Fixed Point to Floating Point

RogerInHawaii

I have a fixed point variable in my Spin code. It has 16 bits of integer value and 16 bits of fractional value. It's coming from an external device in Fixed Point format so I have no direct control over its format. I want to be able to use that in some floating point operations, such as FloatMath.FMul(MyFixedPointVal, MyFLoatingPointVal).

But that clearly doesn't work since the FMul expects both arguments to already be floating point values.

How can I convert my Fixed Point variable into the Floating Point format?

Timmoore

Receive it as 2 integers (separated by .). Convert both to float, then divide fractional part by 2**16 and then add integer prt.

John Abshier

I think we need more info on the format. Are numbers positive only? This may work; it is off the top of my head and not tested.
MyFloatInt := F.FFloat(MFP >> 16) MyFloat is the integer portion of the value. Or
MyFloatInt := F.FFloat(MFP ~> 16) if the first bit is a sign
temp := MFP & $FFFF get the lower 16 bits
MyFloatFrac := F.FDiv(F.FFloat(temp), F.FFloat(65536)) could be 65535
MyFloat := F.FAdd(MyFloatInt, MyFloatFrac)

John Abshier

RogerInHawaii

Yes, the values are always positive.

Those approaches, using FFloat to convert and then combine, looks good!

I'll give it a try.

Thanks

mpark

John Abshier said...
I think we need more info on the format. Are numbers positive only? This may work; it is off the top of my head and not tested.
MyFloatInt := F.FFloat(MFP >> 16) MyFloat is the integer portion of the value. Or
MyFloatInt := F.FFloat(MFP ~> 16) if the first bit is a sign
temp := MFP & $FFFF get the lower 16 bits
MyFloatFrac := F.FDiv(F.FFloat(temp), F.FFloat(65536)) could be 65535
MyFloat := F.FAdd(MyFloatInt, MyFloatFrac)

John Abshier

Wouldn't it work just as well simply to divide MFP by 65536? I.e., F.FDiv( F.FFloat(MFP), 65536.0 )

Jay Kickliter

If you can, the best thing would be to keep it in fixed point. You'd save a lot of resources by not having to go FP. But, if you're already using FP, then mpark's solution is the way to go.

Jay