Re: [gnugo-devel] use fixed point arithmetics in accumulat_influence

[Arend Bayer]

Gunnar wrote:

> Maybe you misunderstood me somewhat.
Aha indeed.

> I was thinking of something that would roughly look like
>
> static void
> compute_influence(struct influence_data *q, int color, int pos,
>                 char no_influence[BOARDMAX], char saved_stones[BOARDMAX])
> {
>   [...]
>
>   [Convert needed variables from float to fixed point.]
>
>   for (ii = BOARDMIN; ii < BOARDMAX; ii++)
>     if (ON_BOARD(ii)) {
>       if (q->white_strength[ii] > 0.0)
>       accumulate_influence(q, ii, WHITE);
>       if (q->black_strength[ii] > 0.0)
>       accumulate_influence(q, ii, BLACK);
>     }
>
>   [Convert fixed point values back to float.]
>
>   [...]
> }
>
> This should minimize the impact of the fixed point arithmetics. The
> question is whether it would be efficient enough.

Ok, to spell this out more precisely:

static void
accumulat_influence(struct influence_data *q, int strength, int
                    strength_pos, int color, const int permeabilities[BOARDMAX],
                    int influence[BOARDMAX])

  [Generated converted fixed point permeability tables]

  for (ii = BOARDMIN; ii < BOARDMAX; ii++)
    if (ON_BOARD(ii)) {
      if (q->white_strength[ii] > 0.0)
        accumulate_influence(q, q->white_strength[ii] * GG_ONE, ii, WHITE,
                             white_permeabilities, white_influence[...]);
      if (q->black_strength[ii] > 0.0)
        accumulate_influence(q, q->black_strength[ii] * GG_ONE, ii, BLACK,
                             black_permeabilities, black_influence[..]);
  [Convert back black_/white_influence]

Then fixed point would be limited to the function parameters and the
internal variables of accumulate_influence(). Sounds better than my
current patch, definitively.

Arend