[INFO] Compilation of Math for your bot

[xalcon]

but even a puny one as me knows that the more tasks you give to the processor the slower it runs

Oh, realy? Didn't knew that :3 my CPU doesn't run slower, the more it needs to do :P It's always running at 2.6 GHz...

ok, back to topic:
well, there are alot of ways to get the nearest object compared to another object. I've done some benchmarks some weeks ago, and you SHOULD implement a getDistance()-Method, which returns the non-sqrt value of the distance. It's around 100% faster, then calculating the "real" distance.. well, at least when using the loop version and a very high amount of objects :P

Nearest object - Objectcount int posList: 1000000
Nearest via LinqFind - with square root: 731ms
Nearest via LinqFind - without square root: 687ms
Nearest via LoopFind - with square root: 75ms
Nearest via LoopFind - without square root: 37ms

Using Microsoft Linq may not be the fastest, but the easiest to change and understand:

Vector3D nearest =
                (from v in posList
                 orderby v.getDistance(me)
                 select v).First();

[suicidity]

Funny, because my results are drastically different. Looping over 1000000 objects like you..

Length() took 67.7275ms
Shortest length 0.01256103
LengthSq() took 7.6812ms
Shortest length 0.0001577795

That's 88.65% faster.

[streppel]

Just did try it myself:
~0.022ms per object(including creating random numbers for the coords!) with sqrt
~0.0217ms per object without sqrt...(and with rnd coords again)

isn't sqrt a cpu function? so i shouldn't be too much of a problem for the cpu to calculate it

[Bananenbrot]

But even the CPU has to approximate them. I really doubt there is a lookup for that.

Ahh.. google says c++ practical computational complexity of <cmath> SQRT() - Stack Overflow the answer is enlightening.

Why the heck are you actually benchmarking object distance computations? :P

[streppel]

cause they were arguing about wether it was faster to do object sorting by distance WITH or w/o squareroot...so this now means that either way it's so damned fast that you won't ever need to care about it in a real world application

[Bananenbrot]

Yeah I know... and I wasn't blaming anyone (especially you) for it^^
But isn't it enough to know THAT it's faster, rather than measure it on a hypothetical 1.000.000 objects (whereever you get them from)?

[streppel]

ofc it is...now let's cut this discussion and better stick to the topic before one of them mods forces us to do so

[Musah]

*snip*

Not using C# So Am using a loop not Linq ^_^
Thanks for benchmarking those results for me. I'll get rid of the sqrt ASAP.

---------- Post added at 07:50 PM ---------- Previous post was at 07:47 PM ----------

Funny, because my results are drastically different. Looping over 1000000 objects like you..

Length() took 67.7275ms
Shortest length 0.01256103
LengthSq() took 7.6812ms
Shortest length 0.0001577795

That's 88.65% faster.

This leaves me confused.

[xalcon]

1000000 objects were only a test. My intention was to test how fast i could get the nearest object out of aprox 5000 objects... it was done in around 1 ms without sqrt and 3ms with sqrt... but this was on my notebook :P

Not using C# So Am using a loop not Linq ^_^
Thanks for benchmarking those results for me. I'll get rid of the sqrt ASAP.

---------- Post added at 07:50 PM ---------- Previous post was at 07:47 PM ----------


This leaves me confused.

me too o.O" why is your sqrt() distance function faster then the function without it? what the heck are you doin? are you sure that you've benched the right functions? ^^

[Bananenbrot]

Length() is obviously sqrt(LengthSq()). So this shouldn't confuse you that much.

Interpreting suicidity's and your results and considering my SO link, your differences could stem from the fact that suicidity used a generally faster (read more optimized) algorithm (maybe some stuff like stackalloc etc.), but had an older CPU (sqrt uses more cycles in relation to other copmutations).

[Musah]

Guess it's subjective to the computer. Only use what's fastest for you. Will benchmark results.

011-09-30 09:58:26.105 DevSpicket[73693:707] Let's populate 100000 false objects.
2011-09-30 09:58:26.302 DevSpicket[73693:707] Population complete after 0.197572 seconds.
2011-09-30 09:58:26.398 DevSpicket[73693:707] Found the closest Object after 0.093588 seconds without sqrt.
2011-09-30 09:58:26.509 DevSpicket[73693:707] Found the closest Object after 0.110844 seconds using sqrt.

[streppel]

well,in a real world program i think i would go for doing it without the sqrt to sort the objects by distance and afterwards only calculate the real distance with sqrt for the 4 or 5 nearest objects so i already know my next targets and their distance,travel time,etc