[PATCH] Pathfind performance

[tuan kuranes]

Pathfind code optimisations without any single behaviour change(no hash change). 30% improvements, as tested directly on pathfind code.

(pathfind_test_branch_vanilla_update.path is for vanilla 0AD svn benchmark adding for fair comparison)

[historic_bruno]

Thanks for the work on this patch. Even though it's somewhat of an improvement (more like 5-15% reduction in short path computation time, depending on testing method, this matches comments from #1923), the short pathfinder algorithm is not scalable, so anything short of redesigning it doesn't have much value - we know this because it can easily consume 1-2s per turn, when a turn is only 200ms... This patch unfortunately makes the existing logic harder to understand, complicating a rewrite, and does yucky things like moving locals to global variables, which even though it may reduce allocations, is a problem that can be solved in a better way by redesign of the algorithm.

Reviewing the patch, it's hard to differentiate the changes that impact performance from those that are simply personal convention, in fact going line-by-line through it, I was able to revert most of the changes without changing performance at all (it was time consuming, but I compared them, building the game with VS2010 and profiling with Very Sleepy). I think some of the suggestions are things we should keep in mind when designing the new pathfinder algorithm and writing future code, but maintaining readable code is also important if it can be helped (given profiling data, it can be helped in this case).