Thread: C++ and memory

I need help understanding a little better about certain aspects of memory management.

1. Why isn't bool 1 bit? (possibly related with 2.)

2. All types I sizeof() seem to be 4 bytes or larger. 32 bits seem to be the minimum. Is this what defines 32 bit machines?

3. Simple classes with just data members don't reflect on memory the sum of their data members sizes. Is there an overhead to a class type?

4. If so, is this a known fixed value? My experiments seem to show a variable overhead.

5. What is the exact formula to calculate the significant digits on integral and floating point numbers?

first all sizes of types (bool, chars, ints, etc) are platform dependant. the standard doesn't actually specify the exact sizes anywhere, merely the relativ sizes (i.e. sizeof(char) <= sizeof(short) <= sizeof(int) and I'm not even sure that's it <= and not < )
All my answers refer to the most common case of win32.

1. because it's more computationally expensive to access a single bit than a byte. In the standard library a vector<bool> is actually a specialisation that stores each bool as a bit. whether this is a good thing is a matter of some debate.

2. sizeof(char) on win32 should return 1. sizeof(short) should return 2. can you post code?

3 and 4. C++ allows structs and classes to be aligned on a certain boundaries. this is usually the sizeof the largest pod. so

Code:

struct X
{
   int a;
   char b;
};

will be 8 bytes on win32. You can override this in the compiler settings or by using #pragma pack. Also you cannot instantiate a class of zero size. so

Code:

class empty
{};

will be at least 1 byte. This doesn't apply to base classes (often known as the "empty base class optimisation")

5. can't remember off the top of my head.

This is definitely one of my least known areas in computation.

You are right ChaosEngine. My bad. Char and Short are 1 and 2 bytes respectively. I didn't think testing these.

As for classes... I found a pattern after you mentioned they are aligned around a boundary. They seem to allways be multiples of 4 bytes. That helped also, thanks.

have a look in project properties->c/C++->code generation->struct member alignment
that lets you change the alignment (that's for vs2003 but I think it's the same for 2005)

2. sizeof(char) on win32 should return 1.

sizeof(char) is always 1, as is sizeof(unsigned char) and of course sizeof(signed char).

the standard doesn't actually specify the exact sizes anywhere, merely the relativ sizes (i.e. sizeof(char) <= sizeof(short) <= sizeof(int) and I'm not even sure that's it <= and not < )

Because zen is to discover that it is about ranges of values, not about sizes of objects containing a range of values.

Ok. Thanks. I'm doing some research on this and now it's starting to make sense to me.