jfdctint.c

M8手机图片查看器

    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[4]);
    tmp3 = GETJSAMPLE(elemptr[3]);

    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[6]);
    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[5]);
    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[4]);

    z1 = tmp0 + tmp2;
    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((z1 + tmp1 + tmp3 - 7 * CENTERJSAMPLE) << PASS1_BITS);
    tmp3 += tmp3;
    z1 -= tmp3;
    z1 -= tmp3;
    z1 = MULTIPLY(z1, FIX(0.353553391));                /* (c2+c6-c4)/2 */
    z2 = MULTIPLY(tmp0 - tmp2, FIX(0.920609002));       /* (c2+c4-c6)/2 */
    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.314692123));       /* c6 */
    dataptr[2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS-PASS1_BITS);
    z1 -= z2;
    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.881747734));       /* c4 */
    dataptr[4] = (DCTELEM)
      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.707106781)), /* c2+c6-c4 */
	      CONST_BITS-PASS1_BITS);
    dataptr[6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS-PASS1_BITS);

    /* Odd part */

    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(0.935414347));   /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.170262339));   /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.378756276)); /* -c1 */
    tmp1 += tmp2;
    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.613604268));   /* c5 */
    tmp0 += tmp3;
    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(1.870828693));   /* c3+c1-c5 */

    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS);
    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS);
    dataptr[5] = (DCTELEM) DESCALE(tmp2, 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 (8/7)**2 = 64/49, which we fold
   * into the constant multipliers:
   * cK now represents sqrt(2) * cos(K*pi/14) * 64/49.
   */

  dataptr = data;
  for (ctr = 0; ctr < 7; ctr++) {
    /* Even part */

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

    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*6];
    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*5];
    tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*4];

    z1 = tmp0 + tmp2;
    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(z1 + tmp1 + tmp3, FIX(1.306122449)), /* 64/49 */
	      CONST_BITS+PASS1_BITS);
    tmp3 += tmp3;
    z1 -= tmp3;
    z1 -= tmp3;
    z1 = MULTIPLY(z1, FIX(0.461784020));                /* (c2+c6-c4)/2 */
    z2 = MULTIPLY(tmp0 - tmp2, FIX(1.202428084));       /* (c2+c4-c6)/2 */
    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.411026446));       /* c6 */
    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS+PASS1_BITS);
    z1 -= z2;
    z2 = MULTIPLY(tmp0 - tmp1, FIX(1.151670509));       /* c4 */
    dataptr[DCTSIZE*4] = (DCTELEM)
      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.923568041)), /* c2+c6-c4 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+PASS1_BITS);

    /* Odd part */

    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.221765677));   /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.222383464));   /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.800824523)); /* -c1 */
    tmp1 += tmp2;
    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.801442310));   /* c5 */
    tmp0 += tmp3;
    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(2.443531355));   /* c3+c1-c5 */

    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS);
    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 */
  }
}


/*
 * Perform the forward DCT on a 5x5 sample block.
 */

GLOBAL(void)
jpeg_fdct_5x5 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1, tmp2;
  INT32 tmp10, tmp11;
  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. */
  /* cK represents sqrt(2) * cos(K*pi/10). */

  dataptr = data;
  for (ctr = 0; ctr < 5; ctr++) {
    elemptr = sample_data[ctr] + start_col;

    /* Even part */

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

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;

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

    /* Apply unsigned->signed conversion */
    dataptr[0] = (DCTELEM)
      ((tmp10 + tmp2 - 5 * CENTERJSAMPLE) << (PASS1_BITS+1));
    tmp11 = MULTIPLY(tmp11, FIX(0.790569415));          /* (c2+c4)/2 */
    tmp10 -= tmp2 << 2;
    tmp10 = MULTIPLY(tmp10, FIX(0.353553391));          /* (c2-c4)/2 */
    dataptr[2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS-PASS1_BITS-1);
    dataptr[4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS-PASS1_BITS-1);

    /* Odd part */

    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(0.831253876));    /* c3 */

    dataptr[1] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.513743148)), /* c1-c3 */
	      CONST_BITS-PASS1_BITS-1);
    dataptr[3] = (DCTELEM)
      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.176250899)), /* c1+c3 */
	      CONST_BITS-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 (8/5)**2 = 64/25, which we partially
   * fold into the constant multipliers (other part was done in pass 1):
   * cK now represents sqrt(2) * cos(K*pi/10) * 32/25.
   */

  dataptr = data;
  for (ctr = 0; ctr < 5; ctr++) {
    /* Even part */

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

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;

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

    dataptr[DCTSIZE*0] = (DCTELEM)
      DESCALE(MULTIPLY(tmp10 + tmp2, FIX(1.28)),        /* 32/25 */
	      CONST_BITS+PASS1_BITS);
    tmp11 = MULTIPLY(tmp11, FIX(1.011928851));          /* (c2+c4)/2 */
    tmp10 -= tmp2 << 2;
    tmp10 = MULTIPLY(tmp10, FIX(0.452548340));          /* (c2-c4)/2 */
    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS+PASS1_BITS);

    /* Odd part */

    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(1.064004961));    /* c3 */

    dataptr[DCTSIZE*1] = (DCTELEM)
      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.657591230)), /* c1-c3 */
	      CONST_BITS+PASS1_BITS);
    dataptr[DCTSIZE*3] = (DCTELEM)
      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.785601151)), /* c1+c3 */
	      CONST_BITS+PASS1_BITS);

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


/*
 * Perform the forward DCT on a 4x4 sample block.
 */

GLOBAL(void)
jpeg_fdct_4x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
{
  INT32 tmp0, tmp1;
  INT32 tmp10, tmp11;
  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 must also scale the output by (8/4)**2 = 2**2, which we add here. */
  /* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. */

  dataptr = data;
  for (ctr = 0; ctr < 4; ctr++) {
    elemptr = sample_data[ctr] + start_col;