[Ffmpeg-devel] Re: [PATCH] Machine endian bytestream functions

Måns Rullgård mans
Sat Apr 14 13:28:35 CEST 2007

matthieu castet <castet.matthieu at free.fr> writes:

> Hi,
> M?ns Rullg?rd wrote:
>>> +    *pdata = 0x01234567;
>>> +    return !(*pdata == 0x01234567);
>>> +EOF
>> This entire check is actually rather pointless.  On many systems that
>> do not support unaligned accesses, the OS will by default fix it up in
>> a trap handler.  There is no way to detect this, but it will run very
>> slowly.  Furthermore, even on x86 aligned accesses are faster than
>> unaligned.
> Is there arches where unaligned access are faster than aligned one ?

I very much doubt it, and I've certainly never seen one.  Think about
why unaligned accesses are usually impossible or slow, and it should
become clear why this is unlikely.

> BTW it's nice to have all these configure checks, but they can't be
> used with cross compilation and may be give wrong result if the host
> and target aren't the same machine.

Good point.

M?ns Rullg?rd
mans at mansr.com

More information about the ffmpeg-devel mailing list