Re: [xine-devel] Mandrake patches including 64-bit support

[Miguel F. <mi...@ce...>]

Hi Gwenole,

On Wed, 2004-02-18 at 09:35, Gwenole Beauchesne wrote:
> On Wed, 18 Feb 2004, Goetz Waschk wrote:
> 
> > Am Di 17 Feb 2004 23:30:27 GMT schrieb Miguel Freitas:
> > > >   http://cvs.mandrakesoft.com/cgi-bin/cvsweb.cgi/SPECS/xine-lib
> > > i'm no configuration/compiler expert but they look very good to me.
> > > i would take a little care with the 64bit fixes patch though. 
> >  
> > > i'm not sure the goom part is correct, using uintptr_t (is it the same
> > > as unsigned int *?) might give a different pointer math that old (wrong)
> > > code that cast to (unsigned int). shouldn't (char *) be used instead?
> 
> uintptr_t is an unsigned integer large enough to hold all bits of a
> pointer. On 64-bit systems, casting the pointer to (unsigned int) will
> result in trashing the higher word. Besides, ~2047 will be a 32-bit value,
> thus the cast to uintptr_t so that we get the desired bits in the upper
> part.

fair enough. so my assumption that uintptr_t == (unsigned int *) was
wrong, and your fix is correct ;-)


> > > and while i agree with the change to alphablend.c, why is the old code
> > > wrong?
> 
> Well, iirc, that one could cause aliasing problems. But I have not come up 
> with real world problems as I don't actually know when it is used. Now 
> that AMD64 hardware is common, people could try it too. ;-)

well, you can force using this code by playing a dvd with
subtitles/menus and enabling tvtime. make sure tvtime's cheap_mode=0.
anyway it looks cleaner with this patch applied.

regards,

Miguel

ps: Rocky, no objections for committing Mandrake's patches. we should
submit the goom part back to maintainers though...