[PIC] MikroC > Float to string routine …

[Eric_O]

Here under the beginning of my routine that runs …

void float2ascii_v3_2 (float f, unsigned char *buf, char decimal) // Converts a floating point number to an ASCII string.
{ //
unsigned int remainder;
unsigned char *s, length; // unsigned char *s is the pointer to the string
long i;
char j;

length = 0;
j = 0;

i = (long)((float)f * pow(10, decimal)); // float type is from 3,4 x 10 -38 to 3,4 x 10 38

…

But …

When I run the routine step by step, with MikroC Debbuger in order to monitor variables and checking loops conditions … when decimal = 3, f = 1234,567, the Debbuger shows f = 1,234567EXP003 (ok) and i = 1234565, instead of 1234567.

Why ?

Merci.

Éric

 

[KlausST]

Hi

Most probably what you see is the resolution (problem) of a float.
You say your

float type is from 3,4 x 10 -38 to 3,4 x 10 38

Is this according IEEE standard?
--> Read the wikipedia article about float numbers.

i = 1234565, instead of 1234567.

1234567 is a 21 bit binary integer number. Missing 2 LSB values let's me guess your float value uses just 19 significand bits.

Klaus

Added: Maybe the missing 2 LSB values are (additionally) caused by calculation errors.

 

[FvM]

Did you check against sprintf(), the standard method for float to string conversion?

Float function pow() might cause the loss of accuracy, did you check?

 

[Eric_O]

Did you check against sprintf(), the standard method for float to string conversion?

Float function pow() might cause the loss of accuracy, did you check?

Thanks for your advice. Writing a float to string routine is for my understanding and practicing pointers for a LCD voltmeter, thermometer project or something else …
Presently my 16F877 and 887 can not use sprintf. I will soon switch to a 18F …
Probably pow() function is responsible.
Because before, in the previous version of my routine I didn’t use « decimal » parameter and I wrote 100 instead of pow(10, decimal), and fixed at 2 decimals. I will test with 1000.

 

[KlausST]

Hi,

generally I avoid float with small microcontrollers at all. (whenever possible. And it was possible almost every time)

16 bit integer (fixed point) should be suitable for 4 digits.

Klaus