/*************************************************************
Copyright (C) 1990, 1991, 1993 Andy C. Hung, all rights reserved.
PUBLIC DOMAIN LICENSE: Stanford University Portable Video Research
Group. If you use this software, you agree to the following: This
program package is purely experimental, and is licensed "as is".
Permission is granted to use, modify, and distribute this program
without charge for any purpose, provided this license/ disclaimer
notice appears in the copies.  No warranty or maintenance is given,
either expressed or implied.  In no event shall the author(s) be
liable to you or a third party for any special, incidental,
consequential, or other damages, arising out of the use or inability
to use the program for any purpose (or the loss of data), even if we
have been advised of such possibilities.  Any public reference or
advertisement of this source code should refer to it as the Portable
Video Research Group (PVRG) code, and not by any author(s) (or
Stanford University) name.
*************************************************************/
/*
************************************************************
chendct.c

A simple DCT algorithm that seems to have fairly nice arithmetic
properties.

W. H. Chen, C. H. Smith and S. C. Fralick "A fast computational
algorithm for the discrete cosine transform," IEEE Trans. Commun.,
vol. COM-25, pp. 1004-1009, Sept 1977.

************************************************************
*/

/*LABEL chendct.c */

/*PUBLIC*/

extern void ChenDct();
extern void ChenIDct();

/*PRIVATE*/

/* Standard Macros */

#define NO_MULTIPLY

#ifdef NO_MULTIPLY
#define LS(r,s) ((r) << (s))
#define RS(r,s) ((r) >> (s))       /* Caution with rounding... */
#else
#define LS(r,s) ((r) * (1 << (s)))
#define RS(r,s) ((r) / (1 << (s))) /* Correct rounding */
#endif

#define MSCALE(expr)  RS((expr),9)

/* Cos constants */

#define c1d4 362L

#define c1d8 473L
#define c3d8 196L

#define c1d16 502L
#define c3d16 426L
#define c5d16 284L
#define c7d16 100L

/*
  VECTOR_DEFINITION makes the temporary variables vectors.
  Useful for machines with small register spaces.

  */

#ifdef VECTOR_DEFINITION
#define a0 a[0]
#define a1 a[1]
#define a2 a[2]
#define a3 a[3]
#define b0 b[0]
#define b1 b[1]
#define b2 b[2]
#define b3 b[3]
#define c0 c[0]
#define c1 c[1]
#define c2 c[2]
#define c3 c[3]
#endif

/*START*/
/*BFUNC

ChenDCT() implements the Chen forward dct. Note that there are two
input vectors that represent x=input, and y=output, and must be
defined (and storage allocated) before this routine is called.

EFUNC*/

void ChenDct(x,y)
     int *x;
     int *y;
{
  register int i;
  register int *aptr,*bptr;
#ifdef VECTOR_DEFINITION
  register int a[4];
  register int b[4];
  register int c[4];
#else
  register int a0,a1,a2,a3;
  register int b0,b1,b2,b3;
  register int c0,c1,c2,c3;
#endif

  /* Loop over columns */

  for(i=0;i<8;i++)
    {
      aptr = x+i;
      bptr = aptr+56;

      a0 = LS((*aptr+*bptr),2);
      c3 = LS((*aptr-*bptr),2);
      aptr += 8;
      bptr -= 8;
      a1 = LS((*aptr+*bptr),2);
      c2 = LS((*aptr-*bptr),2);
      aptr += 8;
      bptr -= 8;
      a2 = LS((*aptr+*bptr),2);
      c1 = LS((*aptr-*bptr),2);
      aptr += 8;
      bptr -= 8;
      a3 = LS((*aptr+*bptr),2);
      c0 = LS((*aptr-*bptr),2);

      b0 = a0+a3;
      b1 = a1+a2;
      b2 = a1-a2;
      b3 = a0-a3;

      aptr = y+i;

      *aptr = MSCALE(c1d4*(b0+b1));
      aptr[32] = MSCALE(c1d4*(b0-b1));

      aptr[16] = MSCALE((c3d8*b2)+(c1d8*b3));
      aptr[48] = MSCALE((c3d8*b3)-(c1d8*b2));

      b0 = MSCALE(c1d4*(c2-c1));
      b1 = MSCALE(c1d4*(c2+c1));

      a0 = c0+b0;
      a1 = c0-b0;
      a2 = c3-b1;
      a3 = c3+b1;

      aptr[8] = MSCALE((c7d16*a0)+(c1d16*a3));
      aptr[24] = MSCALE((c3d16*a2)-(c5d16*a1));
      aptr[40] = MSCALE((c3d16*a1)+(c5d16*a2));
      aptr[56] = MSCALE((c7d16*a3)-(c1d16*a0));
    }

  for(i=0;i<8;i++)
    {       /* Loop over rows */
      aptr = y+LS(i,3);
      bptr = aptr+7;

      c3 = RS((*(aptr)-*(bptr)),1);
      a0 = RS((*(aptr++)+*(bptr--)),1);
      c2 = RS((*(aptr)-*(bptr)),1);
      a1 = RS((*(aptr++)+*(bptr--)),1);
      c1 = RS((*(aptr)-*(bptr)),1);
      a2 = RS((*(aptr++)+*(bptr--)),1);
      c0 = RS((*(aptr)-*(bptr)),1);
      a3 = RS((*(aptr)+*(bptr)),1);

      b0 = a0+a3;
      b1 = a1+a2;
      b2 = a1-a2;
      b3 = a0-a3;

      aptr = y+LS(i,3);

      *aptr = MSCALE(c1d4*(b0+b1));
      aptr[4] = MSCALE(c1d4*(b0-b1));
      aptr[2] = MSCALE((c3d8*b2)+(c1d8*b3));
      aptr[6] = MSCALE((c3d8*b3)-(c1d8*b2));

      b0 = MSCALE(c1d4*(c2-c1));
      b1 = MSCALE(c1d4*(c2+c1));

      a0 = c0+b0;
      a1 = c0-b0;
      a2 = c3-b1;
      a3 = c3+b1;

      aptr[1] = MSCALE((c7d16*a0)+(c1d16*a3));
      aptr[3] = MSCALE((c3d16*a2)-(c5d16*a1));
      aptr[5] = MSCALE((c3d16*a1)+(c5d16*a2));
      aptr[7] = MSCALE((c7d16*a3)-(c1d16*a0));
    }

  /* We have an additional factor of 8 in the Chen algorithm. */

  for(i=0,aptr=y;i<64;i++,aptr++)
    *aptr = (((*aptr<0) ? (*aptr-4) : (*aptr+4))/8);
}


/*BFUNC

ChenIDCT() implements the Chen inverse dct. Note that there are two
input vectors that represent x=input, and y=output, and must be
defined (and storage allocated) before this routine is called.

EFUNC*/

void ChenIDct(x,y)
     int *x;
     int *y;
{
  register int i;
  register int *aptr;
#ifdef VECTOR_DEFINITION
  register int a[4];
  register int b[4];
  register int c[4];
#else
  register int a0,a1,a2,a3;
  register int b0,b1,b2,b3;
  register int c0,c1,c2,c3;
#endif

  /* Loop over columns */

  for(i=0;i<8;i++)
    {
      aptr = x+i;
      b0 = LS(*aptr,2);
      aptr += 8;
      a0 = LS(*aptr,2);
      aptr += 8;
      b2 = LS(*aptr,2);
      aptr += 8;
      a1 = LS(*aptr,2);
      aptr += 8;
      b1 = LS(*aptr,2);
      aptr += 8;
      a2 = LS(*aptr,2);
      aptr += 8;
      b3 = LS(*aptr,2);
      aptr += 8;
      a3 = LS(*aptr,2);

      /* Split into even mode  b0 = x0  b1 = x4  b2 = x2  b3 = x6.
   And the odd terms a0 = x1 a1 = x3 a2 = x5 a3 = x7.
   */

      c0 = MSCALE((c7d16*a0)-(c1d16*a3));
      c1 = MSCALE((c3d16*a2)-(c5d16*a1));
      c2 = MSCALE((c3d16*a1)+(c5d16*a2));
      c3 = MSCALE((c1d16*a0)+(c7d16*a3));

      /* First Butterfly on even terms.*/

      a0 = MSCALE(c1d4*(b0+b1));
      a1 = MSCALE(c1d4*(b0-b1));

      a2 = MSCALE((c3d8*b2)-(c1d8*b3));
      a3 = MSCALE((c1d8*b2)+(c3d8*b3));

      b0 = a0+a3;
      b1 = a1+a2;
      b2 = a1-a2;
      b3 = a0-a3;

      /* Second Butterfly */

      a0 = c0+c1;
      a1 = c0-c1;
      a2 = c3-c2;
      a3 = c3+c2;

      c0 = a0;
      c1 = MSCALE(c1d4*(a2-a1));
      c2 = MSCALE(c1d4*(a2+a1));
      c3 = a3;

      aptr = y+i;
      *aptr = b0+c3;
      aptr += 8;
      *aptr = b1+c2;
      aptr += 8;
      *aptr = b2+c1;
      aptr += 8;
      *aptr = b3+c0;
      aptr += 8;
      *aptr = b3-c0;
      aptr += 8;
      *aptr = b2-c1;
      aptr += 8;
      *aptr = b1-c2;
      aptr += 8;
      *aptr = b0-c3;
    }

  /* Loop over rows */

  for(i=0;i<8;i++)
    {
      aptr = y+LS(i,3);
      b0 = *(aptr++);
      a0 = *(aptr++);
      b2 = *(aptr++);
      a1 = *(aptr++);
      b1 = *(aptr++);
      a2 = *(aptr++);
      b3 = *(aptr++);
      a3 = *(aptr);

      /*
  Split into even mode  b0 = x0  b1 = x4  b2 = x2  b3 = x6.
  And the odd terms a0 = x1 a1 = x3 a2 = x5 a3 = x7.
  */

      c0 = MSCALE((c7d16*a0)-(c1d16*a3));
      c1 = MSCALE((c3d16*a2)-(c5d16*a1));
      c2 = MSCALE((c3d16*a1)+(c5d16*a2));
      c3 = MSCALE((c1d16*a0)+(c7d16*a3));

      /* First Butterfly on even terms.*/

      a0 = MSCALE(c1d4*(b0+b1));
      a1 = MSCALE(c1d4*(b0-b1));

      a2 = MSCALE((c3d8*b2)-(c1d8*b3));
      a3 = MSCALE((c1d8*b2)+(c3d8*b3));

      /* Calculate last set of b's */

      b0 = a0+a3;
      b1 = a1+a2;
      b2 = a1-a2;
      b3 = a0-a3;

      /* Second Butterfly */

      a0 = c0+c1;
      a1 = c0-c1;
      a2 = c3-c2;
      a3 = c3+c2;

      c0 = a0;
      c1 = MSCALE(c1d4*(a2-a1));
      c2 = MSCALE(c1d4*(a2+a1));
      c3 = a3;

      aptr = y+LS(i,3);
      *(aptr++) = b0+c3;
      *(aptr++) = b1+c2;
      *(aptr++) = b2+c1;
      *(aptr++) = b3+c0;
      *(aptr++) = b3-c0;
      *(aptr++) = b2-c1;
      *(aptr++) = b1-c2;
      *(aptr) = b0-c3;
    }

  /*
    Retrieve correct accuracy. We have additional factor
    of 16 that must be removed.
   */

  for(i=0,aptr=y;i<64;i++,aptr++)
    *aptr = (((*aptr<0) ? (*aptr-8) : (*aptr+8)) /16);
}

/*END*/