That's because resolution of a floating point data format is decreasing while numbers increase (because of finite number of decimal places in mantissa). Simply speaking adding very small number to a very high does not produce acurate result.

Hi
All this problem stems due to fact that we have only 32 bits to represent a fractional number.
In Step 7 these 32 bits are distributed like:
31st bit is Sign, Bit 23 to Bit 30 represents Exponent, Bit 0 to Bit 22 represents Mentissa.

All this is then cubled together to represent a Floating number by using formula (IEEE Standard):
Number = 1.m * 2 to the power of (e - bias)

So all this leads to funny approximations. From step 7 Help:

The floating-point numbers in STEP 7 are accurate to 6 decimal places. You can therefore only specify a maximum of 6 decimal places when entering floating-point constants.

The calculation accuracy of 6 decimal places means, for example, that the addition of number1 + number2 = number1 if number1 is greater than number2 * 10 to the power of y, where y>6:

100 000 000 + 1 = 100 000 000.

Your error started because of method being implemented to represent a floating number as said above.You have mentionedthat your problem started from 134217800.0 which after conversion got converted to 134217792, but everything was perfect 134217600.0. Fact is even if you try to convert 134217601.0, it will be represented as 134217600, so there is a error much before 134217800.0

If try to convert 134217640.0 then it will be converted without any error, again error will come at 134217641.0 !

manmeet

If try to convert 134217640.0 then it will be converted without any error, again error will come at 134217641.0 !

New question published by norimane is split to a separate thread with the subject How to round REAL.

Best regards
Min_Moderator