Optimization IS important

[Kilinich]

I just can't stand that! WTF is that in "Some extra math functions":

Code: Select all

scene.my.sign = (number) =>
{
    sign := 0;
    (number<0)?{sign=(-1)}:false;
    (number>0)?{sign=1}:false;
    sign;
};
scene.my.abs = (number) =>
{
    number*scene.my.sign(number);
};

it's crazy. who did that?
I'll just re-write it:

Code: Select all

scene.my.sign = (x) => {x < 0 ? -1 : 1};
scene.my.abs = (x) => {x < 0 ? -x : x};


See the difference? KISS, and your thyme code will work as fast as a rocket.

[Livirus]

See the difference?

Yeah your sign(0) returns -1.

Code: Select all

scene.my.sign = (x) => {x < 0 ? -1 : 1};
scene.my.abs = (x) => { (x^2)^.5 }

...might be preferable, since sign(0) would now return 1 and abs does no longer require an if-statement.

[KarateBrot]

...might be preferable, since sign(0) would now return 1

sign(0) should return 0 by definition.

[Livirus]

...might be preferable, since sign(0) would now return 1

sign(0) should return 0 by definition.

Ye I know. But if you can live without the 0-exception, I think returning 1 is preferable. It's situational though.

[tatt61880]

Added Kilinich's definitions.
Some extra math functions

[Livirus]

I was playing around with math and noticed (0^x)^x returns 1 for all positives, 0 for zero, and +inf for all negatives. So all we need is an exception for x<0 and we're done :

Code: Select all

scene.my.sign = (x) => { x < 0 ? {-1} : 0^(0^x) };

It seem to run perfectly, including on zero.

Short?
Simple?
Fast?

EDIT:
Here's an option:

Code: Select all

scene.my.sign = (x) => { x != 0 ? x/(x^2)^.5 : 0 };

[tatt61880]

I think, which original scene.my.sign is defined only for scene.my.abs.
So, we seldom use sign() function which requires sign(0) = 0, don't we?

[Livirus]

I think, which original scene.my.sign is defined only for scene.my.abs.
So, we seldom use sign() function which requires sign(0) = 0, don't we?

Are you saying abs(0)=0*sign(0)=0 despite sign(0)=1, thus sign(0)=0 is not required?
As both Kilinich and I have demonstrated more efficient abs-functions that are independant of the sign-function, how does that make any sense?
And why would the abs-function be the only application of the sign-function?

[tatt61880]

I think, which original scene.my.sign is defined only for scene.my.abs.
So, we seldom use sign() function which requires sign(0) = 0, don't we?

Are you saying abs(0)=0*sign(0)=0 despite sign(0)=1, thus sign(0)=0 is not required?
As both Kilinich and I have demonstrated more efficient abs-functions that are independant of the sign-function, how does that make any sense?
And why would the abs-function be the only application of the sign-function?

When do you use sign() function?
I think sign() which sign(0) should be 0 is very rare to be used.
Even sign() which allow sign(0)!=0 is rare to be used.

[KarateBrot]

sorry i didn't want to start a discussion i just read that for programming 0 often gets let out so that it only takes one bit for calculations so there's no problem

[highly]

See the difference?

Yeah your sign(0) returns -1.

Code: Select all

scene.my.sign = (x) => {x < 0 ? -1 : 1};
scene.my.abs = (x) => { (x^2)^.5 }

...might be preferable, since sign(0) would now return 1 and abs does no longer require an if-statement.

How is scene.my.abs = (x) => { (x^2)^.5 } preferable? it's MUCH less efficient than other methods :/

Scene.my.abs = (x) => {x < 0 ? (x * (-1)):x} works for every situation. I also ran these through multiple iterations tests. for the purpose of the test, I use abs1 = (x) => {x < 0 ? (x * (-1)):x} and abs2 = (x) => { (x^2)^.5 }

Code: Select all

for 1000 iterations, 2000 tests each per iteration:
     abs1: 404ms
     abs2: 1157ms

I used an array of random floating point numbers in the range of -1000 to 1000 to test. EACH function was tested 2,000,000 times, and I calculated the computation time by using milliseconds from the system date and time.

As you can see, your proposed "preferred" method takes twice as long as the method I've shown. When you do (x^2)^.5, you perform relatively expensive power operations, and the .5 translates into a square root calculation, which is INCREDIBLY expensive to compute.

The method I used for abs1 is also all inclusive, it works for every possible number.

To show the difference between the methods, here's a test that goes through 5,000,000 iterations:

Code: Select all

for 2500 iterations, 2000 tests each per iteration:
     abs1: 966ms
     abs2: 2853ms

Over a larger sample we see that the abs function you seem to consider "preferable" is very close to 3 times slower than mine. Not including the "if" statement in NO way makes it preferable, "ease of writing" should never be favored if you're giving up almost 3 times as much processing time.

[Kilinich]

I've fix sign(0), thanks.
I usually call it as {value^2 * sign(value)} so sign(0) is not important in that case...