yptr[ 0 + 6], wptr[6]);
    MAD_F_MLA(hi, lo, yptr[12 + 0], wptr[0]);

    z[i + 12] = MAD_F_MLZ(hi, lo);

    MAD_F_ML0(hi, lo, yptr[12 + 6], wptr[6]);
    MAD_F_MLA(hi, lo, yptr[24 + 0], wptr[0]);

    z[i + 18] = MAD_F_MLZ(hi, lo);

    z[i + 24] = mad_f_mul(yptr[24 + 6], wptr[6]);
    z[i + 30] = 0;

    ++yptr;
    ++wptr;
  }
}

/*
 * NAME:	III_overlap()
 * DESCRIPTION:	perform overlap-add of windowed IMDCT outputs
 */
static
void III_overlap(mad_fixed_t const output[36], mad_fixed_t overlap[18],
		 mad_fixed_t sample[18][32], unsigned int sb)
{
  unsigned int i;

# if defined(ASO_INTERLEAVE2)
  {
    register mad_fixed_t tmp1, tmp2;

    tmp1 = overlap[0];
    tmp2 = overlap[1];

    for (i = 0; i < 16; i += 2) {
      sample[i + 0][sb] = output[i + 0] + tmp1;
      overlap[i + 0]    = output[i + 0 + 18];
      tmp1 = overlap[i + 2];

      sample[i + 1][sb] = output[i + 1] + tmp2;
      overlap[i + 1]    = output[i + 1 + 18];
      tmp2 = overlap[i + 3];
    }

    sample[16][sb] = output[16] + tmp1;
    overlap[16]    = output[16 + 18];
    sample[17][sb] = output[17] + tmp2;
    overlap[17]    = output[17 + 18];
  }
# elif 0
  for (i = 0; i < 18; i += 2) {
    sample[i + 0][sb] = output[i + 0] + overlap[i + 0];
    overlap[i + 0]    = output[i + 0 + 18];

    sample[i + 1][sb] = output[i + 1] + overlap[i + 1];
    overlap[i + 1]    = output[i + 1 + 18];
  }
# else
  for (i = 0; i < 18; ++i) {
    sample[i][sb] = output[i] + overlap[i];
    overlap[i]    = output[i + 18];
  }
# endif
}

/*
 * NAME:	III_overlap_z()
 * DESCRIPTION:	perform "overlap-add" of zero IMDCT outputs
 */
static inline
void III_overlap_z(mad_fixed_t overlap[18],
		   mad_fixed_t sample[18][32], unsigned int sb)
{
  unsigned int i;

# if defined(ASO_INTERLEAVE2)
  {
    register mad_fixed_t tmp1, tmp2;

    tmp1 = overlap[0];
    tmp2 = overlap[1];

    for (i = 0; i < 16; i += 2) {
      sample[i + 0][sb] = tmp1;
      overlap[i + 0]    = 0;
      tmp1 = overlap[i + 2];

      sample[i + 1][sb] = tmp2;
      overlap[i + 1]    = 0;
      tmp2 = overlap[i + 3];
    }

    sample[16][sb] = tmp1;
    overlap[16]    = 0;
    sample[17][sb] = tmp2;
    overlap[17]    = 0;
  }
# else
  for (i = 0; i < 18; ++i) {
    sample[i][sb] = overlap[i];
    overlap[i]    = 0;
  }
# endif
}

/*
 * NAME:	III_freqinver()
 * DESCRIPTION:	perform subband frequency inversion for odd sample lines
 */
static
void III_freqinver(mad_fixed_t sample[18][32], unsigned int sb)
{
  unsigned int i;

# if 1 || defined(ASO_INTERLEAVE1) || defined(ASO_INTERLEAVE2)
  {
    register mad_fixed_t tmp1, tmp2;

    tmp1 = sample[1][sb];
    tmp2 = sample[3][sb];

    for (i = 1; i < 13; i += 4) {
      sample[i + 0][sb] = -tmp1;
      tmp1 = sample[i + 4][sb];
      sample[i + 2][sb] = -tmp2;
      tmp2 = sample[i + 6][sb];
    }

    sample[13][sb] = -tmp1;
    tmp1 = sample[17][sb];
    sample[15][sb] = -tmp2;
    sample[17][sb] = -tmp1;
  }
# else
  for (i = 1; i < 18; i += 2)
    sample[i][sb] = -sample[i][sb];
# endif
}

/*
 * NAME:	III_decode()
 * DESCRIPTION:	decode frame main_data
 */
static
enum mad_error III_decode(struct mad_bitptr *ptr, struct mad_frame *frame,
			  struct sideinfo *si, unsigned int nch)
{
  struct mad_header *header = &frame->header;
  unsigned int sfreqi, ngr, gr;

  {
    unsigned int sfreq;

    sfreq = header->samplerate;
    if (header->flags & MAD_FLAG_MPEG_2_5_EXT)
      sfreq *= 2;

    /* 48000 => 0, 44100 => 1, 32000 => 2,
       24000 => 3, 22050 => 4, 16000 => 5 */
    sfreqi = ((sfreq >>  7) & 0x000f) +
             ((sfreq >> 15) & 0x0001) - 8;

    if (header->flags & MAD_FLAG_MPEG_2_5_EXT)
      sfreqi += 3;
  }

  /* scalefactors, Huffman decoding, requantization */

  ngr = (header->flags & MAD_FLAG_LSF_EXT) ? 1 : 2;

  for (gr = 0; gr < ngr; ++gr) {
    struct granule *granule = &si->gr[gr];
    unsigned char const *sfbwidth[2];
    mad_fixed_t xr[2][576];
    unsigned int ch;
    enum mad_error error;

    for (ch = 0; ch < nch; ++ch) {
      struct channel *channel = &granule->ch[ch];
      unsigned int part2_length;

      sfbwidth[ch] = sfbwidth_table[sfreqi].l;
      if (channel->block_type == 2) {
	sfbwidth[ch] = (channel->flags & mixed_block_flag) ?
	  sfbwidth_table[sfreqi].m : sfbwidth_table[sfreqi].s;
      }

      if (header->flags & MAD_FLAG_LSF_EXT) {
	part2_length = III_scalefactors_lsf(ptr, channel,
					    ch == 0 ? 0 : &si->gr[1].ch[1],
					    header->mode_extension);
      }
      else {
	part2_length = III_scalefactors(ptr, channel, &si->gr[0].ch[ch],
					gr == 0 ? 0 : si->scfsi[ch]);
      }

      error = III_huffdecode(ptr, xr[ch], channel, sfbwidth[ch], part2_length);
      if (error)
	return error;
    }

    /* joint stereo processing */

    if (header->mode == MAD_MODE_JOINT_STEREO && header->mode_extension) {
      error = III_stereo(xr, granule, header, sfbwidth[0]);
      if (error)
	return error;
    }

    /* reordering, alias reduction, IMDCT, overlap-add, frequency inversion */

    for (ch = 0; ch < nch; ++ch) {
      struct channel const *channel = &granule->ch[ch];
      mad_fixed_t (*sample)[32] = &frame->sbsample[ch][18 * gr];
      unsigned int sb, l, i, sblimit;
      mad_fixed_t output[36];

      if (channel->block_type == 2) {
	III_reorder(xr[ch], channel, sfbwidth[ch]);

# if !defined(OPT_STRICT)
	/*
	 * According to ISO/IEC 11172-3, "Alias reduction is not applied for
	 * granules with block_type == 2 (short block)." However, other
	 * sources suggest alias reduction should indeed be performed on the
	 * lower two subbands of mixed blocks. Most other implementations do
	 * this, so by default we will too.
	 */
	if (channel->flags & mixed_block_flag)
	  III_aliasreduce(xr[ch], 36);
# endif
      }
      else
	III_aliasreduce(xr[ch], 576);

      l = 0;

      /* subbands 0-1 */

      if (channel->block_type != 2 || (channel->flags & mixed_block_flag)) {
	unsigned int block_type;

	block_type = channel->block_type;
	if (channel->flags & mixed_block_flag)
	  block_type = 0;

	/* long blocks */
	for (sb = 0; sb < 2; ++sb, l += 18) {
	  III_imdct_l(&xr[ch][l], output, block_type);
	  III_overlap(output, (*frame->overlap)[ch][sb], sample, sb);
	}
      }
      else {
	/* short blocks */
	for (sb = 0; sb < 2; ++sb, l += 18) {
	  III_imdct_s(&xr[ch][l], output);
	  III_overlap(output, (*frame->overlap)[ch][sb], sample, sb);
	}
      }

      III_freqinver(sample, 1);

      /* (nonzero) subbands 2-31 */

      i = 576;
      while (i > 36 && xr[ch][i - 1] == 0)
	--i;

      sblimit = 32 - (576 - i) / 18;

      if (channel->block_type != 2) {
	/* long blocks */
	for (sb = 2; sb < sblimit; ++sb, l += 18) {
	  III_imdct_l(&xr[ch][l], output, channel->block_type);
	  III_overlap(output, (*frame->overlap)[ch][sb], sample, sb);

	  if (sb & 1)
	    III_freqinver(sample, sb);
	}
      }
      else {
	/* short blocks */
	for (sb = 2; sb < sblimit; ++sb, l += 18) {
	  III_imdct_s(&xr[ch][l], output);
	  III_overlap(output, (*frame->overlap)[ch][sb], sample, sb);

	  if (sb & 1)
	    III_freqinver(sample, sb);
	}
      }

      /* remaining (zero) subbands */

      for (sb = sblimit; sb < 32; ++sb) {
	III_overlap_z((*frame->overlap)[ch][sb], sample, sb);

	if (sb & 1)
	  III_freqinver(sample, sb);
      }
    }
  }

  return MAD_ERROR_NONE;
}

/*
 * NAME:	layer->III()
 * DESCRIPTION:	decode a single Layer III frame
 */
int mad_layer_III(struct mad_stream *stream, struct mad_frame *frame)
{
  struct mad_header *header = &frame->header;
  unsigned int nch, priv_bitlen, next_md_begin = 0;
  unsigned int si_len, data_bitlen, md_len;
  unsigned int frame_space, frame_used, frame_free;
  struct mad_bitptr ptr;
  struct sideinfo si;
  enum mad_error error;
  int result = 0;