Subject:
|
WARNING WRONG ARITHMETICS ON HB !!!
|
Newsgroups:
|
lugnet.robotics.handyboard
|
Date:
|
Mon, 21 Feb 2000 10:13:33 GMT
|
Viewed:
|
832 times
|
| |
|
|
Working with the free version of IC for Linux 2.860 BETA I discovered
faulty operations on my handyboard. Please download the following
program to check your own soft- and hardware.
Can anybody tell me the result with the commercial version of IC.
If the problems are solved this would make it worth paying for it.
DIRK
/*
Testprogram for your Handyboard-Configuration
---------------------------------------------
This program shows faulty arithmetic operations on
Interactive C for 6811. Version 2.860 BETA (Dec 7 1994)
(Linux-Version)
Pcoder on Board was either
V 2.81 9/28/93 or
V 2.82 6/11/97 (smooth power routines)
Dirk Stueker (stueker@gmx.de)
University of Oldenburg, Department of Computer Science.
*/
void main(void) {
/* LONG-PROBLEMS */
if (-2147483647L==-2147483648L) beep();
/*
Shouldn't beep, test it, my board does beep!
I experienced, that any value less than -2147483647L is treated as
-2147483647L although actually -2147483648L is -maxlong (32bits)!
*/
if (((long) -1)==65535L)
printf("\nWhat's this? -1 == 65536!");
else
printf("\nResult ok!");
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
/*
Strange behavior, isn't it?
Whereas positive ints can be converted to long without any
problem,
negative ints are converted to 65536-value.
Conversion of float to long is done in an expected way.
*/
/* FLOAT-PROBLEMS */
printf("\n %f", 65537.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
/*
Should be somewhat greater than 1.
My board displays: .5000153
*/
printf("\n %f", 131074.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
printf("\n %f", 131075.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
/*
Both results should be somewhat greater than 2.
My board displays: 1.000031 and 1.000046
Even more surprising: The result for 131073.0 / 65536.0 is ok!
*/
printf("\n %f", 262148.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
printf("\n %f", 262149.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
printf("\n %f", 262150.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
printf("\n %f", 262151.0 / 65536.0);
while (!start_button()); while (start_button()); /* wait until start
pressed
and released */
/*
All results should be somewhat greater than 4.
My board displays: 2.000061, 2.000076, 2.000092 and 2.000107
Same behavior as above: result for 262147.0 / 65536.0 is correct.
After testing a lot, I discovered some systematic error.
n Errors are found at the following numbers:
2^(n-1) * 65536 + 2^(n-1) / n = 1,2,3, ...
See above:
First error at 1 * 65536 + 1 = 65537.
Next TWO errors at: 2 * 65536 + 2 = 131074 and 131075
Next FOUR errors at: 4 * 65536 + 4 = 262148 - 262151
...
In each case the value must be multiplied by 2 to get the correct
result!
The same behavior can be found if the negative values are used.
*/
printf("\n-terminated-");
}
--
------------------------------------------------------------------------
Dirk Stueker Zietenstr. 36 26131 Oldenburg Germany
Department of Computer Science University of Oldenburg
Tel. & Fax: +49 / 441 / 5 39 35
========================================================================
Home-P: http://condor.informatik.uni-oldenburg.de/stueker
E-Mail: stueker@gmx.de (=Dirk.Stueker@Informatik.Uni-Oldenburg.de)
------------------------------------------------------------------------
|
|
1 Message in This Thread:
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
