/* TomsFastMath, a fast ISO C bignum library.
 * 
 * This project is meant to fill in where LibTomMath
 * falls short.  That is speed ;-)
 *
 * This project is public domain and free for all purposes.
 * 
 * Tom St Denis, tomstdenis@gmail.com
 */

#define TFM_DEFINES
#include "fp_sqr_comba.c"

/* generic comba squarer */
void fp_sqr_comba(fp_int *A, fp_int *B)
{
  int       pa, ix, iz;
  fp_digit  c0, c1, c2;
  fp_int    tmp, *dst;
#ifdef TFM_ISO
  fp_word   tt;
#endif    

  /* get size of output and trim */
  pa = A->used + A->used;
  if (pa >= FP_SIZE) {
     pa = FP_SIZE-1;
  }

  /* number of output digits to produce */
  COMBA_START;
  CLEAR_CARRY;

  if (A == B) {
     fp_zero(&tmp);
     dst = &tmp;
  } else {
     fp_zero(B);
     dst = B;
  }

  for (ix = 0; ix < pa; ix++) { 
      int      tx, ty, iy;
      fp_digit *tmpy, *tmpx;

      /* get offsets into the two bignums */
      ty = MIN(A->used-1, ix);
      tx = ix - ty;

      /* setup temp aliases */
      tmpx = A->dp + tx;
      tmpy = A->dp + ty;

      /* this is the number of times the loop will iterrate,
         while (tx++ < a->used && ty-- >= 0) { ... }
       */
      iy = MIN(A->used-tx, ty+1);

      /* now for squaring tx can never equal ty 
       * we halve the distance since they approach 
       * at a rate of 2x and we have to round because 
       * odd cases need to be executed
       */
      iy = MIN(iy, (ty-tx+1)>>1);

      /* forward carries */
      CARRY_FORWARD;

      /* execute loop */
      for (iz = 0; iz < iy; iz++) {
          fp_digit _tmpx = *tmpx++;
          fp_digit _tmpy = *tmpy--;
          SQRADD2(_tmpx, _tmpy);
      }

      /* even columns have the square term in them */
      if ((ix&1) == 0) {
          fp_digit _a_dp = A->dp[ix>>1];
          SQRADD(_a_dp, A->dp[ix>>1]);
      }

      /* store it */
      COMBA_STORE(dst->dp[ix]);
  }

  COMBA_FINI;

  /* setup dest */
  dst->used = pa;
  fp_clamp (dst);
  if (dst != B) {
     fp_copy(dst, B);
  }
}

/* $Source$ */
/* $Revision$ */
/* $Date$ */