 |
 |
| View previous topic :: View next topic |
| Author |
Message |
picit
Guest
|
| Read/write float to ext eeprom |
 Posted: Thu Feb 11, 2010 5:08 am |
 |
|
Hi, I have a problem with read/write floats to an external ee-prom.
It seems that it will only work up to a "signed 16-bit float", or 32767.0
I'm using CCS ver 4.068, a PIC18F2525.
An M95256 eeprom, and bitbanging it. Everything else is working ok, 8-bits, 16-bits but not floats!
| Code: | void write_ext_eeprom(EEPROM_ADDRESS address, BYTE data) {
BYTE cmd[4];
BYTE i;
BYTE wren;
wren=0x06;
cmd[0]=data;
cmd[1]=address;
cmd[2]=(address>>8);
cmd[3]=0x02;
// Wait until the eeprom is done with a previous write
while(!ext_eeprom_ready());
oMEM_ON();
for(i=0; i<8; ++i)
{
output_bit(EEPROM_DI, shift_left(&wren,1,0));
output_high(EEPROM_CLK);
output_low(EEPROM_CLK);
}
oMEM_OFF();
oMEM_ON();
for(i=0; i<32; ++i)
{
output_bit(EEPROM_DI, shift_left(cmd,4,0));
output_high(EEPROM_CLK);
output_low(EEPROM_CLK);
}
oMEM_OFF();
}
BYTE read_ext_eeprom(EEPROM_ADDRESS address) {
BYTE cmd[3];
BYTE i,data;
cmd[0]=address;
cmd[1]=(address>>8);
cmd[2]=0x03;
// Wait until the eeprom is done with a previous write
while(!ext_eeprom_ready());
oMEM_ON();
for(i=0; i<24; ++i)
{
output_bit(EEPROM_DI, shift_left(cmd,3,0));
output_high(EEPROM_CLK);
output_low(EEPROM_CLK);
}
for(i=0; i<8; ++i)
{
shift_left(&data,1,input(EEPROM_DO));
output_high(EEPROM_CLK);
output_low(EEPROM_CLK);
}
oMEM_OFF();
return(data);
}
////////////////////////////////////////////////////////////////////////
void WRITE_FLOAT_EXT_EEPROM(long int n, float data) {
int i;
for (i = 0; i < 4; i++)
write_ext_eeprom(i + n, *((int8 *)(&data) + i) );
}
float READ_FLOAT_EXT_EEPROM(long int n) {
int i;
float data;
for (i = 0; i < 4; i++)
*((int8 *)(&data) + i) = read_ext_eeprom(i + n);
return(data);
}
//////////////////////////////////////////////////////////////////////////////
void WRITE_LONG_EXT_EEPROM(long n, long data)
{
int i;
for (i = 0; i < 2; i++)
write_ext_eeprom(i + n, *((int8 *)(&data) + i) );
} |
|
|
 |
PCM programmer
Joined: 06 Sep 2003
Posts: 21708
|
|
| Posted: Thu Feb 11, 2010 12:37 pm |
|
|
Post the test program that calls these routines. This program should
write a floating point value to the eeprom, and then read it back, and
display it in a terminal window on a PC. Post the floating point value
that you see in the terminal window.
The test program should be very short, and it must compile without
errors. (Verify that before you post it).
Also, is this test being done in real hardware, or is this a Proteus project ? |
|
|
picit
Guest
|
|
| Posted: Tue Feb 16, 2010 6:08 am |
|
|
just found out that 32767.001 dont work but 32767.01 do work!
So it seems that CCS floats only can handle 8 "signs" long floats??
It isnt in the eeprom, its in the program itself.
I managed to get around the problem by using less decimals..
Altough I cant understand why its so.. |
|
|
Ttelmah
Guest
|
|
| Posted: Tue Feb 16, 2010 6:31 am |
|
|
Because of the size of a 'float'.
The single precision 'float' in a PC, or CCS, uses just four bytes. An 8 bit 'exponent' saying where the 'binary point' is. A sign bit, for the mantissa, and a 23bit mantissa. The mantissa, is treated as always having a leading '1', giving it effectively 24bits.
This gives just under 7 digits of useable decimal precision. Anyhing beyond this is 'luck'....
This is always specified for you, in 'float.h' on any C, which specifies the guaranteed precision (only 6 digits!....), and the sizes of the mantissa etc...
Switching to PCD, offers you a double precision 'foat' with 15digits of useable precision.
The 'overhead' of using float, is always that several bits of the binary number are 'lost' to the precision, to handle the exponent.
This is why scaled integers are 'better' for anything working in a limited part of the number line....
Best Wishes |
|
|
FvM
Joined: 27 Aug 2008
Posts: 2337
Location: Germany
|
|
| Posted: Tue Feb 16, 2010 6:36 am |
|
|
Unfortunately, you didn't give a complete example code, so we can only guess, what's exactly not working in
your code. If you only mean, that 32767.001 isn't exactly reproduced, Ttelmah has given the complete answer.
If you observe other problems than a simple rounding error, you should clarify. |
|
|
picit
Guest
|
|
| Posted: Tue Feb 16, 2010 9:46 am |
|
|
Thanks Ttelmah, exactly the thing.
Its only possible to have a float number up to 32767.000
unless decreasing the precision. I have never thought about it earlier.
Thanks again. |
|
|
|
|
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
|
FAQ 
Search
Register
Profile 
Log in to check your private messages 
Log in
