Custom Float Power Function

[LB]

Hello, I am trying to make a custom floating point power function of some sort. I need to do this because I am using a bignum library that lets me declare integers and floats with any number of bits I want, and I am dealing with very big numbers. I have absolutely now idea how to make a power function for this; I'm good at math, and OK at programming, but I can't get the two to mesh. I have done a lot of searching but I cannot find anything that will work correctly or accurately when I implement it. The bignum lib I use supports most things except for the power operators (^=, ^) and the modulo operators (%=, %) for floating point types.

If someone could set me on the right track for making a function for a super-big, super-accurate floating point power function, I'd highly appreciate it.

Thanks,
-LB

[gumaze]

Are you looking for modpow operations?

I found this project about big integers in c++. I don't know if they implement correctly the modpow algorithm, but worth trying.

The bad news is there isn't documentation :(

https://mattmccutchen.net/bigint/

[LB]

No, I want something that allows me to take a float as a base, and raise it to a float as the power, and return a float. I can't use any of the inbuilt C++ power functions because I have no source to them to change for my big float types. I was simply saying what the lib was capable of when I mentioned the modulo operators. (The lib is not finished but I was hoping to use a workaround for it until it was finished)

[helios]

Sorry, I don't really have an answer for you, but

super-big, super-accurate floating point

if you don't want to sound like a complete tool, I suggest you use the term "arbitrary precision".

[LB]

I was literally using the name of the type of variable defined by the library. It's literally called "BigInt" and such.

[ne555]

If the exponent is an integer you can use divide and conquer a^n = a^(n/2) * a^(n/2)

For float exponent, maybe this X^Y = e^(ln(X^Y)) = e^(Y*ln(X))

[LB]

Could you explain the floating point one more? I'm not sure which of those variables are integers and which are floating point types.

[Athar]

You should drop your current bignum library and use gmp instead.

[LB]

GMP doesn't work for me because my compiler always complains about it being in 'mac format" or whatever.

[Athar]

Then download the appropriate version for your platform.

[helios]

If the exponent is an integer you can use divide and conquer a^n = a^(n/2) * a^(n/2)

What? If n is an integer, you can just multiply a by itself n times. What are you talking about divide an conquer?

[ne555]

What? If n is an integer, you can just multiply a by itself n times. What are you talking about divide an conquer?

That will be O(n). Divide and conquer is O(log n).
If the number has a lot of digits the multiplication will cost you.

[LB]

Then download the appropriate version for your platform.

There are only two download links for GNP and both are in this "Mac Format". :(