Need a pow() function without math.h

>> there are function-name conflicts, particularly with "log", and I don't have leave to go and change all the conflicting function names in this program

I would make the time to do that, since you'll have to make the changes anyway if you use your own code to implement log and pow. Conflicts aren't magically resolved by providing a different implementation ;)

Or maybe I'm misunderstanding the issue.

In any case, if you want your own implementation of pow, you could do exp(b * log(a)) which is the same as pow(a, b).

Or you can take the code from glibc compiler for example.

Greetings Phoenix, thanks for taking the time to respond.

I had reviewed that response; unfortunately sunnycoder failed to mention that the #define referenced other #defines which he/she didn't include (I see __tgmath_real_type  at a glance).   The "accepted solution" was a forced-accept which led to a broken link (or useless link anyway)

Honestly I tried to do exactly that last night but failed because I also grabbed the #define pow() part which seemed to reference itself and wouldn't compile.  If necessary I'm going to try again tonight with just the # define __TGMATH_BINARY_REAL_IMAG(Val1, Val2, Fct, Cfct) code and not the #define pow() part if it isn't critical.

Hi Infinity08, thanks for taking the time to respond.

For once, my reluctance to do as you suggest isn't simply rank laziness.  I really don't have liberty to change the name of the conflicting functions.

I'd be very satisfied with a "copy" of the standard glibc version renamed to pow_stolen() or something to be honest, therefore, conflict magically resolved.  (Note I don't 'need' the math.h library itself, just the one function)

Your alternate solutions require exp and log which are also parts of math.h :)

Thanks kindly!

In the link I posted, the pow function is written, however it's not just to cut and paste in the code, that's part of the glibc that I8 is mentioning.

As for the link, I have found the file which I'm attaching for you to review, however is just the math part, no code is written in the file.

I would use the log implementing a series, also the exp can be defined as a Taylor series.

>> Your alternate solutions require exp and log which are also parts of math.h :)

I know. But you mentioned conflicts with log, so I assumed that you already had your own implementation for that.

In any case, it's relatively easy to implement the exp and log functions.

The exp function can be approximated to arbitrary precision using Euler's continued fraction for example :

        http://en.wikipedia.org/wiki/Exponential_function#Continued_fractions_for_ex

The log function can be approximated to arbitrary precision using a series like :

        http://en.wikipedia.org/wiki/Logarithm#More_efficient_series

The ease of computation of both the exp and log functions is the reason why I suggested using that formula to implement pow.

>> I would use the log implementing a series, also the exp can be defined as a Taylor series.

Precisely :)

Since your code will hardly be redistributable if it conflicts with naming conventions in math.h, I assume you are targeting only for a single type of CPU and not for the "general audience".
In this case a bit of assembler might help, esp. for Intel CPUs, the F2XM1, FYL2X, FYL2XP1 instructions (they compute 2^x-1, y*ld(x), y*ld(x+1), hence the instruction sequcne for x^y is: load y, load x, FYL2X, F2XM1, FLD1, FADDP)

Bit of explanation:  The code isn't my own and I can only change parts of it.  The code does not include many standard header files.  It obviously does not include math.h (to my despair).

I8: conflicts with log are by name only: the application "logs" things in hundreds of places throughout the code calling a "log()" function (not my devising.)

Thus far the responses have been far above my level of comprehension and I'm concerned about wasting everyone's time.   Is there a way I can flag the question to indicate a lower level of knowledge and garner responses tailored to my level?  Ideally I'm hoping to see a response including an actual function/macro/define that performs x^y (and if that's irresponsible I'll kill the question and figure out how to ensure everyone who has responded gets a share of the points.)

Ideally I'm hoping to see a response including an actual function/macro/define that performs x^y WHERE y can be a fractional number (I've seen integer-only functions that work just fine)

I8,

You're very patient, thank you.  You don't think including mymath.h, which includes math.h, won't simply bring up the same conflicts?  (I won't be able to confirm either way for 3 hours).  I gather I'd also need to include mymath.c in the Makefile.

This is just the sort of clever response I was hoping for, I'm crossing my fingers it works :)

>> You don't think including mymath.h, which includes math.h

The thing is that mymath.h doesn't include math.h. The only file that includes math.h is mymath.c. That way, it's completely hidden from the rest of the code.


>> I gather I'd also need to include mymath.c in the Makefile.

Indeed.


>> I'm crossing my fingers it works :)

Let me know :)

Ok,

here is what I found... and I think it's going to work for you :)...

Here is the link when I found the library, no time to implement a pow function at work, this one seems to work fine.

http://www.hxa.name/articles/content/fast-pow-adjustable_hxa7241_2007.html

you will be able to download the following file that implements the pow function:
fast-pow-adjustable_hxa7241_2007.zip

Hope it helps

Hey l8, that's just a trick :P, but well, it will work to trick someone that doesn't know that the math.h is linked to that library.

Actually the link in the mentioned forced accept that you referenced (first few posts) wasn't broken, just outdated.  Here's a new link to the same PDF

http://math.ucalgary.ca/~kwooding/papers/hardware/gordon98survey.pdf

Hey l8, what you think about the link I posted for the library, seems fine to me :)...

>> Hey l8, that's just a trick :P, but well, it will work to trick someone that doesn't know that the math.h is linked to that library.

I'm not sure what you mean. But it's not a "trick" ... it's a valid design pattern called 'adapter'.

It's also not intended to "trick" somebody - it's intended to provide an easy way to get around the limitation described by tknudsen-qec.


Note that the problem was not with the math library, but with the name conflicts (introduced by including the math.h header file in the same file that uses a conflicting function).

In case you are only developing for x86 cpu's (and use the gnu compiler), see attached code that implements pow() using assembler.

And here's yet another way to include <math.h> without disturbing log:

#define log some_obscure_name
#include <math.h>
#undef log

#include <stdio.h>
 
float mypow(float x, float y) {
        float result;
        asm __volatile(
                "flds %2; flds %1; fyl2x;"
                "fld %%st;" //"flds %2; flds %1; fyl2x;"
                "frndint;fxch %%st(1);"
                "fsub %%st(1),%%st; f2xm1;"
                "fld1;faddp;fscale;fstp %0;faddp;"
                : "=g"(result)
                : "g"(x), "g"(y)
        );
        return result;
}
 
int main() {
        int i;
        float y = 3.14;
        for (i=1; i<10; ++i) {
                printf("%d ^ %f == %f\n", i,y,mypow(i,y));
        }
        return 0;
}
 
 
Output:
1 ^ 3.140000 == 1.000000
2 ^ 3.140000 == 8.815242
3 ^ 3.140000 == 31.489140
4 ^ 3.140000 == 77.708481
5 ^ 3.140000 == 156.590668
6 ^ 3.140000 == 277.584381
7 ^ 3.140000 == 450.409912
8 ^ 3.140000 == 685.019043
9 ^ 3.140000 == 991.565918
                                              

1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
29:
30:
31:
32:
33:
34:
35:
36:

>> Hey l8, what you think about the link I posted for the library, seems fine to me :)...

It could be (i didn't look at it in detail). But I honestly prefer to use the math.h library whenever possible - it's easier, more portable, and probably more efficient too.

Infinity08

I would have responded to this sooner but our satellite linkup failed in a storm last night.

I'm pleased that not only do I have access to the pow() function, I can hook into any of the math functions as needed.

I asked for change for the bus and you handed me a blank cheque.  THANKS :)