[Ffmpeg-cvslog] CVS: ffmpeg/libavcodec bitstream.h,1.151,1.152

[Måns Rullgård]

Michael Niedermayer <michaelni at gmx.at> writes:

> Hi
>
> On Thursday 12 May 2005 22:37, M?ns Rullg?rd wrote:
>> Michael Niedermayer <michaelni at gmx.at> writes:
>> > Hi
>> >
>> > On Thursday 12 May 2005 21:11, M?ns Rullg?rd wrote:
>> >> michael at mplayerhq.hu (Michael Niedermayer CVS) writes:
>> >> > -#ifdef ARCH_SPARC
>> >> > +#if defined(ARCH_SPARC) || defined(ARCH_ARMV4L)
>> >> >  #define UNALIGNED_STORES_ARE_BAD
>> >> >  #endif
>> >>
>> >> Why doesn't this bite other non-x86 CPUs?
>> >
>> > because they either can do unaligned accesses directly or the kernel
>> > catches and emulates them, AFAIK in the ARM case this isnt done by
>> > default
>>
>> On Alpha, you get a message in the kernel log if there are unaligned
>> accesses.  Running ffmpeg, I don't get any.  
>
> well, maybe you are running it with the wrong parameters, look at
> the original bugreport if not then maybe the whole
> UNALIGNED_STORES_ARE_BAD thing can be removed

Ah, indeed.  When writing JPEG images I do get such reports.  I think
we should be setting UNALIGNED_STORES_ARE_BAD #ifndef ARCH_X86
instead.

>> On SPARC/Solaris unaligned accesses cause a SIGBUS and core dump.
>
> on ARM you get pseudo random data silently

That's not nice at all.  How does powerpc behave?

-- 
M?ns Rullg?rd
mru at inprovide.com