1. Re: Math Shortcuts
- Posted by Hawke <mdeland at NWINFO.NET>
Jul 03, 1998
-
Last edited Jul 04, 1998
what if:
Xdis = floor(x2-x1)
Xdis = power(Xdis,2)
--now in theory, Xdis _should_ be 'integer', ie:
--Xdis should have no 'real' or fractional part
Ydis = floor(y2-y1)
Xdis = power(Ydis,2)
--Zdis = power(floor(z2-z1),2)
DIS = SQRT( Xdis + Ydis ) -- + Zdis
since you're starting with an 'integer' value
sqrt should be faster... but if its not actually
faster, then since it is an integer, you could in
theory have a lookup table for the squareroots
of all the integers in the range most likely
for DIS to fall in... then if you used a binary
search algorithm for that lookup table...after all
the lookup table is of course inherently presorted...
and binary searchs are -fast-...can you gain a
speed increase over just plain
dis = sqrt ( power(x2-x1,2) + power(y2-y1,2))????
no...i dont think so...i dont think any of the tricks
i've mentioned or seen mentioned will actually improve
speed noticeably (except of course R.McDougal optimization
for deltaX=deltaY and his algorithm for longhanding it)
unless youre in the middle of a loop
iterating at least several thousand times...finding
several thousand distances...then you may very well
be able to see a difference...sqrt is not that slow...
im not being pessimistic...well okay i am :/
im just saying that sometimes tweaking for speed
is not worth the implementation effort to gain
0.000000001 % in a program that only takes 1/4 sec
to run anyway :) i've caught myself doing tweaking
like that...bonk myself later for it...
and btw, nowadays, raising a number to a power as
opposed to multiplying it times itself (as in a loop)
won't hamper you... most power implementations will
actually do better than loop multiplying since
power can be optimized in various neat ways...
*runs back to doghouse where his wife is kind enuff
to let him sleep in :P *--Hawke'
