How many digit after the comma a negative float variable can take – 9 days old

[yvesmazzon]

Hello Picsters,

How many digit after the comma a negative float variable can take or what precision?

Regards,

Yves

[tumbleweed]

If you're talking about the decimal point position, it really doesn't matter much where it is until you start to approach the extreme end of the range (E+-38, or there abouts). You get 6-7 decimal digits total.

Keep that in mind if you do operations with variables that cover a wide range (ie large number + small number).

[John Drew]

Yves,
The important thing to understand is Significant Figures.
The accuracy of floating point maths available in PDS runs out after 6 or 7 significant figures.
Look up "Rules for significant figures" in Google then choose the www.nku.edu option as a useful one.
The tricky part is understanding when zero is significant or not.
John

[Les]

That's one of the reasons I created 64-bit floating point in Proton24. I'm sure it was John, who was developing astronomical equipment and he wanted a bit more accuracy, so I decided to create the 64-bit floats. Because they operate on the 16-bit architecture of the PIC24 and dsPIC devices, they are very fast to operate and very accurate. In Proton24, they are named "Double"

An 8-bit device is not suitable for 64-bit floating point IMO, because they take so much room up and are quite complex so will be quite slow on any 8-bit microcontroller platform.

[John Drew]

The 64 bit floats in Proton 24 work a treat and give great accuracy. The orbit of the moon is really complex and there are hundreds of lines of code so accuracy is critical. The moon is is one of the most difficult astronomical bodies to track. Because it's moving the calculations also need to be fast because1 sec updates are necessary. The P24 series are ideal with Les's double floats.
John

If any users make use of Floats I recommend reading this article in Microchip Developers:http://microchipdeveloper.com/c:unde...epresentations

[yvesmazzon]

Thank you all

Yves

[normnet]

For expected precision on today's larger and faster PICs see NBitMath for calculation values of up to 31 bytes wide with no rounding down etc.

For display and step by step debugging see macroFORMAT which converts the up to 31 bytes wide values into a string.

Norm