Floating Point calculation

Jkane · 2015-03-12 04:53

Hello,

I have a question on floating point,

I using floating point,

tMath : "FloatMath"
fString : "FloatString"

and performing this calculation

(0.1 * 13) / 13 which should give me 0.1 as output

but I am getting 0.07692308

Hz: 13 GPM: 0.07692308

I attached the code:

{Object_Title_and_Purpose}


CON
        _clkmode = xtal1 + pll16x                                               'Standard clock mode * crystal frequency = 80 MHz
        _xinfreq = 5_000_000

VAR
                
  long Transition_counter
  long pin_state
  long previous_pin_state
  long GPM
  long GPM_temp
 
  

OBJ
   term   : "FullDuplexSerial"
   Math : "FloatMath"
  '  fMath : "Float32full"   
  String : "FloatString"
  
PUB Main | t  

term.start(31, 30, 0, 115200)



dira[8]~  ' set to input := 0

'Hz is frequeny per second
' count the number of transisions from high->low->high

repeat
  term.tx(13) 
  term.str(string("Hz:  "))
   transition_counter := 13
  term.dec(transition_counter)
  'GPM := (0.1 * transition_counter) / 13
  fmath.FFloat(transition_counter)
  GPM_temp := fmath.FMul(0.10, transition_counter)
  GPM := fmath.FDiv(GPM_temp,13)
  term.str(string(" GPM:  ")) 
  term.str(fstring.FloatToString(GPM))
  
  GPM_temp := 0
  GPM := 0
  Transition_counter := 0           ' reset  Transision counter
  
   t := clkfreq + cnt               ' get next second
  repeat while cnt < t              ' loop until second is up
     waitcnt(clkfreq/1000 + cnt)    ' wait one millisecond
     pin_state := ina[8]         ' check high or low
    
        if pin_state == previous_pin_state
             'do nothing
        else
            previous_pin_state := pin_state   ' make new pin state
            transition_counter++              ' and count the transision

regards

Jeff

Leon · 2015-03-12 05:09

Use code tags!

Jkane · 2015-03-12 05:24

Ok,

Jeff

Bean · 2015-03-12 05:56

Jeff,
I think the line

fmath.FFloat(transition_counter)

should be

transition_counter := fmath.FFloat(transition_counter)

Bean

Jkane · 2015-03-12 06:08

Bean

that yields

Hz: 13 GPM: 5.444519e+38

regards

Jeff

Bean · 2015-03-12 06:14

Okay,
Oh, I see. I think this needs changed too.

GPM := fmath.FDiv(GPM_temp,13)

to

GPM := fmath.FDiv(GPM_temp,FFloat(13))

Basically you can't use integers ANYWHERE you are doing floating point stuff. Without using FFloat.
I'm not sure, but you might be able to just use 13.0 too.

Bean

P.S. Yes, I tried it. using 13.0 will work.

ersmith · 2015-03-12 06:14

The line

 GPM := fmath.FDiv(GPM_temp,13)

is mixing floating point and integer -- the 13 should probably be 13.0. That error is partially masked by the one Bean noticed above, where transition_counter is never updated to be in floating point. The result is that you're doing all the math using the floating point number whose binary representation is 0x0000000d (13), which is a very tiny floating point number -- small enough so that you're encountering significant round off errors. The floating point representation of 13.0 is actually 0x41500000.

Jkane · 2015-03-12 06:43

Yes, it worked, picky, picky, picky.

anyway, the changes

'GPM := (0.1 * transition_counter) / 13
  transition_counter := fmath.FFloat(transition_counter)
  GPM_temp := fmath.FMul(0.10, transition_counter)
  GPM := fmath.FDiv(GPM_temp ,13.0)
  term.str(string(" GPM: ")) 
  term.str(fstring.FloatToString(GPM))

output is now correct

Hz: 13 GPM: 0.1

thanks for the help

regards

Jeff

Duane Degn · 2015-03-12 10:10

The object "F32" is a new and improved floating point object. Lonesock sped up the methods and was able to fit what used to take two cogs into a single cog.

The latest version (i think it's 1.6) is in the OBEX. F32 is also included in the archive attached to my integer vs floating point trig speed tests.

Floating Point calculation

Comments