Function xsum_large_add_dot

xs/sources/xsum.c:1806–1963

/* ADD DOT PRODUCT OF VECTORS OF FLOATING-POINT NUMBERS TO LARGE ACCUMULATOR. */

void xsum_large_add_dot (xsum_large_accumulator *restrict lacc,
                         const xsum_flt *vec1,
                         const xsum_flt *vec2,
                         xsum_length n)
{
  if (xsum_debug) c_printf("\nLARGE ADD_DOT OF %ld VALUES\n",(long)n);

# if OPT_LARGE_DOT
  {
    xsum_lcount count;
    xsum_expint ix;
    xsum_uint uintv;
    double fltv;

    while (n > 3)
    {
      fltv = *vec1 * *vec2;
      COPY64 (uintv, fltv);
      vec1 += 1; vec2 += 1;

      ix = uintv >> XSUM_MANTISSA_BITS;

      count = lacc->count[ix] - 1;

      if (count < 0)
      { xsum_large_add_value_inf_nan (lacc, ix, uintv);
      }
      else
      { lacc->count[ix] = count;
        lacc->chunk[ix] += uintv;
      }

      fltv = *vec1 * *vec2;
      COPY64 (uintv, fltv);
      vec1 += 1; vec2 += 1;

      ix = uintv >> XSUM_MANTISSA_BITS;

      count = lacc->count[ix] - 1;

      if (count < 0)
      { xsum_large_add_value_inf_nan (lacc, ix, uintv);
      }
      else
      { lacc->count[ix] = count;
        lacc->chunk[ix] += uintv;
      }

      fltv = *vec1 * *vec2;
      COPY64 (uintv, fltv);
      vec1 += 1; vec2 += 1;

      ix = uintv >> XSUM_MANTISSA_BITS;

      count = lacc->count[ix] - 1;

      if (count < 0)
      { xsum_large_add_value_inf_nan (lacc, ix, uintv);