![](templates/subSilver/images/CCSLogo.jpg) |
![CCS C Software and Maintenance Offers](templates/subSilver/images/forumAd6.jpg) |
View previous topic :: View next topic |
Author |
Message |
RVR
Joined: 04 Jan 2012 Posts: 18
|
float math operations problems |
Posted: Sat Jan 28, 2012 4:10 pm |
|
|
Hi everyone,
I'm trying to operate with precision numbers (ex: 146516.728, 0.0.52118, etc) in a math operations, and with algorithm that use a data type "double".
Before to program the main algorithm, I have some troubles that must to solve.
1.- floor function problem,
Code: |
#Include <18f2550.h>
#use delay (clock=48000000)
#use Rs232( baud=9600,xmit=Pin_c6,rcv=Pin_c7,parity=n,bits=8)
#INCLUDE <math.h>
#Include <stdio.h>
#Include <stdlib.h>
#Include <float.h>
void main(){
float aux1;
year_aux=2011.0;
aux1= (365.25*(year_aux+2000.0));
printf("\r aux1:%8.3f",aux1);
}
|
When I simulate the answer was:
1465017.75 (calculator)
1465016.728 (response algorithm)
Why occurs this error, if i try with float numbers, and the representation are until 3.4*10^38 ?
Moreover, but when I used a floor function
1465017 (obviously)
1465016.960 (algorithm response, error)
Why???
2.- If I try to operate with a fixed point ex.
int32 _fixed(4) aux1, could be operate "fixed point" and float numbers simultaneously?? Operate with +,-,/,floor(), sin(), cos(),etc..
3.- When I can learn about float point operation with ccs??
Thanks for your time
RVR
I'm using a Pic18F2550, simulator proteus 7.6, PCWH 4.1. |
|
![](templates/subSilver/images/spacer.gif) |
Ttelmah
Joined: 11 Mar 2010 Posts: 19382
|
|
Posted: Sat Jan 28, 2012 4:42 pm |
|
|
Key thing, is that there _isn't_ a 'double' type. Double _only_ exists for things like DSP chips. In standard CCS C, the keyword is for 'compatibility only'. So you have a 6.5digit effective _single precision_ float only.
'Why occurs this error'. Simple you are trying to use precision that is not there in the available maths....
Best Wishes |
|
![](templates/subSilver/images/spacer.gif) |
RVR
Joined: 04 Jan 2012 Posts: 18
|
|
Posted: Sat Jan 28, 2012 5:39 pm |
|
|
Ttelmah wrote: | Key thing, is that there _isn't_ a 'double' type. Double _only_ exists for things like DSP chips. In standard CCS C, the keyword is for 'compatibility only'. So you have a 6.5digit effective _single precision_ float only.
'Why occurs this error'. Simple you are trying to use precision that is not there in the available maths....
Best Wishes |
Hi Ttelmah,
I had some suspicions that the magnitude of the number was the error source (a few digits numbers had been tested ) but did not know the amount of numbers (precision) that supported the compiler (6.5), and so appreciate your help.
But i need to work at these conditions, even if it losing precision.
Any idea?
For example fixing a decimals at three (for ex) but i don't know if it is possible using a fixed point and if it works with the math.c library? |
|
![](templates/subSilver/images/spacer.gif) |
Douglas Kennedy
Joined: 07 Sep 2003 Posts: 755 Location: Florida
|
|
Posted: Sun Jan 29, 2012 1:50 am |
|
|
Quote: | I'm using a Pic18F2550, simulator proteus 7.6, PCWH 4.1. |
There is a lot that could happen here between a proteus simulation and calculations using real PIC registers. We all learned math using base ten notation and most calculators use BCD so they also use base ten notation. This makes us accept only decimal notation as accurate. The PIC uses binary notation accurately.
That means a PIC's accurate binary notation is faithfully translatable to decimal only for power series of 2 up to the bit precision used. It is not that the floating point binary gets an erroneous result ( it is accurate in binary notation ) its just that an accurate decimal translation is impossible. We encounter this all the time in decimal Ex. 1/3 0.3333... can't be accurately notated in decimal yet 1/2 0.5 can.
For the PIC translational issues may occur on the way into binary and on the way back to decimal. The simple solution is to stay with power series of 2 so use fixed point arithmetic. Otherwise use float and accept the accurate base 2 calculation and the translation error to decimal that shows in the last significant digits. Now the PIC is capable of doing binary floating point to any precision assuming there is sufficient RAM to hold the results. Since binary float math is CPU intensive the CCS compiler only does this for a 23 bit mantissa and an 8 bit exponent which in decimal means 6 or 7 significant digits. A PC might do 64 bit precision binary still with translation to decimal issues but only in numbers with many decimal digits of significance. |
|
![](templates/subSilver/images/spacer.gif) |
Ttelmah
Joined: 11 Mar 2010 Posts: 19382
|
|
Posted: Sun Jan 29, 2012 3:04 am |
|
|
float.h, tells you all the precision and range limits for the numeric types.
Double is shown here, because it is available in PCD. If you look at the PCH manual, you get:
"double
Is a reserved word but is not a supported data type."
The PCD manual, offers you float, float48, and float64 types.
Best Wishes |
|
![](templates/subSilver/images/spacer.gif) |
|
|
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
|