CCS C Software and Maintenance Offers
FAQFAQ   FAQForum Help   FAQOfficial CCS Support   SearchSearch  RegisterRegister 

ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

CCS does not monitor this forum on a regular basis.

Please do not post bug reports on this forum. Send them to support@ccsinfo.com

float printing BUG workaround
Goto page 1, 2  Next
 
Post new topic   Reply to topic    CCS Forum Index -> General CCS C Discussion
View previous topic :: View next topic  
Author Message
kda406



Joined: 17 Sep 2003
Posts: 97
Location: Atlanta, GA, USA

View user's profile Send private message

float printing BUG workaround
PostPosted: Tue Mar 15, 2005 3:11 pm     Reply with quote

I just can't get floats to print out correctly with CCS. At first I thought it was a math problem, but after days of messing with it, I believe it is a float or printf problem. To illustrate my point I put together this simple program:

Code:
   int32 w = 3600;
   float f;
   f = w;
   fprintf(TERM,"%1.2f\n\r",f);


I expect it to print 3600.00, but it always prints 3599.99. No matter what value I put in w, it always prints out wrong value in f.

How can I work around this unacceptable problem?

-Kyle
Mark



Joined: 07 Sep 2003
Posts: 2838
Location: Atlanta, GA

View user's profile Send private message Send e-mail

PostPosted: Tue Mar 15, 2005 3:22 pm     Reply with quote

Has to do with floating point accuracy. You are not alone and if you search the forum you will find this posted many times. Maybe even within the last week. ttelmah posted a way to round this recently.
SherpaDoug



Joined: 07 Sep 2003
Posts: 1640
Location: Cape Cod Mass USA

View user's profile Send private message

PostPosted: Tue Mar 15, 2005 3:28 pm     Reply with quote

What happens if you try:
Code:

   int32 w = 3600;
   float f;
   f = w;
   f+= 0.005;
   fprintf(TERM,"%1.2f\n\r",f);

That moves the truncation point up a little. I would guess that f=3599.9999 or so and it is just truncating at an unfortunate place. To find out what is REALLY happening you should print f in hex and see what the exponent and mantissa are.

Also tell us what compiler version you are using.
_________________
The search for better is endless. Instead simply find very good and get the job done.
Mark



Joined: 07 Sep 2003
Posts: 2838
Location: Atlanta, GA

View user's profile Send private message Send e-mail

PostPosted: Tue Mar 15, 2005 6:47 pm     Reply with quote

http://www.ccsinfo.com/forum/viewtopic.php?t=22116&highlight=printf+float

http://www.ccsinfo.com/forum/viewtopic.php?t=21368&highlight=printf+float
Guest








PostPosted: Tue Mar 15, 2005 8:40 pm     Reply with quote

SherpaDoug wrote:
What happens if you try:
Code:

   int32 w = 3600;
   float f;
   f = w;
   f+= 0.005;
   fprintf(TERM,"%1.2f\n\r",f);



in stead of using f+= 0.005, I would like to use f *= 1.000001

besides, floating point multiplication is faster than addition!
Douglas Kennedy



Joined: 07 Sep 2003
Posts: 755
Location: Florida

View user's profile Send private message AIM Address

PostPosted: Wed Mar 16, 2005 2:23 am     Reply with quote

Any whole number less than 2^23 can be stored with perfect precision in the 4 byte floating point format CCS uses. 3600 certainly fits this criteria.
Now if you add .1 base ten or 1 to any negative power of ten then there is inherent differences due the the fundamental differences between the binary and the decimal systems.
The error has to be in the printf conversion to characters for display...its going to do divisions by ten against the float to tease out the digits and the error is accumulating in the sucessive divisions. Did you try casting back the result to a Int32 and printing the int32 out instead of the float?
kda406



Joined: 17 Sep 2003
Posts: 97
Location: Atlanta, GA, USA

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 8:36 am     Reply with quote

Thanks for all the replies. Adding 0.005 produces a result I might be able to live with. Multiplying like this "f *= 1.000001" didn't work. Perhaps I need to mess with the fudge factor, but I've wasted enough time on this stupid issue already.

Although I appreciate the information about the inherent differences in binary and decimal which I am painfully aware of, it is no excuse for the compiler not doing its job. Not one other C compiler I use has this problem. This is CRAP, CCS. Mad

The number 3600 in the example represents 36.00 degrees farenheight. Likewise, 14278 would be 142.78. I guess I can work out a 10-20 line program to print out this int32 as a pair of unrelated numbers with a dumb period in the middle, but geez, it really shouldn't be that hard.

FWIW, the main processor on this board (the one doing this printing) is an 18F8720. I'm not sure which compiler that uses in the PCWH system, PCH perhaps? Other C compilers are able to comiple highly optimized code for multiple different platforms (not just processors) by simply passing command line flags. I'm not sure why CCS feels like they need three different programs to compile for this one family of processors when others can handle altogether different plaforms from within a single compiler. But that's another shortcoming for another thread....

Again, thanks for the work arounds.

-Kyle
bluetooth



Joined: 08 Jan 2005
Posts: 74

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 8:39 am     Reply with quote

Perhaps you should save yourself all this frustration and use one of those other compilers! Rolling Eyes
kda406



Joined: 17 Sep 2003
Posts: 97
Location: Atlanta, GA, USA

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 9:33 am     Reply with quote

I suppose I am too idealistic in expecting a company I pay for a product to actually do their job and produce a working, comparable product. Hey I do software for a living, I know bugs are inevitable. The difference is, I fix bugs in my code. Printing floating point and doing FP math is not a new feature in CCS, which means they have left it incomplete for years.

As far as your comment about using another compiler, every compiler company has their initial cost and annual fees. With all the other fingers in the pie, my company's budget can only sustain the single PIC compiler I have already purchased, so I am stuck with CCS. Are you suggesting that everybody that has this problem change compilers too? Maybe that would get them to fix it.

Qutoing Mark:
Quote:
You are not alone and if you search the forum you will find this posted many times. Maybe even within the last week.


I am not out in left field here. Aparently lots of people are having this problem. If other compilers can overcome this problem, then so should CCS. I simply expect them to do their job, but I suppose you feel that is too much to ask.

-Kyle
bluetooth



Joined: 08 Jan 2005
Posts: 74

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 10:01 am     Reply with quote

Kyle:

Calm down - there's no need to get upset!

No, I am not suggesting that everyone switch compilers... it's just that all the energy that seems to being going in to these posts would be better spent just working around the problem.

As the quote from Mark points out, this is not a new problem. Most folks have a different version of this problem because they don't understand how floats are represented. This one appears to be related to how printf does the conversion. There are lots of ways to deal with this.

All environments have their features and limitations - you just have to weigh these against each other to best fit your needs. The amount of money CCS charges for this environment is, in my opinion, paltry. It performs way beyond what I would expect from such an inexpensive tool.

In the mean time, report the problem to CCS and perhaps they will address it in an upcoming release.

Laughing
mvaraujo



Joined: 20 Feb 2004
Posts: 59
Location: Brazil

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 11:33 am     Reply with quote

Kyle,

Out of the compiler discussion, is it really necessary going for float on the code?

I always keep it as simplest integer and use integer operations to get decimal parts.

It goes quick in most cases. And it can get quicker if you represent in power of 2 numbers.

Best regards,

Marcus
Guest








PostPosted: Wed Mar 16, 2005 12:00 pm     Reply with quote

kda406 wrote:
I suppose I am too idealistic in expecting a company I pay for a product to actually do their job and produce a working, comparable product. Hey I do software for a living, I know bugs are inevitable. The difference is, I fix bugs in my code. Printing floating point and doing FP math is not a new feature in CCS, which means they have left it incomplete for years.

As far as your comment about using another compiler, every compiler company has their initial cost and annual fees. With all the other fingers in the pie, my company's budget can only sustain the single PIC compiler I have already purchased, so I am stuck with CCS. Are you suggesting that everybody that has this problem change compilers too? Maybe that would get them to fix it.

Qutoing Mark:
Quote:
You are not alone and if you search the forum you will find this posted many times. Maybe even within the last week.


I am not out in left field here. Aparently lots of people are having this problem. If other compilers can overcome this problem, then so should CCS. I simply expect them to do their job, but I suppose you feel that is too much to ask.

-Kyle


Hey Microsoft make a huge bussiness doing this. If everything was perfect, there would be no real reason to upgrade.
Guest








PostPosted: Wed Mar 16, 2005 12:11 pm     Reply with quote

bluetooth wrote:
Kyle:

Calm down - there's no need to get upset!

No, I am not suggesting that everyone switch compilers... it's just that all the energy that seems to being going in to these posts would be better spent just working around the problem.

As the quote from Mark points out, this is not a new problem. Most folks have a different version of this problem because they don't understand how floats are represented. This one appears to be related to how printf does the conversion. There are lots of ways to deal with this.

All environments have their features and limitations - you just have to weigh these against each other to best fit your needs. The amount of money CCS charges for this environment is, in my opinion, paltry. It performs way beyond what I would expect from such an inexpensive tool.

In the mean time, report the problem to CCS and perhaps they will address it in an upcoming release.

Laughing


Maybe not in the situation of a PIC compiler, but I have seen freeware work just as good (even better) than payware. To say "you really can't expect more" is a cop-out and is why companies can get away with not fixing it.

I guess this problem is just another example of why open source is becoming so much more popular these days. Instead of waiting for a company to get their butt is gear with a fix, a community could have created a fix for it. Not just a hack!
bluetooth



Joined: 08 Jan 2005
Posts: 74

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 1:44 pm     Reply with quote

To the 1:11 pm Guest:

No cop out - you get what you pay for. I think CCS gives you a LOT of value for what you pay for - if you wanted to, there are hundreds of little things with this (and any) tool suite that you could complain about. Make no mistake - compilers are TOUGH to implement, and good ones are even tougher.

Idea
Maybe someone will organize an effort to make a free, open source PIC environment that's superior to CCS, Microchip, Hi-Tech et. al. For starters, it better handle 256 bit floats Shocked perfectly and better darn well have pointers to constants that use NO extra code space to implement Cool . And it better work equally well on a 12C508 as an 18F8720! And don't forget - FREE!!!!!

I can't wait!

Wink
PCM programmer



Joined: 06 Sep 2003
Posts: 21708

View user's profile Send private message

PostPosted: Wed Mar 16, 2005 2:23 pm     Reply with quote

I think there might be a bug.

According to Microchip AN575, if a decimal number consists
of the sums of powers of 2, then it can be represented exactly
in their floating point notation. This is true for 3600:
Code:

  2048
+ 1024
+  512
+   16
--------
  3600


To find out what the floating point representation of 3600 is,
go to this page, and download the file called floatconv10.zip.
http://www.sxlist.com/techref/microchip/math/fpconvert.htm
Unzip the executable program called floatconv.exe and run it.

Select "Microchip 32 bit" on the left side and type in 3600 in the
top box. Then click "Convert to Byte". The results are 8A 61 00 00.

Also, compile the following test program with PCM 3.221, and look
at the .LST file. It also has 8A 61 00 00.
Code:
#include <16F877.H>
#fuses XT, NOWDT, NOPROTECT, BROWNOUT, PUT, NOLVP
#use delay(clock = 4000000)
#use rs232(baud=9600, xmit=PIN_C6, rcv=PIN_C7, ERRORS)
//============================
main()
{
float f;

f = 3600.00;

printf("%f \n\r", f);

while(1);
}


Here is part of the .LST file. It's storing 8A 61 00 00.
Code:
.................... f = 3600.00; 
*
017B:  CLRF   24    // 00
017C:  CLRF   23    // 00
017D:  MOVLW  61    // 61
017E:  MOVWF  22
017F:  MOVLW  8A    // 8A
0180:  MOVWF  21


So the number 3600 is being stored correctly. The problem must
be in the CCS runtime code for printf, that converts from floating point
to ASCII. Microchip has AN670, which shows ASM code to convert
from floating point to ASCII base 10. I haven't gone through that code
and compared it to the CCS routines, but that would be the next step.
Display posts from previous:   
Post new topic   Reply to topic    CCS Forum Index -> General CCS C Discussion All times are GMT - 6 Hours
Goto page 1, 2  Next
Page 1 of 2

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group