Floating point math and strange results in the terminal

turbosupra · 2012-02-22 11:40

Parallax terminal reads: wrote:

additionAnswer is: 2
subtractionAnswer is: -2
multiplicationAnswer is: 16
divisionAnswer is: 3
serialDataExample1 is: 1
serialDataExample2 is: 1

I'm unsure as to why I'm receiving "-2" as the result for the subtractionAnswer variable, and why I'm required to put pst.Str(fs.FloatToString(additionAnswer)) and pst.Str(fs.FloatToString(subtractionAnswer)) but then for division and multiplication I have to put pst.dec(multiplicationAnswer) and pst.dec(divisionAnswer) ? I'm also unsure as to why I cannot get the serialDataExample1 and serialDataExample2 to divide correctly? I can upload the objects later if someone would like them, just not from work because of the firewall.

CON
_clkmode = xtal1 + pll16x
_xinfreq = 5_000_000
 

VAR

long numerator
long denominator
long constantMultiplier 
long additionAnswer
long subtractionAnswer
long multiplicationAnswer
long divisionAnswer

  

OBJ
pst : "Parallax Serial Terminal"
strings : "Strings"
f : "Float32"
fm : "FloatMath"
fs : "FloatString"



PUB Main

  {Addition: result = a + b
   Subtraction: result = a - b
   Multiplication: result = a * b
   Division: result = a / b}

  'numerator := 1640.0
  'denominator := 2.0
  'constantMultiplier := 1.221

  pst.Start(115_200)                                    ' starts with passed on variable of the baud rate                        
  pst.Clear 
  f.start

  pst.Str(String("Start" ))
  pst.Str(String(pst#NL))

 repeat

                             'x := f.FAdd(x, y) is correct - from the propeller floating point math pdf file

     additionAnswer := f.FAdd(2.0, 4) 
     subtractionAnswer := f.FSub(4, 2.0)
     multiplicationAnswer := f.FMul(8, 2.0) 
     divisionAnswer := f.Fdiv(6, 2.0)
     serialDataExample1 := (f.FMul(820, 0.001221))
     serialDataExample2 := ((f.FMul(1638, 1.221)/1000)) 
      
     pst.Str(String("additionAnswer is: " ))
     pst.Str(fs.FloatToString(additionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("subtractionAnswer is: " ))
     pst.Str(fs.FloatToString(subtractionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("multiplicationAnswer is: " ))
     pst.dec(multiplicationAnswer)
     pst.Str(String(pst#NL))

     pst.Str(String("divisionAnswer is: " ))
     pst.dec(divisionAnswer)
     pst.Str(String(pst#NL))

     pst.Str(String("serialDataExample1 is: " ))
     pst.Dec((serialDataExample1))
     pst.Str(String(pst#NL))

     pst.Str(String("serialDataExample2 is: " ))
     pst.Dec((serialDataExample2))
     pst.Str(String(pst#NL))

    WaitCnt((clkfreq) + cnt)

ElectricAye · 2012-02-22 11:59

I can't remember for sure about the use of decimal points in this object, but should

divisionAnswer := f.Fdiv(6, 2.0)

be written as divisionAnswer := f.Fdiv(6.0, 2.0)

????

Duane Degn · 2012-02-22 12:03

ElectricAye is right. You need ".0" on all the constants you're using as floating point.

Beside the "6" just mentioned, I also see a "4" without a ".0" on it. (And a 1638, 820 etc.)

Duane Degn · 2012-02-22 12:04

I bet

serialDataExample2 := ((f.FMul(1638, 1.221)/1000))

doesn't do what you want it to. You can't mix floating point and interger math like that.

turbosupra · 2012-02-22 12:21

Thanks! You both were right, the only issue now with the code below is that it truncates all values after the decimal point, what should I do to obtain 1 decimal place?

ElectricAye wrote: »

I can't remember for sure about the use of decimal points in this object, but should

divisionAnswer := f.Fdiv(6, 2.0)

be written as divisionAnswer := f.Fdiv(6.0, 2.0)

????

Duane Degn wrote: »

ElectricAye is right. You need ".0" on all the constants you're using as floating point.

Beside the "6" just mentioned, I also see a "4" without a ".0" on it. (And a 1638, 820 etc.)

Duane Degn wrote: »
I bet
serialDataExample2 := ((f.FMul(1638, 1.221)/1000)) 
doesn't do what you want it to. You can't mix floating point and interger math like that.

CON
_clkmode = xtal1 + pll16x
_xinfreq = 5_000_000
 

VAR

long numerator
long denominator
long constantMultiplier 
long additionAnswer
long subtractionAnswer
long multiplicationAnswer
long divisionAnswer
long serialDataExample1
long serialDataExample2

  

OBJ
pst : "Parallax Serial Terminal"
strings : "Strings"
f : "Float32"
fm : "FloatMath"
fs : "FloatString"



PUB Main

  {Addition: result = a + b
   Subtraction: result = a - b
   Multiplication: result = a * b
   Division: result = a / b}

  'numerator := 1640.0
  'denominator := 2.0
  'constantMultiplier := 1.221
  

  pst.Start(115_200)                                    ' starts with passed on variable of the baud rate                        
  pst.Clear 
  f.start

  pst.Str(String("Start" ))
  pst.Str(String(pst#NL))

 repeat

                             'x := f.FAdd(x, y) is correct - from the propeller floating point math pdf file

     additionAnswer := f.FAdd(2.0, 4.0) 
     subtractionAnswer := f.FSub(4.0, 2.0)
     multiplicationAnswer := f.FMul(8.0, 2.0) 
     divisionAnswer := f.Fdiv(6.0, 2.0)
     serialDataExample1 := (f.FMul(820, 0.001221))
     serialDataExample2 := ((f.FMul(1638, 0.001221))) 
      
     pst.Str(String("additionAnswer is: " ))
     pst.Str(fs.FloatToString(additionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("subtractionAnswer is: " ))
     pst.Str(fs.FloatToString(subtractionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("multiplicationAnswer is: " ))
     pst.Str(fs.FloatToString(multiplicationAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("divisionAnswer is: " ))
     pst.Str(fs.FloatToString(divisionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("serialDataExample1 is: " ))
     pst.Dec((serialDataExample1))
     pst.Str(String(pst#NL))

     'pst.dec(serialDataExample2)
     'pst.Str(String(pst#NL))
     pst.Str(String("serialDataExample2 is: " ))
     pst.Dec((serialDataExample2))
     pst.Str(String(pst#NL))

    WaitCnt((clkfreq) + cnt)

Duane Degn · 2012-02-22 12:38

Try using the SetPrecision method of the FloatString object.

turbosupra · 2012-02-22 12:58

Hi Duane,

I put this line right under f.start

fs.SetPrecision(4)

and still nothing. I tried f.FRound in a few places as well without luck? Am I putting the fs.SetPrecision(4) in the correct location? If not, can you quote my code and show me where please?

Duane Degn wrote: »

Try using the SetPrecision method of the FloatString object.

Duane Degn · 2012-02-22 13:29

One of the SHT11 humidity object has some sort of floating point formatting.

I'm usually not worried about the formatting of my floating point projects.

You could use the "round" method and then output with a method like this:

PUB DecPoint(value, denominator) 
  if value < 0
    tx("-")
    -value
      
  if value => denominator
    result := value / denominator
    dec(result)
    value //= denominator     
  else    
    tx("0")
  tx(".")  
  repeat while denominator > 1
    denominator /= 10
    if value => denominator
      result := value / denominator
      dec(result)
      value //= denominator
    else
      tx("0")

You could add the method to your serial object. Just make sure and change all the "tx" calls to what ever the method is called in your serial object. For PST, all the tx calls will need to be changed to char calls.

So if you have a variable "x" with a number you want to display to one digit past the decimal, do this.

x := F.FMul(x, 10.0)
  x := F.FRound(x)
  Pst.DecPoint(x, 10)

This assumes you've added the DecPoint method to PST.

I think finding a floating point formatting method would be less awkward though.

Oh, lose the "Float32" and "FloatMath". Those objects are so last decade! F32 is the way to go. It's faster and fits in one cog. You shouldn't need to modify your code to use it.

The latest version of F32 is here. The OBEX version has a bug.

Edit (July 12, 2012): I fixed a bug the DecPoint method above. Thank you Gunstar1 for pointing it out.

turbosupra · 2012-02-22 14:10

Thank you.

I don't know why, but I cannot get this to work, it seems like it should be a very simple thing too. I did as you suggested with modifying the pst object, without luck and I cannot for the life of me get it to round up??? Below is the current status, I appreciate you sticking with me on this.

Parallax Serial Terminal reads wrote:

1
1
serialDataExample2 is: 1
0.0

CON
_clkmode = xtal1 + pll16x
_xinfreq = 5_000_000
 

VAR

long numerator
long denominator
long constantMultiplier 
long additionAnswer
long subtractionAnswer
long multiplicationAnswer
long divisionAnswer
long serialDataExample1
long serialDataExample2
long serialDataExample1b
long serialDataExample2b

  

OBJ
pst : "Parallax Serial Terminal"
strings : "Strings"
'f : "Float32"
f : "F32"
fm : "FloatMath"
fs : "FloatString"



PUB Main

  {Addition: result = a + b
   Subtraction: result = a - b
   Multiplication: result = a * b
   Division: result = a / b}

  'numerator := 1640.0
  'denominator := 2.0
  'constantMultiplier := 1.221
  

  pst.Start(115_200)                                    ' starts with passed on variable of the baud rate                        
  pst.Clear 
  f.start
  fs.SetPrecision(4)
  
  pst.Str(String("Start" ))
  pst.Str(String(pst#NL))

 repeat

                             'x := f.FAdd(x, y) is correct - from the propeller floating point math pdf file
     
     serialDataExample1 := (f.FMul(820, 0.001221))
     serialDataExample2 := (f.FMul(1638, 0.001221))
     pst.Dec((serialDataExample2))
     pst.Str(String(pst#NL))
     serialDataExample2b := f.FRound(f.FFloat(serialDataExample2))
     pst.Dec((serialDataExample2b))
     pst.Str(String(pst#NL))
    
     pst.Str(String("serialDataExample2 is: " ))
     pst.Dec(f.FRound(f.FFloat(serialDataExample2)))
     pst.Str(String(pst#NL))
     pst.DecPoint((serialDataExample2), 10)
     pst.Str(String(pst#NL))
     pst.Str(String(pst#NL))

      'serialDataExample2b := f.FFloat(serialDataExample2)

     'pst.dec(serialDataExample2)
     'pst.Str(String(pst#NL))
     
    WaitCnt((clkfreq) + cnt)

Duane Degn wrote: »
One of the SHT11 humidity object has some sort of floating point formatting.

I'm usually not worried about the formatting of my floating point projects.

You could use the "round" method and then output with a method like this:
PUB DecPoint(value, denominator) 
  if value < 0
    tx("-")
    -value
      
  if value > denominator
    result := value / denominator
    dec(result)
    value //= denominator     
  else    
    tx("0")
  tx(".")  
  repeat while denominator > 1
    denominator /= 10
    if value > denominator
      result := value / denominator
      dec(result)
      value //= denominator
    else
      tx("0")
      
You could add the method to your serial object. Just make sure and change all the "tx" calls to what ever the method is called in your serial object. For PST, all the tx calls will need to be changed to char calls.

So if you have a variable "x" with a number you want to display to one digit past the decimal, do this.
x := F.FMul(x, 10.0)
  x := F.FRound(x)
  Pst.DecPoint(x, 10)
The below was added to pst.spin
PUB DecPoint(value, denominator) 
  if value < 0
    Char("-")
    -value
      
  if value > denominator
    result := value / denominator
    dec(result)
    value //= denominator     
  else    
    Char("0")
  Char(".")  
  repeat while denominator > 1
    denominator /= 10
    if value > denominator
      result := value / denominator
      dec(result)
      value //= denominator
    else
      Char("0")
  
This assumes you've added the DecPoint method to PST.

I think finding a floating point formatting method would be less awkward though.

Oh, lose the "Float32" and "FloatMath". Those objects are so last decade! F32 is the way to go. It's faster and fits in one cog. You shouldn't need to modify your code to use it.

The latest version of F32 is here. The OBEX version has a bug.

Duane Degn · 2012-02-22 14:13

You still have numbers you're treating as floating point that don't end with ".0". Did you multply the floating point number by 10.0 before rounding?

Duane Degn · 2012-02-22 14:14

Duane Degn wrote: »

(And a 1638, 820 etc.)

Duane Degn · 2012-02-22 14:22

Floating point is useful for trig and log functions but if your just using multiply and divide you can use pseudo real numbers.

Instead of "0.001221" use 1221 and just remember it has been multiplied by 1,000,000 and compensate later when you do other calculations on it.

You would use DecPoint to display it with 1,000,000 as the denominator.

Duane Degn · 2012-02-22 14:28

I found the FloatString object with the format method. Watch out, it has the same name as the other object.

turbosupra · 2012-02-22 14:46

Ok, that was it. I'm going to have to force my ADC values to a float I guess with the f.FFloat method? Thanks for pointing that out. I can get a rounded version and a decimal version now ... excellent! Thanks again. I'll post the code in case this thread turns up in someones search.

Parallax Serial Terminal reads wrote:

additionAnswer is: 6
subtractionAnswer is: 2
multiplicationAnswer is: 16
divisionAnswer is: 3
serialDataExample1 is: 1065363450
1
serialDataExample2b is: 2
serialDataExample2 is: 2
1987788
1.9

repeat

                             'x := f.FAdd(x, y) is correct - from the propeller floating point math pdf file
     
     additionAnswer := f.FAdd(2.0, 4.0) 
     subtractionAnswer := f.FSub(4.0, 2.0)
     multiplicationAnswer := f.FMul(8.0, 2.0) 
     divisionAnswer := f.Fdiv(6.0, 2.0)
     serialDataExample1 := (f.FMul(820.0, 0.001221))
     serialDataExample2 := (f.FMul(1628.0, 0.001221))
     serialDataExample3 := (f.FMul(1628.0, 1221))
     'serialDataExample1 := (f.FMul(f.FFloat(820), 0.001221))
     'serialDataExample2 := (f.FMul(f.FFloat(1628), 0.001221))
     'serialDataExample3 := (f.FMul(f.FFloat(1628), 1221))
     pst.Dec(f.FTrunc(serialDataExample2))
     pst.Str(String(pst#NL))
     serialDataExample2b := f.FRound((serialDataExample2))
     pst.Str(String("serialDataExample2b is: " ))
     pst.Dec((serialDataExample2b))
     pst.Str(String(pst#NL))
    
     pst.Str(String("serialDataExample2 is: " ))
     pst.Dec(f.FRound(serialDataExample2))
     pst.Str(String(pst#NL))
     pst.Dec(serialDataExample3)
     pst.Str(String(pst#NL))
     pst.DecPoint((serialDataExample3/100000), 10)
     pst.Str(String(pst#NL))
     pst.Str(String(pst#NL))

    pst.Str(String("additionAnswer is: " ))
     pst.Str(fs.FloatToString(additionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("subtractionAnswer is: " ))
     pst.Str(fs.FloatToString(subtractionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("multiplicationAnswer is: " ))
     pst.Str(fs.FloatToString(multiplicationAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("divisionAnswer is: " ))
     pst.Str(fs.FloatToString(divisionAnswer))
     pst.Str(String(pst#NL))

     pst.Str(String("serialDataExample1 is: " ))
     pst.Dec((serialDataExample1))
     pst.Str(String(pst#NL)) 


   WaitCnt((clkfreq) + cnt)

Duane Degn wrote: »

Floating point is useful for trig and log functions but if your just using multiply and divide you can use pseudo real numbers.

Instead of "0.001221" use 1221 and just remember it has been multiplied by 1,000,000 and compensate later when you do other calculations on it.

You would use DecPoint to display it with 1,000,000 as the denominator.

Paul K. · 2012-02-22 21:35

I had the same problem. I use mpc3208 a/d's and I float them all. You have to float the adc values if you use it in a floating equation. Or use the adc value in a non floating equation. Or trunc the other value before using the adc value in a non floating equation.

F.ffloat method is correct. You can throw in to a equation like. Sum := f.fadd(f.ffloat(adc1),2.0) or just float them as you read the value in.

turbosupra · 2012-02-23 08:16

Thanks Paul.

Do you have any other 3208 interface tips, so I can save myself the headache?

I'm hoping this post will come up in a search to save future users the headache of banging their head through this process like I did. It's a lot easier to work backwards from a working set of code, then to work through all of the issues to get to that working set of code.

Floating point math and strange results in the terminal

Comments