after getting fed up with the slow speed of freeciv in a large game I
did some profiling. One thing I noticed is that there are MACROS that
use one of their arguments twice. With the change to make the
player_index a function call they are not called with a function
call as argument instead of an interger. Using the argument twice
means the function call is executed twice.

In the server 3% of the total time where spend in map_is_known:

[56]     3.0    5.43   12.74 152585010         map_is_known [56]
                8.89    0.00 305170020/565345748     player_number [62]
                3.85    0.00 305170020/541877534     player_index [103]

Notice how player_number and player_index is called twice for every

---[ server/maphand.c ]------------------------------------------------------

  Return whether the player knows the tile.  Knowing a tile means
  seen it once (as opposed to seeing a tile which means you can see it
bool map_is_known(const struct tile *ptile, const struct player
  return BV_ISSET(ptile->tile_known, player_index(pplayer));

---[ utils/shared.h ]------------------------------------------------------

#define BV_ISSET(bv, bit) \
  (_BV_ASSERT(bv, bit), \
   ((bv).vec[_BV_BYTE_INDEX(bit)] & _BV_BITMASK(bit)) != 0)

---[ common/player.c ]-----------------------------------------------------

  Return the player index.

  Currently same as player_number(), paired with player_count()
  indicates use as an array index.
int player_index(const struct player *pplayer)
  return player_number(pplayer);

  Return the player index/number/id.
int player_number(const struct player *pplayer)
  return pplayer - game.players;


Is there any reason BV_ISSET (and similar ones) isn't written as a
static inline function? That one macro alone would save 1.5%.

Inlining the player_number() and player_index() functions (and other
similary trivial and often called functions) would save a lot of time
too. Any reason such frequent one-liners aren't declared static inline
in the header file?


Freeciv-dev mailing list

Reply via email to