dtoa?

[pcmattman]

Does anyone know a (simple) implementation of the function dtoa, to convert a double type to a string?

I've got this as a base:

Code: Select all

// converts a double to an ascii string
void dtoa( double d, char* buff )
{
	// a float type
	union dpresc {
		struct sdat_t
		{
			unsigned long long mantissa : 52;
			unsigned long long exp : 11;
			unsigned long long sign : 1;
		} sdat __attribute((packed));
		double ddat;
	};
	
	union dpresc doub;
	doub.ddat = d;
	
	if( doub.sdat.sign == 0x0 )
		cons.Output( "POSITIVE" );
	else
		cons.Output( "NEGATIVE" );
}

Now how do I put the exponent and the mantissa together?

[Combuster]

You might want to read on IEEE Float.

A simple approach is to just load the mantissa, append the leading one, then shift based on the exponent and hope it doesnt overflow a long long int.

Try browsing a libc's source to find out a more complete approach.

[pcmattman]

Hmmm... I just realized a problem. I can't do arithmetic on 64-bit numbers. I get complaints about undefined functions - umod??

I would like to be able to have a ulltoa function.

I'll look into the site you gave.

Solar, does your PDCLIB have 64-bit arithmetic and a dtoa function??

[Brynet-Inc]

I found this.. "extensive" example of dtoa.. (Not the best coding style..)
http://netlib2.cs.utk.edu/fp/dtoa.c

I later found this for ulltoa and lltoa..
http://www.uwsg.iu.edu/hypermail/linux/ ... /0809.html

I didn't test them though... The dtoa has BSD-style licensing.. The other is unknown.

You can also play around with: http://www.koders.com/

[pcmattman]

I tried koders, it's pretty much my first reference point.

I'll look into those other links.

[Solar]

Solar, does your PDCLIB have 64-bit arithmetic and a dtoa function??

No float logic in PDCLib yet. 64bit-support is up to the compiler environment, dtoa() isn't a standard function - you'd have to use sprintf(), or use whatever internal function I will be using for the actual conversion - which would make your code PDCLib-specific and non-portable.

The problem with dtoa()-like functions is that they have to handle values off the sunny path (INF, NaN, ...).

[Combuster]

Hmmm... I just realized a problem. I can't do arithmetic on 64-bit numbers. I get complaints about undefined functions - umod??

Have you got / built (cross-compiler) libgcc and have you tried linking it in?

[pcmattman]

No, I'll have to try that...