[cxx-abi-dev] Question about tail padding

[Nathan Sidwell]

"Nelson, Clark" wrote:

> The problem Scott pointed out was about packing bit-fields from derived and
> base classes into the same byte. Intel's compiler doesn't do that. We
> instead wind up with the bit-field in the derived class in a poorly-aligned
> container, which is a smaller performance problem.
ah, what you're saying is that A::a is bits 0& 1 of byte 4,
and B::b is bits 0 & 1 of byte 5.

So, the intel compiler doesn't consider that an int bitfield
reserves 4 bytes of an allocation unit, but takes the unused
bytes of that as tail padding?

I've just checked g++ with the attached, it shows that B::b is
allocated in bits 2 & 3 of byte 4.

nathan

-- 
Dr Nathan Sidwell   ::   http://www.codesourcery.com   ::   CodeSourcery LLC
         'But that's a lie.' - 'Yes it is. What's your point?'
nathan at codesourcery.com : http://www.cs.bris.ac.uk/~nathan/ : nathan at acm.org
-------------- next part --------------
#include <stdio.h>
#include <string.h>

struct A 
{
  virtual ~A (){}

  int a : 2;
};

struct B : A
{
  int b : 2;
};
struct C
{
  A a;
  
  int c : 2;
};


int main ()
{
  B b;
  char *ptr = (char *)(void *)&b;
  
  printf ("A %u, B %u C %u\n", sizeof (A), sizeof (B), sizeof (C));
  memset (ptr, 0, sizeof (B));
  b.b = 3;
  
  printf ("byte 4 %x, byte 5 %x\n", ptr[4], ptr[5]);
  
}