Thread: Float and double size

I was reading K&R, in section 2.2 they mentioned that

float is single precision floating point, and
double is double precision floating point

When I checked wiki, it says that "IEEE single-precision floating point computer numbering format, is a binary computing format that occupies 4 bytes (32 bits) in computer memory."

But in the second last paragraph of section 2.2 in K&R, it says that size of float,double and long double is also implementation defined.

How can that be possible because wiki says it always take 4 byte for single precision or 8 bytes for double precision.

Please help

They mention that just in case some hardware vendors don't conform to the IEEE standards.

The C standard and IEEE 754 are two totally different things. The C type "float" does not mean "IEEE 754 single-precision", nor C "double" mean "IEEE 754 double-precision", though the names do suggest a correlation. The C standard does not say that implementations must use the IEEE 754 floating point format. They generally do, but it is not a requirement. Thus, the C standard describes floating point types for C, regardless of the underlying implementation.

Actually K&R says that float has single-precision, what I want to know is that whether single-precision is attached with IEEE or not, AFAIK single precision term always comes with IEEE.

Please clarify.

Originally Posted by _arjun

Actually K&R says that float has single-precision, what I want to know is that whether single-precision is attached with IEEE or not, AFAIK single precision term always comes with IEEE.

No, it is not attached, as I said before. If it were, K&R would say something like "A single-precision 'float' type in C must conform to the IEEE 754 standard for single-precision floating point." But it doesn't say that because they are not connected. The single/double precision in C just refers to relative size. The C99 standard (6.2.5p10) states that the set of values of type float is a subset of the set of values of type double, meaning that a double can represent every value a float can and possibly more. Similarly, a long double must hold everything a double can, and possibly more. It also says (Annex F) that an implementation may define __STDC_IEC_559__, and if it does, it must conform to IEC 60559, which is basically IEEE 754, but nothing about how implementations that do not define __STDC_IEC_559__ must represent floats.

Originally Posted by anduril462

No, it is not attached, as I said before. If it were, K&R would say something like "A single-precision 'float' type in C must conform to the IEEE 754 standard for single-precision floating point." But it doesn't say that because they are not connected. The single/double precision in C just refers to relative size. The C99 standard (6.2.5p10) states that the set of values of type float is a subset of the set of values of type double, meaning that a double can represent every value a float can and possibly more. Similarly, a long double must hold everything a double can, and possibly more. It also says (Annex F) that an implementation may define __STDC_IEC_559__, and if it does, it must conform to IEC 60559, which is basically IEEE 754, but nothing about how implementations that do not define __STDC_IEC_559__ must represent floats.

Thank you so much for clarifying my doubt