[FFmpeg-devel] [PATCH] Use posix_memalign() instead of memalign()

[Måns Rullgård]

"Ramiro Polla" <ramiro.polla at gmail.com> writes:

> On Tue, Oct 7, 2008 at 4:24 AM, M?ns Rullg?rd <mans at mansr.com> wrote:
>> "Ramiro Polla" <ramiro.polla at gmail.com> writes:
>>
>>> Hi,
>>>
>>> On Sun, Sep 28, 2008 at 10:17 AM, M?ns Rullg?rd <mans at mansr.com> wrote:
>>>> Diego 'Flameeyes' Petten? <flameeyes at gmail.com> writes:
>>>>
>>>>> While on GLIBC the memalign() function is declared in malloc.h, on
>>>>> OpenSolaris its definiton is in stdlib.h; include the file so that an
>>>>> implicit declaration can be avoided.
>>>>> ---
>>>>>
>>>>>  libavutil/mem.c |    1 +
>>>>>  1 files changed, 1 insertions(+), 0 deletions(-)
>>>>>
>>>>> diff --git a/libavutil/mem.c b/libavutil/mem.c
>>>>> index 960074c..da75638 100644
>>>>> --- a/libavutil/mem.c
>>>>> +++ b/libavutil/mem.c
>>>>> @@ -34,6 +34,7 @@
>>>>>  #ifdef HAVE_MALLOC_H
>>>>>  #include <malloc.h>
>>>>>  #endif
>>>>> +#include <stdlib.h>
>>>>>
>>>>>  /* you can redefine av_malloc and av_free in your project to use your
>>>>>     memory allocator. You do not need to suppress this file because the
>>>>
>>>> #including stdlib.h obviously can't do any harm, it being a standard
>>>> header.
>>>
>>>> There is, however, a deeper issue here: memalign() is not
>>>> defined by any standard, and some systems that have it provide no
>>>> means to free such allocations.  Is there any reason we can't use the
>>>> standard posix_memalign() instead?
>>>
>>> Attached patch should spark some discussion...
>>>
>>
>> [...]
>>
>>> Index: libavutil/mem.c
>>> ===================================================================
>>> --- libavutil/mem.c   (revision 15575)
>>> +++ libavutil/mem.c   (working copy)
>>> @@ -35,6 +35,10 @@
>>>  #include <malloc.h>
>>>  #endif
>>
>> #include <malloc.h> wouldn't be needed with posix_memalign(), but see
>> below.
>
> Which systems use malloc.h?

BSD systems predating posix_memalign().  I wouldn't be surprised if
OpenBSD is still missing it.

>>> +#ifdef HAVE_POSIX_MEMALIGN
>>> +#include <stdlib.h>
>>> +#endif
>>
>> It is always save to #include <stdlib.h>
>
> Now I see that the stdlib.h include comes from common.h.
>
> And another thing, I remember there was some discussion regarding
> whether we should include only the necessary files or whether it was
> ok to just include common.h everywhere, but I forgot the outcome.

I don't think there ever was a consensus.  IMHO, it is good practise
to #include standard headers where needed, even if common.h includes
them as well.  Things that are conditionally included in
common.h/internal.h are a different matter, since duplicating checks
all over the code is not pretty.

>>>  /* you can redefine av_malloc and av_free in your project to use your
>>>     memory allocator. You do not need to suppress this file because the
>>>     linker will do it automatically */
>>> @@ -57,8 +61,9 @@
>>>      diff= ((-(long)ptr - 1)&15) + 1;
>>>      ptr = (char*)ptr + diff;
>>>      ((char*)ptr)[-1]= diff;
>>> -#elif defined (HAVE_MEMALIGN)
>>> -    ptr = memalign(16,size);
>>> +#elif defined (HAVE_POSIX_MEMALIGN)
>>> +    if(posix_memalign(&ptr,16,size))
>>> +        ptr = NULL;
>>
>> It might make sense to use memalign() as a fall-back for old BSD
>> systems and the like without posix_memalign().
>
> Would that be a manually entered list like is done for need_memalign?

No, it would be tested by configure.

-- 
M?ns Rullg?rd
mans at mansr.com