[sdcc-devel] [ sdcc-Bugs-1166494 ] dont get proper result for while (x > 10) where x is float

[SourceForge.net <no...@so...>]

Bugs item #1166494, was opened at 2005-03-19 07:28
Message generated for change (Comment added) made by epetrich
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1166494&group_id=599

Category: msc51(8051) target
Group: non bugs
Status: Closed
Resolution: Rejected
Priority: 5
Submitted By: shailesh arya (shailarya)
Assigned to: Nobody/Anonymous (nobody)
Summary: dont get proper result for while (x > 10) where x is float

Initial Comment:
i have attached my source file with it T2.c. it compiled 
properly and generate hex code but it goes in following 
loop, but it should not go in that way.



----------------------------------------------------------------------

>Comment By: Erik Petrich (epetrich)
Date: 2005-03-24 01:19

Message:
Logged In: YES 
user_id=635249

This part of your program:
    for(;a>1;)
     {
     a=a*10;
     w++;
     *p++='0';
     }
overflows your str[16] buffer when a>1, causing other
variables to be modified in undefined ways.

If I delete the above code, after the program exits I find
that str[16] contains "-101199". Is your concern that this
is not "-101200"?


----------------------------------------------------------------------

Comment By: shailesh arya (shailarya)
Date: 2005-03-23 23:20

Message:
Logged In: YES 
user_id=825443

i have sdcc 2.4.8 977. there is a big problem. still i think it is 
not correct as compre to normal logic. pls consider small 
code as follow. still the compre function doe not work as per 
the logic. for the +/- value of a. try with different value of a less 
then 1.  it give worng result.
dont go for sbuf. this will work properly. 

#include<at89x52.h>
#include<stdio.h>
#include<stdlib.h>
#include<math.h>
#include<float.h>

float compre (float a); 
void putchar (char c); 

unsigned char str[16], k;
data unsigned char *p;
unsigned char i,w;
float a;

void main  (void)
{
TH1 = 0xfd ;
TL1 = 0xfd ;
PCON &= 0x7f ;
SCON = 0x50 ;
TMOD = 0x25 ;
TCON = 0x40 ;
TR1=1;
p=str;
a =-.10120;
a = compre(a);
for(w=0;w<6;w++)			
{
     a=a*10;
    for(;a>1;)
     {
     a=a*10;
     w++;
     *p++='0';
     }
  k=a;
  a=a-k;
  *p++=k+'0';
 } 
for (i=0;i<16;i++)
{
SBUF = str[i];		
}
printf_fast("\n");
}

float compre (float a)
{
  if ( a < 0.0)   
	{ 
	a = - a; 
	*p++ = '-';
	}
return a;
}

----------------------------------------------------------------------

Comment By: Erik Petrich (epetrich)
Date: 2005-03-22 10:55

Message:
Logged In: YES 
user_id=635249

I still don't see a problem. With your new example, I get
the output:

3
90
1

Which appears to be correct since (unsigned
int)(2000*(2.0/10)) is 0x190. I get the same output when the
variables are declared globally or locally.

You did not specify the version number of the compiler you
are using. My results are with SDCC 2.4.8 #980.


----------------------------------------------------------------------

Comment By: shailesh arya (shailarya)
Date: 2005-03-22 02:36

Message:
Logged In: YES 
user_id=825443

epetrich
thanks. but even you dont go to full function like initialize and 
putchar. just simply do as following then also it gives a 
problem. valuse of a is change. if we remove print_fast then 
only it goes properly.

data unsigned char e[6];
data unsigned char *p;
unsigned char c,d;
float a;
unsigned int i;
e[0]=3;
p=e;
a=2.0;
printf_fast ("\n%d",*p);
if (a>0.0)
{
a=a/10;
i=(2000*a);
c=i;
d=(i>>8);
printf_fast ("\n%x",c);
printf_fast ("\n%x",d);
}
else 
{
i=-(300*a);
c=i;
d=(i>>8);
printf_fast ("\n%x",c);
printf_fast ("\n%x",d);
}
}

----------------------------------------------------------------------

Comment By: shailesh arya (shailarya)
Date: 2005-03-22 02:33

Message:
Logged In: YES 
user_id=825443

epetrich
thanks. but even you dont go to full function like initialize and 
putchar. just simply do as following then also it gives a 
problem. valuse of a is change. if we remove print_fast then 
only it goes properly.

data unsigned char e[6];
data unsigned char *p;
unsigned char c,d;
float a;
unsigned int i;
e[0]=3;
p=e;
a=2.0;
printf_fast ("\n%d",*p);
if (a>0.0)
{
a=a/10;
i=(2000*a);
c=i;
d=(i>>8);
printf_fast ("\n%x",c);
printf_fast ("\n%x",d);
}
else 
{
i=-(300*a);
c=i;
d=(i>>8);
printf_fast ("\n%x",c);
printf_fast ("\n%x",d);
}
}

----------------------------------------------------------------------

Comment By: Erik Petrich (epetrich)
Date: 2005-03-21 23:53

Message:
Logged In: YES 
user_id=635249

I can't find a problem with the compiler, but I can find
several problems with your example code. The first two are
the most serious.

1) In the call ftoa(a,str1,5,'f') the pointer str1 has not
been assigned a value. Since it is a global variable, it
will default to pointing to memory location 0. Writing to
this pointer will unexpectedly modify the values in the
first register bank, causing very strange behaviour. 

2) In the function initialize() you have the statement
SP=0x55. This modifies the stack pointer. Thus when the
initialize() function ends, it will fetch the return address
from this new stack pointer location and likely jump to some
random location rather than the expected location in main().
 If you must control the stack pointer location manually,
use the --stack-loc and ==no-pack-iram command line options.
By default, the compiler will automatically initialize the
stack pointer to a reasonable location for you.

3) Perhaps this was added just to try debug the problem, but
declaring ndiag at 0x50 is generally not a good idea, since
this will cause the specified address to be used for this
variable even if another variable was allocated there.

4) You are writing to the SBUF sfr without checking to see
if any previous character has completed transmission. Thus
some characters may be lost if the writes occur too quickly.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1166494&group_id=599