kusano 7d535a
/*
kusano 7d535a
 * jfdctflt.c
kusano 7d535a
 *
kusano 7d535a
 * Copyright (C) 1994-1996, Thomas G. Lane.
kusano 7d535a
 * Modified 2003-2009 by Guido Vollbeding.
kusano 7d535a
 * This file is part of the Independent JPEG Group's software.
kusano 7d535a
 * For conditions of distribution and use, see the accompanying README file.
kusano 7d535a
 *
kusano 7d535a
 * This file contains a floating-point implementation of the
kusano 7d535a
 * forward DCT (Discrete Cosine Transform).
kusano 7d535a
 *
kusano 7d535a
 * This implementation should be more accurate than either of the integer
kusano 7d535a
 * DCT implementations.  However, it may not give the same results on all
kusano 7d535a
 * machines because of differences in roundoff behavior.  Speed will depend
kusano 7d535a
 * on the hardware's floating point capacity.
kusano 7d535a
 *
kusano 7d535a
 * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
kusano 7d535a
 * on each column.  Direct algorithms are also available, but they are
kusano 7d535a
 * much more complex and seem not to be any faster when reduced to code.
kusano 7d535a
 *
kusano 7d535a
 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
kusano 7d535a
 * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
kusano 7d535a
 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
kusano 7d535a
 * JPEG textbook (see REFERENCES section in file README).  The following code
kusano 7d535a
 * is based directly on figure 4-8 in P&M.
kusano 7d535a
 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
kusano 7d535a
 * possible to arrange the computation so that many of the multiplies are
kusano 7d535a
 * simple scalings of the final outputs.  These multiplies can then be
kusano 7d535a
 * folded into the multiplications or divisions by the JPEG quantization
kusano 7d535a
 * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
kusano 7d535a
 * to be done in the DCT itself.
kusano 7d535a
 * The primary disadvantage of this method is that with a fixed-point
kusano 7d535a
 * implementation, accuracy is lost due to imprecise representation of the
kusano 7d535a
 * scaled quantization values.  However, that problem does not arise if
kusano 7d535a
 * we use floating point arithmetic.
kusano 7d535a
 */
kusano 7d535a
kusano 7d535a
#define JPEG_INTERNALS
kusano 7d535a
#include "jinclude.h"
kusano 7d535a
#include "jpeglib.h"
kusano 7d535a
#include "jdct.h"		/* Private declarations for DCT subsystem */
kusano 7d535a
kusano 7d535a
#ifdef DCT_FLOAT_SUPPORTED
kusano 7d535a
kusano 7d535a
kusano 7d535a
/*
kusano 7d535a
 * This module is specialized to the case DCTSIZE = 8.
kusano 7d535a
 */
kusano 7d535a
kusano 7d535a
#if DCTSIZE != 8
kusano 7d535a
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
kusano 7d535a
#endif
kusano 7d535a
kusano 7d535a
kusano 7d535a
/*
kusano 7d535a
 * Perform the forward DCT on one block of samples.
kusano 7d535a
 */
kusano 7d535a
kusano 7d535a
GLOBAL(void)
kusano 7d535a
jpeg_fdct_float (FAST_FLOAT * data, JSAMPARRAY sample_data, JDIMENSION start_col)
kusano 7d535a
{
kusano 7d535a
  FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
kusano 7d535a
  FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
kusano 7d535a
  FAST_FLOAT z1, z2, z3, z4, z5, z11, z13;
kusano 7d535a
  FAST_FLOAT *dataptr;
kusano 7d535a
  JSAMPROW elemptr;
kusano 7d535a
  int ctr;
kusano 7d535a
kusano 7d535a
  /* Pass 1: process rows. */
kusano 7d535a
kusano 7d535a
  dataptr = data;
kusano 7d535a
  for (ctr = 0; ctr < DCTSIZE; ctr++) {
kusano 7d535a
    elemptr = sample_data[ctr] + start_col;
kusano 7d535a
kusano 7d535a
    /* Load data into workspace */
kusano 7d535a
    tmp0 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]));
kusano 7d535a
    tmp7 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]));
kusano 7d535a
    tmp1 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]));
kusano 7d535a
    tmp6 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]));
kusano 7d535a
    tmp2 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]));
kusano 7d535a
    tmp5 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]));
kusano 7d535a
    tmp3 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]));
kusano 7d535a
    tmp4 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]));
kusano 7d535a
kusano 7d535a
    /* Even part */
kusano 7d535a
kusano 7d535a
    tmp10 = tmp0 + tmp3;	/* phase 2 */
kusano 7d535a
    tmp13 = tmp0 - tmp3;
kusano 7d535a
    tmp11 = tmp1 + tmp2;
kusano 7d535a
    tmp12 = tmp1 - tmp2;
kusano 7d535a
kusano 7d535a
    /* Apply unsigned->signed conversion */
kusano 7d535a
    dataptr[0] = tmp10 + tmp11 - 8 * CENTERJSAMPLE; /* phase 3 */
kusano 7d535a
    dataptr[4] = tmp10 - tmp11;
kusano 7d535a
kusano 7d535a
    z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
kusano 7d535a
    dataptr[2] = tmp13 + z1;	/* phase 5 */
kusano 7d535a
    dataptr[6] = tmp13 - z1;
kusano 7d535a
kusano 7d535a
    /* Odd part */
kusano 7d535a
kusano 7d535a
    tmp10 = tmp4 + tmp5;	/* phase 2 */
kusano 7d535a
    tmp11 = tmp5 + tmp6;
kusano 7d535a
    tmp12 = tmp6 + tmp7;
kusano 7d535a
kusano 7d535a
    /* The rotator is modified from fig 4-8 to avoid extra negations. */
kusano 7d535a
    z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
kusano 7d535a
    z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
kusano 7d535a
    z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
kusano 7d535a
    z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
kusano 7d535a
kusano 7d535a
    z11 = tmp7 + z3;		/* phase 5 */
kusano 7d535a
    z13 = tmp7 - z3;
kusano 7d535a
kusano 7d535a
    dataptr[5] = z13 + z2;	/* phase 6 */
kusano 7d535a
    dataptr[3] = z13 - z2;
kusano 7d535a
    dataptr[1] = z11 + z4;
kusano 7d535a
    dataptr[7] = z11 - z4;
kusano 7d535a
kusano 7d535a
    dataptr += DCTSIZE;		/* advance pointer to next row */
kusano 7d535a
  }
kusano 7d535a
kusano 7d535a
  /* Pass 2: process columns. */
kusano 7d535a
kusano 7d535a
  dataptr = data;
kusano 7d535a
  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
kusano 7d535a
    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
kusano 7d535a
    tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
kusano 7d535a
    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
kusano 7d535a
    tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
kusano 7d535a
    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
kusano 7d535a
    tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
kusano 7d535a
    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
kusano 7d535a
    tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
kusano 7d535a
kusano 7d535a
    /* Even part */
kusano 7d535a
kusano 7d535a
    tmp10 = tmp0 + tmp3;	/* phase 2 */
kusano 7d535a
    tmp13 = tmp0 - tmp3;
kusano 7d535a
    tmp11 = tmp1 + tmp2;
kusano 7d535a
    tmp12 = tmp1 - tmp2;
kusano 7d535a
kusano 7d535a
    dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
kusano 7d535a
    dataptr[DCTSIZE*4] = tmp10 - tmp11;
kusano 7d535a
kusano 7d535a
    z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
kusano 7d535a
    dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
kusano 7d535a
    dataptr[DCTSIZE*6] = tmp13 - z1;
kusano 7d535a
kusano 7d535a
    /* Odd part */
kusano 7d535a
kusano 7d535a
    tmp10 = tmp4 + tmp5;	/* phase 2 */
kusano 7d535a
    tmp11 = tmp5 + tmp6;
kusano 7d535a
    tmp12 = tmp6 + tmp7;
kusano 7d535a
kusano 7d535a
    /* The rotator is modified from fig 4-8 to avoid extra negations. */
kusano 7d535a
    z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
kusano 7d535a
    z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
kusano 7d535a
    z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
kusano 7d535a
    z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
kusano 7d535a
kusano 7d535a
    z11 = tmp7 + z3;		/* phase 5 */
kusano 7d535a
    z13 = tmp7 - z3;
kusano 7d535a
kusano 7d535a
    dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
kusano 7d535a
    dataptr[DCTSIZE*3] = z13 - z2;
kusano 7d535a
    dataptr[DCTSIZE*1] = z11 + z4;
kusano 7d535a
    dataptr[DCTSIZE*7] = z11 - z4;
kusano 7d535a
kusano 7d535a
    dataptr++;			/* advance pointer to next column */
kusano 7d535a
  }
kusano 7d535a
}
kusano 7d535a
kusano 7d535a
#endif /* DCT_FLOAT_SUPPORTED */