[MPlayer-dev-eng] [PATCH] memalign hack should use av_malloc

[Hervé W.]

On 23/07/06, Gianluigi Tiesi wrote:
> On Sun, Jul 23, 2006 at 02:43:56AM +0200, Hervé W. wrote:
[...]
> > If memalign gets defined as av_malloc, doesn't free have to become
> > av_free (or av_freep) ? I think I saw some lines in liba52/parse.c and
> > in libswscale/swscale.c where "memaligned" memory gets free'd and I
> > think there are more locations.
> > And if free becomes av_free, all the rest of the real malloc, calloc
> > and realloc call should become av_*alloc. It has the added benifit
> > that enabling SSE in MinGW builds shouldn't cause any problems
> > anymore.
> >
> I'm making sse enabled mingw builds and I had no negative
> feedbacks about, I'll ask to add mingw to supported sse arch
> if I will not get negative responses from users.

Fine. No problem. That wasn't the main thing I was trying to say, though.

av_malloc, with MEMALIGN_HACK defined, returns ptr2 for memory like:
  __________________________
 |              |       |                       |
 |              | diff  |   data              |
 |________|____|_____________|
 |                      |
ptr1                 ptr2

and some "memaligned" memory gets free'd (not av_free'd or
av_freep'd). Would calling free on anything but ptr1 work? and if it
does, is that behaviour we should rely upon?

-V