Double Precision Floating Point Accuracy Problem

desouza said:

I then checked the actual memory value on one of the many IEEE-754 data converters around on the internet and they agree with CCS: the 64-bit value in memory (0x40DF782760000000) translates to 32224.615234375.

Agree with the result of the IEEE-754 data converter. However, the hex value had the last 7 hex digits as all zeros, which suggests there might be some truncation occurring somewhere.

If I feed the original posters value of 32224.61593030503 into a IEEE-754 data converter the 64-bit double hex value reported in 40DF78276B66F12E which has the first 9 hex digits matching your value, but non-zero values in the last 7 hex digits.

Therefore, is the error that the  a = 32224.61593030503  assignment is loosing some precision?

Hamza Ahmed24 said:

The Problem is when i declare variable "a" of type long double and store some value in it a = 32224.61593030503 in it.

The problem is that the unsuffixed literal constant 32224.61593030503 is implicitly double precision.

A "L" suffix needs to be added to the constant to make the constant explicitly long double precision.

e.g. with the following program on a TMS320F28335:

#include <stdio.h>

long double  a = 32224.61593030503L;
long double  b = 32224.61593030503;

int main(void) {
    printf ("sizeof(a)=%lu  sizeof(b)=%lu\n", sizeof(a), sizeof(b));
    printf ("a=%.16Lg\n", a);
    printf ("b=%.16Lg\n", b);
	
	return 0;
}

The CCS 6.1 debugger watch window reports the value of a as "32224.61593030503", and the value of b as "32224.615234375".