[FFmpeg-devel] [PATCH] make sure initialization happens even after goto fires

[Reimar Döffinger]

Hello,
On Thu, Jul 17, 2008 at 07:55:57PM +0100, M?ns Rullg?rd wrote:
> > --- libavcodec/h264_parser.c	(revision 14263)
> > +++ libavcodec/h264_parser.c	(working copy)
> > @@ -59,10 +59,7 @@
> >              if(v==7 || v==8 || v==9){
> >                  if(pc->frame_start_found){
> >                      i++;
> > -found:
> > -                    pc->state=7;
> > -                    pc->frame_start_found= 0;
> > -                    return i-(state&5);
> > +                    goto found;
> >                  }
> >              }else if(v==1 || v==2 || v==5){
> >                  if(pc->frame_start_found){
> > @@ -80,6 +77,11 @@
> >      }
> >      pc->state= state;
> >      return END_NOT_FOUND;
> > +
> > +found:
> > +    pc->state=7;
> > +    pc->frame_start_found= 0;
> > +    return i-(state&5);
> >  }
> 
> The question is, is this as fast?  I agree it's nicer structurally.

Since gcc 4.3.1 generates exactly the same code, for this case the
answer is a clear yes.
Actually I would find the current code to be more likely for the
compiler to mess up and do something stupid like doing the stack
allocation for v upon the second goto (though with that statement I
probably only show my ignorance about writing compilers ;-) ).
Either way I my real intention was not to propose this as a real patch
(though I will apply it if asked), but to show how to "fix" this
properly if one was hell-bent to do it. I started to try to explain it
and thought a patch would be much simpler to understand ;-)
Note: tests and benchmarks when done by me as always have a reliability
close to 0, so better test yourselves if you want to be sure.

Greetings,
Reimar D?ffinger