[FFmpeg-devel] Compiling with Sun cc

Roman V. Shaposhnik rvs
Sun Oct 12 04:44:28 CEST 2008


Sorry for the long delay...

On Mon, 2008-10-06 at 08:41 +0200, Reimar D?ffinger wrote:
> On Sun, Oct 05, 2008 at 09:31:56PM -0700, Roman V. Shaposhnik wrote:
> > On Fri, 2008-10-03 at 09:24 +0200, Reimar D?ffinger wrote:
> > > On Thu, Oct 02, 2008 at 08:10:22PM -0700, Roman V. Shaposhnik wrote:
> > > > On Thu, 2008-10-02 at 11:09 +0200, Diego 'Flameeyes' Petten? wrote:
> > > > > > "libavcodec/mpegvideo_enc.c", line 3288: warning: can not set non-default
> > > > > > alignment for automatic variable
> > > > > 
> > > > > This basically means it does not support alignment for stack variable,
> > > > > which is probably a very bad thing, this is actually where I stopped and
> > > > > waited before continuing...
> > > > 
> > > > Well, it is very controversial thing, for one. The trouble is, that even
> > > > gcc can not really guarantee it, since initial stack assumptions are not
> > > > controlled by the compiler, but rather by the combination of an OS
> > > > and how linking was done.
> > > 
> > > How about 64 bit compilation? Does it work there?
> > 
> > Well, section 3.2.2 of the amd64 ABI guarantees the stack alignment 
> > of 16 bytes.
> 
> My question was not about the ABI, I know the basics of that. I do not
> know how the Sun C compiler behaves though.
> The Sun C compiler message is "arning: can not set non-default
> alignment for automatic variable", which seems to indicate that it just
> ignores the alignment, which would mean the issue still would be the
> same for amd64.

You're right currently even on x64 arc we don't provide alignment for
the stack variables.

> And a third one would be that the message actually should be "assuming
> sufficiently aligned stack for non-default alignment for automatic
> variable", which IMO is the most reasonable way to handle it by default.

That's pretty much the behavior that we would like to get eventually. It
is not there yet, but I'll poke our backend guys to see what's the 
holdup.

Thanks,
Roman.





More information about the ffmpeg-devel mailing list