[PATCH] Long range pathfinder performance improvements in AtGoal()

[kanetaka]

The original AtGoal() calculates the distance to goal every time, but the distance can be got from heuristic cost. This routine is in main loop on CCmpPathfinder::ComputePath(), it requires optimization. To check my improvements, I added some profiling code in both the original and my patch and I have both codes run with replay mode. The commands.txt is from 1 human vs. 1 AI player's match. The result is:

HRT: using name=TSC freq=1796000000.P000000
HRT: counter=TSC freq=1.796e+009 res=5.56793e-010 bits=64

Original
  tc_AtGoal: 431.391 Mc (788205x)
My patch
  tc_AtGoal: 67.5181 Mc (788205x)

The numbers of called are same, which implies this patch doesn't change pathfinding results. The performance improvements about 7 times faster.

[Stan]

That's awesome, keep up the good work :) Do you think a 7time faster calculation would save up a lot of lag ?

[Niek]

Replying to stanislas69:

That's awesome, keep up the good work :) Do you think a 7time faster calculation would save up a lot of lag ?

As kanetaka wanted to say (but got rejected): No, 300 Mc gain is only 1/6 seconds per game. It might differ per game, but I'm not sure. Although I do not completely understand how the engine works, I guess the other components are the bottlenecks.

[leper]

Uses floats in simulation code (breaks network sync). Also the pathfinder is in the process of being rewritten (as bad algorithms can't be fixed with micro-optimizations; and the current pathfinders have mismatches (which is the reason it is being rewritten)). It'd be nice if you'd help with the new wip pathfinder.