[IRCServices Coding] GCC3

[Trevor Talbot]

On Monday, February 25, 2002, at 01:04  PM, Kelmar K. Firesun wrote:

> ----- Original Message -----
> From: "Andrew Church" <achurch at achurch.org>
>
>>>>      The major problem I have with GCC 3.0 is that it reorders 
>>>> structures
>>>> (or at least did at one point), which would break convert-db, and in
>>>> general is a Bad Idea (among other things it prevents you from laying
>>>> structures on top of data in memory).  Does anyone know if this has 
>>>> been
>>>> fixed or if there's a way around it?
>>>
>>> Are you sure it reordered them? It might just have changed the padding
>>> between members, which compilers have been known to do in the past.
>>> Anyhow, I haven't tried convert-db yet but I will when I get the 
>>> chance
>>
>>      To be perfectly honest, that's based on hearsay--I haven't 
>> confirmed
>> it one way or the other.  But I do recall quite a lot of discussion on 
>> that
>> point, so I'd like to have confirmation that it does work before
>> "officially" endorsing it.

Reordering is not permitted by the ANSI/ISO C standards.

> I could certanly see it padding the data structures.
> This would be to optimize memory access by keeping things
> aligned with the CPU's WORD size.  64bits in the case of
> most Intel Pentiums if I recall correctly.

Usually 32 bits, possible exceptions where optimized structures
occur (MMX usage, for example).

> For example, if you have a structure like so:
>
> struct foo
> {
>     int_16 bar;
>     int_8 baz;
> };
>
> The compiler would append or prepend (depending on compiler)
> on an extra 40bits of data to align it in memory on each
> allocation.

Under 32bit x86, it's likely to add 8 bits of padding to the end
of the structure.  The alignment is for the width of the largest
datatype.

> This is a very common optimization for speed in many compilers.
> I'm actually very surprised that GCC would just now be
> implementing it.

Not just for speed; some platforms require aligned types
(such as Motorola 68k and PPC under certain conditions).
It's also well-documented by the C standards.

Partly for this reason, mapping structs onto arbitrary data in
memory results in undefined behavior.

-- Quension