Alien-FreeImage

 view release on metacpan or  search on metacpan

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS);

    dataptr++;			/* advance pointer to next column */
  }
}


/*
 * Perform the forward DCT on a 6x6 sample block.
 */

GLOBAL(void)
jpeg_fdct_6x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2;
  INT32 tmp10, tmp11, tmp12;
  DCTELEM *dataptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pre-zero output coefficient block. */
  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * cK represents sqrt(2) * cos(K*pi/12).
   */

  dataptr = data;
  for (ctr = 0; ctr < 6; ctr++) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);

    tmp10 = tmp0 + tmp2;
    tmp12 = tmp0 - tmp2;

    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS);
    dataptr[2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
	      CONST_BITS-PASS1_BITS);
    dataptr[4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
	      CONST_BITS-PASS1_BITS);

    /* Odd part */

    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
		    CONST_BITS-PASS1_BITS);

    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS));
    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS);
    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS));

    dataptr += DCTSIZE;		/* advance pointer to next row */
  }

  /* Pass 2: process columns.
   * We remove the PASS1_BITS scaling, but leave the results scaled up
   * by an overall factor of 8.
   * We must also scale the output by (8/6)**2 = 16/9, which we fold
   * into the constant multipliers:
   * cK now represents sqrt(2) * cos(K*pi/12) * 16/9.
   */

  dataptr = data;
  for (ctr = 0; ctr < 6; ctr++) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5];
    tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];

    tmp10 = tmp0 + tmp2;
    tmp12 = tmp0 - tmp2;

    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5];
    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)),         /* 16/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp12, FIX(2.177324216)),                 /* c2 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */
	      CONST_BITS+PASS1_BITS);

    /* Odd part */

    tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829));             /* c5 */

    dataptr[DCTSIZE*1] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),   /* 16/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*3] = (DCTELEM)
      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)),    /* 16/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*5] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)),   /* 16/9 */
	      CONST_BITS+PASS1_BITS);

    dataptr++;			/* advance pointer to next column */
  }
}

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

  }
}


/*
 * Perform the forward DCT on a 12x12 sample block.
 */

GLOBAL(void)
jpeg_fdct_12x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  DCTELEM workspace[8*4];
  DCTELEM *dataptr;
  DCTELEM *wsptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT.
   * cK represents sqrt(2) * cos(K*pi/24).
   */

  dataptr = data;
  ctr = 0;
  for (;;) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]);
    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]);
    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]);
    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]);
    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]);

    tmp10 = tmp0 + tmp5;
    tmp13 = tmp0 - tmp5;
    tmp11 = tmp1 + tmp4;
    tmp14 = tmp1 - tmp4;
    tmp12 = tmp2 + tmp3;
    tmp15 = tmp2 - tmp3;

    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]);
    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]);
    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]);
    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]);
    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]);
    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]);

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM) (tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE);
    dataptr[6] = (DCTELEM) (tmp13 - tmp14 - tmp15);
    dataptr[4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */
	      CONST_BITS);
    dataptr[2] = (DCTELEM)
      DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */
	      CONST_BITS);

    /* Odd part */

    tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100);    /* c9 */
    tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865);   /* c3-c9 */
    tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065);   /* c3+c9 */
    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054));   /* c5 */
    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669));   /* c7 */
    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */
	    + MULTIPLY(tmp5, FIX(0.184591911));        /* c11 */
    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */
    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */
	    + MULTIPLY(tmp5, FIX(0.860918669));        /* c7 */
    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */
	    - MULTIPLY(tmp5, FIX(1.121971054));        /* c5 */
    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */
	    - MULTIPLY(tmp2 + tmp5, FIX_0_541196100);  /* c9 */

    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS);
    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS);
    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS);
    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS);

    ctr++;

    if (ctr != DCTSIZE) {
      if (ctr == 12)
	break;			/* Done. */
      dataptr += DCTSIZE;	/* advance pointer to next row */
    } else
      dataptr = workspace;	/* switch pointer to extended workspace */
  }

  /* Pass 2: process columns.
   * We leave the results scaled up by an overall factor of 8.
   * We must also scale the output by (8/12)**2 = 4/9, which we partially
   * fold into the constant multipliers and final shifting:
   * cK now represents sqrt(2) * cos(K*pi/24) * 8/9.
   */

  dataptr = data;
  wsptr = workspace;
  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3];
    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2];
    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1];
    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0];
    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7];
    tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6];

    tmp10 = tmp0 + tmp5;
    tmp13 = tmp0 - tmp5;
    tmp11 = tmp1 + tmp4;
    tmp14 = tmp1 - tmp4;
    tmp12 = tmp2 + tmp3;
    tmp15 = tmp2 - tmp3;

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

    tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2];
    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1];
    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0];
    tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];

    tmp10 = tmp0 + tmp6;
    tmp14 = tmp0 - tmp6;
    tmp11 = tmp1 + tmp5;
    tmp15 = tmp1 - tmp5;
    tmp12 = tmp2 + tmp4;
    tmp16 = tmp2 - tmp4;

    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5];
    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3];
    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2];
    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1];
    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0];
    tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13,
		       FIX(0.653061224)),                 /* 32/49 */
	      CONST_BITS+1);
    tmp13 += tmp13;
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */
	      MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */
	      MULTIPLY(tmp12 - tmp13, FIX(0.575835255)),  /* c8 */
	      CONST_BITS+1);

    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570));    /* c6 */

    dataptr[DCTSIZE*2] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691))   /* c2-c6 */
	      + MULTIPLY(tmp16, FIX(0.400721155)),        /* c10 */
	      CONST_BITS+1);
    dataptr[DCTSIZE*6] = (DCTELEM)
      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725))   /* c6+c10 */
	      - MULTIPLY(tmp16, FIX(0.900412262)),        /* c2 */
	      CONST_BITS+1);

    /* Odd part */

    tmp10 = tmp1 + tmp2;
    tmp11 = tmp5 - tmp4;
    dataptr[DCTSIZE*7] = (DCTELEM)
      DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6,
		       FIX(0.653061224)),                 /* 32/49 */
	      CONST_BITS+1);
    tmp3  = MULTIPLY(tmp3 , FIX(0.653061224));            /* 32/49 */
    tmp10 = MULTIPLY(tmp10, - FIX(0.103406812));          /* -c13 */
    tmp11 = MULTIPLY(tmp11, FIX(0.917760839));            /* c1 */
    tmp10 += tmp11 - tmp3;
    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) +     /* c5 */
	    MULTIPLY(tmp4 + tmp6, FIX(0.491367823));      /* c9 */
    dataptr[DCTSIZE*5] = (DCTELEM)
      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */
	      + MULTIPLY(tmp4, FIX(0.731428202)),         /* c1+c11-c9 */
	      CONST_BITS+1);
    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) +     /* c3 */
	    MULTIPLY(tmp5 - tmp6, FIX(0.305035186));      /* c11 */
    dataptr[DCTSIZE*3] = (DCTELEM)
      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */
	      - MULTIPLY(tmp5, FIX(2.004803435)),         /* c1+c5+c11 */
	      CONST_BITS+1);
    dataptr[DCTSIZE*1] = (DCTELEM)
      DESCALE(tmp11 + tmp12 + tmp3
	      - MULTIPLY(tmp0, FIX(0.735987049))          /* c3+c5-c1 */
	      - MULTIPLY(tmp6, FIX(0.082925825)),         /* c9-c11-c13 */
	      CONST_BITS+1);

    dataptr++;			/* advance pointer to next column */
    wsptr++;			/* advance pointer to next column */
  }
}


/*
 * Perform the forward DCT on a 15x15 sample block.
 */

GLOBAL(void)
jpeg_fdct_15x15 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
  INT32 z1, z2, z3;
  DCTELEM workspace[8*7];
  DCTELEM *dataptr;
  DCTELEM *wsptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT.
   * cK represents sqrt(2) * cos(K*pi/30).
   */

  dataptr = data;
  ctr = 0;
  for (;;) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[14]);
    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[13]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[12]);
    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[11]);
    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[10]);
    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[9]);
    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[8]);
    tmp7 = GETJSAMPLE(elemptr[7]);

    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[14]);
    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[13]);
    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[12]);
    tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[11]);
    tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[10]);

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

   */

  dataptr = data;
  wsptr = workspace;
  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*6];
    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*5];
    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*4];
    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*3];
    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*2];
    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*1];
    tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*0];
    tmp7 = dataptr[DCTSIZE*7];

    tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*6];
    tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*5];
    tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*4];
    tmp13 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*3];
    tmp14 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*2];
    tmp15 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*1];
    tmp16 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*0];

    z1 = tmp0 + tmp4 + tmp5;
    z2 = tmp1 + tmp3 + tmp6;
    z3 = tmp2 + tmp7;
    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(z1 + z2 + z3, FIX(1.137777778)), /* 256/225 */
	      CONST_BITS+2);
    z3 += z3;
    dataptr[DCTSIZE*6] = (DCTELEM)
      DESCALE(MULTIPLY(z1 - z3, FIX(1.301757503)) - /* c6 */
	      MULTIPLY(z2 - z3, FIX(0.497227121)),  /* c12 */
	      CONST_BITS+2);
    tmp2 += ((tmp1 + tmp4) >> 1) - tmp7 - tmp7;
    z1 = MULTIPLY(tmp3 - tmp2, FIX(1.742091575)) -  /* c2+c14 */
         MULTIPLY(tmp6 - tmp2, FIX(2.546621957));   /* c4+c8 */
    z2 = MULTIPLY(tmp5 - tmp2, FIX(0.908479156)) -  /* c8-c14 */
	 MULTIPLY(tmp0 - tmp2, FIX(0.103948774));   /* c2-c4 */
    z3 = MULTIPLY(tmp0 - tmp3, FIX(1.573898926)) +  /* c2 */
	 MULTIPLY(tmp6 - tmp5, FIX(1.076671805)) +  /* c8 */
	 MULTIPLY(tmp1 - tmp4, FIX(0.899492312));   /* (c6+c12)/2 */

    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z3, CONST_BITS+2);
    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(z2 + z3, CONST_BITS+2);

    /* Odd part */

    tmp2 = MULTIPLY(tmp10 - tmp12 - tmp13 + tmp15 + tmp16,
		    FIX(1.393487498));                         /* c5 */
    tmp1 = MULTIPLY(tmp10 - tmp14 - tmp15, FIX(1.530307725)) + /* c3 */
	   MULTIPLY(tmp11 - tmp13 - tmp16, FIX(0.945782187));  /* c9 */
    tmp12 = MULTIPLY(tmp12, FIX(1.393487498));                 /* c5 */
    tmp4 = MULTIPLY(tmp10 - tmp16, FIX(1.600246161)) +         /* c1 */
	   MULTIPLY(tmp11 + tmp14, FIX(1.530307725)) +         /* c3 */
	   MULTIPLY(tmp13 + tmp15, FIX(0.654463974));          /* c11 */
    tmp0 = MULTIPLY(tmp13, FIX(0.541301207)) -                 /* c7-c11 */
	   MULTIPLY(tmp14, FIX(0.584525538)) +                 /* c3-c9 */
	   MULTIPLY(tmp16, FIX(1.934788705)) + tmp4 + tmp12;   /* c1+c13 */
    tmp3 = MULTIPLY(tmp10, - FIX(0.404480980)) -               /* -(c1-c7) */
	   MULTIPLY(tmp11, FIX(2.476089912)) -                 /* c3+c9 */
	   MULTIPLY(tmp15, FIX(0.989006518)) + tmp4 - tmp12;   /* c11+c13 */

    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+2);
    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+2);
    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+2);
    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+2);

    dataptr++;			/* advance pointer to next column */
    wsptr++;			/* advance pointer to next column */
  }
}


/*
 * Perform the forward DCT on a 16x16 sample block.
 */

GLOBAL(void)
jpeg_fdct_16x16 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17;
  DCTELEM workspace[DCTSIZE2];
  DCTELEM *dataptr;
  DCTELEM *wsptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * cK represents sqrt(2) * cos(K*pi/32).
   */

  dataptr = data;
  ctr = 0;
  for (;;) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[15]);
    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[14]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[13]);
    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[12]);
    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[11]);
    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[10]);
    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[9]);
    tmp7 = GETJSAMPLE(elemptr[7]) + GETJSAMPLE(elemptr[8]);

    tmp10 = tmp0 + tmp7;
    tmp14 = tmp0 - tmp7;
    tmp11 = tmp1 + tmp6;
    tmp15 = tmp1 - tmp6;
    tmp12 = tmp2 + tmp5;
    tmp16 = tmp2 - tmp5;
    tmp13 = tmp3 + tmp4;
    tmp17 = tmp3 - tmp4;

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

/*
 * Perform the forward DCT on a 12x6 sample block.
 *
 * 12-point FDCT in pass 1 (rows), 6-point in pass 2 (columns).
 */

GLOBAL(void)
jpeg_fdct_12x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  DCTELEM *dataptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Zero 2 bottom rows of output coefficient block. */
  MEMZERO(&data[DCTSIZE*6], SIZEOF(DCTELEM) * DCTSIZE * 2);

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
   */

  dataptr = data;
  for (ctr = 0; ctr < 6; ctr++) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]);
    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]);
    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]);
    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]);
    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]);

    tmp10 = tmp0 + tmp5;
    tmp13 = tmp0 - tmp5;
    tmp11 = tmp1 + tmp4;
    tmp14 = tmp1 - tmp4;
    tmp12 = tmp2 + tmp3;
    tmp15 = tmp2 - tmp3;

    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]);
    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]);
    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]);
    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]);
    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]);
    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]);

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE) << PASS1_BITS);
    dataptr[6] = (DCTELEM) ((tmp13 - tmp14 - tmp15) << PASS1_BITS);
    dataptr[4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */
	      CONST_BITS-PASS1_BITS);
    dataptr[2] = (DCTELEM)
      DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */
	      CONST_BITS-PASS1_BITS);

    /* Odd part */

    tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100);    /* c9 */
    tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865);   /* c3-c9 */
    tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065);   /* c3+c9 */
    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054));   /* c5 */
    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669));   /* c7 */
    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */
	    + MULTIPLY(tmp5, FIX(0.184591911));        /* c11 */
    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */
    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */
	    + MULTIPLY(tmp5, FIX(0.860918669));        /* c7 */
    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */
	    - MULTIPLY(tmp5, FIX(1.121971054));        /* c5 */
    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */
	    - MULTIPLY(tmp2 + tmp5, FIX_0_541196100);  /* c9 */

    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);

    dataptr += DCTSIZE;		/* advance pointer to next row */
  }

  /* Pass 2: process columns.
   * We remove the PASS1_BITS scaling, but leave the results scaled up
   * by an overall factor of 8.
   * We must also scale the output by (8/12)*(8/6) = 8/9, which we
   * partially fold into the constant multipliers and final shifting:
   * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12) * 16/9.
   */

  dataptr = data;
  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5];
    tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];

    tmp10 = tmp0 + tmp2;
    tmp12 = tmp0 - tmp2;

    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5];
    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)),         /* 16/9 */
	      CONST_BITS+PASS1_BITS+1);
    dataptr[DCTSIZE*2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp12, FIX(2.177324216)),                 /* c2 */
	      CONST_BITS+PASS1_BITS+1);
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */
	      CONST_BITS+PASS1_BITS+1);

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

  }
}


/*
 * Perform the forward DCT on a 6x3 sample block.
 *
 * 6-point FDCT in pass 1 (rows), 3-point in pass 2 (columns).
 */

GLOBAL(void)
jpeg_fdct_6x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2;
  INT32 tmp10, tmp11, tmp12;
  DCTELEM *dataptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pre-zero output coefficient block. */
  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * We scale the results further by 2 as part of output adaption
   * scaling for different DCT size.
   * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  dataptr = data;
  for (ctr = 0; ctr < 3; ctr++) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);

    tmp10 = tmp0 + tmp2;
    tmp12 = tmp0 - tmp2;

    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << (PASS1_BITS+1));
    dataptr[2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
	      CONST_BITS-PASS1_BITS-1);
    dataptr[4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
	      CONST_BITS-PASS1_BITS-1);

    /* Odd part */

    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
		    CONST_BITS-PASS1_BITS-1);

    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << (PASS1_BITS+1)));
    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << (PASS1_BITS+1));
    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << (PASS1_BITS+1)));

    dataptr += DCTSIZE;		/* advance pointer to next row */
  }

  /* Pass 2: process columns.
   * We remove the PASS1_BITS scaling, but leave the results scaled up
   * by an overall factor of 8.
   * We must also scale the output by (8/6)*(8/3) = 32/9, which we partially
   * fold into the constant multipliers (other part was done in pass 1):
   * 3-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/6) * 16/9.
   */

  dataptr = data;
  for (ctr = 0; ctr < 6; ctr++) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*2];
    tmp1 = dataptr[DCTSIZE*1];

    tmp2 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*2];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),        /* 16/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(1.257078722)), /* c2 */
	      CONST_BITS+PASS1_BITS);

    /* Odd part */

    dataptr[DCTSIZE*1] = (DCTELEM)
      DESCALE(MULTIPLY(tmp2, FIX(2.177324216)),               /* c1 */
	      CONST_BITS+PASS1_BITS);

    dataptr++;			/* advance pointer to next column */
  }
}


/*
 * Perform the forward DCT on a 4x2 sample block.
 *
 * 4-point FDCT in pass 1 (rows), 2-point in pass 2 (columns).
 */

GLOBAL(void)
jpeg_fdct_4x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1;
  INT32 tmp10, tmp11;
  DCTELEM *dataptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

src/Source/LibJPEG/jfdctint.c  view on Meta::CPAN

    tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2];
    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1];
    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0];
    tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];

    tmp10 = tmp0 + tmp6;
    tmp14 = tmp0 - tmp6;
    tmp11 = tmp1 + tmp5;
    tmp15 = tmp1 - tmp5;
    tmp12 = tmp2 + tmp4;
    tmp16 = tmp2 - tmp4;

    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5];
    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4];
    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3];
    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2];
    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1];
    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0];
    tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13,
		       FIX(0.653061224)),                 /* 32/49 */
	      CONST_BITS+PASS1_BITS);
    tmp13 += tmp13;
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */
	      MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */
	      MULTIPLY(tmp12 - tmp13, FIX(0.575835255)),  /* c8 */
	      CONST_BITS+PASS1_BITS);

    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570));    /* c6 */

    dataptr[DCTSIZE*2] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691))   /* c2-c6 */
	      + MULTIPLY(tmp16, FIX(0.400721155)),        /* c10 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*6] = (DCTELEM)
      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725))   /* c6+c10 */
	      - MULTIPLY(tmp16, FIX(0.900412262)),        /* c2 */
	      CONST_BITS+PASS1_BITS);

    /* Odd part */

    tmp10 = tmp1 + tmp2;
    tmp11 = tmp5 - tmp4;
    dataptr[DCTSIZE*7] = (DCTELEM)
      DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6,
		       FIX(0.653061224)),                 /* 32/49 */
	      CONST_BITS+PASS1_BITS);
    tmp3  = MULTIPLY(tmp3 , FIX(0.653061224));            /* 32/49 */
    tmp10 = MULTIPLY(tmp10, - FIX(0.103406812));          /* -c13 */
    tmp11 = MULTIPLY(tmp11, FIX(0.917760839));            /* c1 */
    tmp10 += tmp11 - tmp3;
    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) +     /* c5 */
	    MULTIPLY(tmp4 + tmp6, FIX(0.491367823));      /* c9 */
    dataptr[DCTSIZE*5] = (DCTELEM)
      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */
	      + MULTIPLY(tmp4, FIX(0.731428202)),         /* c1+c11-c9 */
	      CONST_BITS+PASS1_BITS);
    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) +     /* c3 */
	    MULTIPLY(tmp5 - tmp6, FIX(0.305035186));      /* c11 */
    dataptr[DCTSIZE*3] = (DCTELEM)
      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */
	      - MULTIPLY(tmp5, FIX(2.004803435)),         /* c1+c5+c11 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*1] = (DCTELEM)
      DESCALE(tmp11 + tmp12 + tmp3
	      - MULTIPLY(tmp0, FIX(0.735987049))          /* c3+c5-c1 */
	      - MULTIPLY(tmp6, FIX(0.082925825)),         /* c9-c11-c13 */
	      CONST_BITS+PASS1_BITS);

    dataptr++;			/* advance pointer to next column */
    wsptr++;			/* advance pointer to next column */
  }
}


/*
 * Perform the forward DCT on a 6x12 sample block.
 *
 * 6-point FDCT in pass 1 (rows), 12-point in pass 2 (columns).
 */

GLOBAL(void)
jpeg_fdct_6x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  DCTELEM workspace[8*4];
  DCTELEM *dataptr;
  DCTELEM *wsptr;
  JSAMPROW elemptr;
  int ctr;
  SHIFT_TEMPS

  /* Pre-zero output coefficient block. */
  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  dataptr = data;
  ctr = 0;
  for (;;) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);

    tmp10 = tmp0 + tmp2;
    tmp12 = tmp0 - tmp2;

    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS);
    dataptr[2] = (DCTELEM)
      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
	      CONST_BITS-PASS1_BITS);
    dataptr[4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
	      CONST_BITS-PASS1_BITS);

    /* Odd part */

    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
		    CONST_BITS-PASS1_BITS);

    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS));
    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS);
    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS));

    ctr++;

    if (ctr != DCTSIZE) {
      if (ctr == 12)
	break;			/* Done. */
      dataptr += DCTSIZE;	/* advance pointer to next row */
    } else
      dataptr = workspace;	/* switch pointer to extended workspace */
  }

  /* Pass 2: process columns.
   * We remove the PASS1_BITS scaling, but leave the results scaled up
   * by an overall factor of 8.
   * We must also scale the output by (8/6)*(8/12) = 8/9, which we
   * fold into the constant multipliers:
   * 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24) * 8/9.
   */

  dataptr = data;
  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++) {
    /* Even part */

    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3];
    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2];
    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1];
    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0];
    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7];
    tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6];

    tmp10 = tmp0 + tmp5;
    tmp13 = tmp0 - tmp5;
    tmp11 = tmp1 + tmp4;
    tmp14 = tmp1 - tmp4;
    tmp12 = tmp2 + tmp3;
    tmp15 = tmp2 - tmp3;

    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*3];
    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*2];
    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*1];
    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*0];
    tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*7];
    tmp5 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*6];

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(0.888888889)), /* 8/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*6] = (DCTELEM)
      DESCALE(MULTIPLY(tmp13 - tmp14 - tmp15, FIX(0.888888889)), /* 8/9 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.088662108)),         /* c4 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*2] = (DCTELEM)



( run in 1.031 second using v1.01-cache-2.11-cpan-39bf76dae61 )