This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: Alignment issue on x86_64?
John Fine wrote:
> Andrew Haley wrote:
>>
>> If so, it's a gcc bug either way. Both caller and callee were compiled
>> by gcc. But we don't have enough information to determine what the cause
>> of the problem really is.
>>
> Sorry that I'm not taking the time to get a copy of the called code and
> I'm instead speculating.
>
> But, it is perfectly possible for the reported symptom to be a bug in
> the called code, rather than in gcc. Here is the simplest bug the
> called code might have fitting all described symptoms (not saying it has
> this bug, just saying we haven't checked that it doesn't).
>
> It could internally use a reference to the passed CARD32 and then cast
> that (CARD32&) to a (long&) before using the value.
Yes, it could.
> In a 32 bit build, that would always work, because a long is 32 bits.
>
> In a 64 bit build, that would usually work, because most ways that the
> compilation of the calling code might place a 32 bit value in a 64 bit
> location would clear the high bits as an (unnecessary I think) side
> effect. But sometimes it would fail. And always it would be a bug,
> because the C++ standard doesn't let you expect defined behavior from
> such an action.
True. Either way, a test case will tell us. Speculation is pointless;
let's see the code.
Andrew.