cost of Math functions

[Disarray-yarrasiD]

If someone could tell me the realitive cost of the following math functions it would be greatly appreciated:

+
-
*
/
Math.sqrt()
Math.pow()
Math.abs()

Thank you.

[Disarray-yarrasiD]

At 5/7/08 10:26 PM, Disarray-yarrasiD wrote: If someone could tell me the realitive cost of the following math functions it would be greatly appreciated:

+
-
*
/
Math.sqrt()
Math.pow()
Math.abs()

Thank you.

Math.sin cos and tan are would be nice to know too.

[authorblues]

At 5/7/08 10:26 PM, Disarray-yarrasiD wrote: +

atomic. single clock cycle.

-

atomic. single clock cycle.

*

minimal. several clock cycles.

/

minimal. several clock cycles.

Math.sqrt()

moderate. employs several arithmetic functions iteratively.
then its splined linearly.

Math.pow()

invokes Math.exp() and a few multiplications.
Math.exp() is about the same cost as Math.sqrt()

Math.abs()

minimal. a few conditionals.

[authorblues]

At 5/7/08 10:34 PM, Disarray-yarrasiD wrote: Math.sin cos and tan are would be nice to know too.

those are just splined table-lookups. moderate.
range reduction algorithms can be costly, but modern compilers optimise the process significantly.

[henke37]

I would not call division moderate, it's fairly slow. This also goes for modulous, %. Try to replace it with a multiply, or even better, a bit shift.

[Disarray-yarrasiD]

thank you very much

[thoughtpolice]

At 5/8/08 02:30 AM, henke37 wrote: I would not call division moderate, it's fairly slow. This also goes for modulous, %. Try to replace it with a multiply, or even better, a bit shift.

Chances are, the JVM (or more specifically, HotSpot) is smart enough to do Strength reduction where appropriate to reduce overhead in cases like these. Nearly any modern compiler these days will. On that note, it might be worth examining.