Re: possible structure layout bug

[Andrew Haley <aph at redhat dot com>]

Stephen Kennedy writes:
 > On Wed, 2004-06-09 at 20:44, Richard Henderson wrote:
 > > On Wed, Jun 09, 2004 at 06:22:24PM +0100, Stephen Kennedy wrote:
 > > > Normally, compilers place members from derived classes starting at
 > > > sizeof(base), but gcc breaks this rule if the class contains a simd type
 > > > and places the derived member in the padding of the base class.
 > > 
 > > Has nothing to do with simd or not.  Has everything to do 
 > > with POD or non-POD type.
 > 
 > Ok, so if I have this correct, POD members padding may not be reused by
 > derived classes but non-POD classes may have their padding reused.
 > 
 > Why then is __m128 not considered a POD type? It is most natural to
 > think of machine native register types as POD.

That's not what he said.  If it's a derived class it's not POD, so the
compiler can pack any way it wants.

Andrew.