[Ffmpeg-devel] [PATCH] fix for libavcodec/armv4l/mpegvideo_armv5te.c

[Siarhei Siamashka]

On Monday 29 January 2007 15:43, you wrote:

> On 1/29/07, Dmitry Antipov <dmantipov at yandex.ru> wrote:
>
> > > Thanks for reporting this. Surely add and sub instructions on ARM have
> > > 3 operands, but both gcc 3.4.4 and gcc 4.1.1 that I have tested,
> > > accepted this code without any problems (maybe it is some kind of
> > > syntax sugar?). Did this code break on your compiler? What version of
> > > gcc do you have?
> >
> > I'm using (cross) gcc 3.4.3. Here is a failure report with gcc 4.0.2:
> > http://lists.mplayerhq.hu/pipermail/mplayer-users/2007-January/065126.htm
> >l (note this user probably has poorly configured compiler, btw.)
>
> Thanks a lot. I'll try to experiment with my compiler today in the
> evening and will try to make it behave more strict in order to detect
> and avoid such problems in the future.

Investigated this issue a bit. Looks like it is binutils version that matters 
(as gnu assembler is part of this package). I have binutils 2.16.91 used 
with gcc 2.4.4 (CodeSourcery ARM 2005q3-2) and also binutils 2.17 used 
with gcc 4.1.1 (gentoo crossdev toolchain). I did not succeed in trying to
find any option that would issue a warning about this nonstandard two 
operand add instruction mnemonic :-(

According to [1], binutils 2.16.92 is the minimal version that supports EABI,
so installing anything older is not an option. So I guess the only way to
avoid such problems in the future is just be more carefull and rely on
bugreports such as yours.

Dmitry, which version of binutils do you have installed by the way? And just
for statistics, what kind of target system uses binaries buit with it?

[1] http://wiki.debian.org/ArmEabiPort