Hi,

On Tue, Dec 18, 2012 at 8:20 AM, Janne Grunau <[email protected]> wrote:
> On 2012-12-18 07:30:03 -0800, Ronald S. Bultje wrote:
>> Hi,
>>
>> On Mon, Dec 17, 2012 at 10:55 PM, Janne Grunau <[email protected]> 
>> wrote:
>> > On 2012-12-17 18:24:37 -0800, Ronald S. Bultje wrote:
>> >> Hi,
>> >> On Dec 17, 2012 1:02 PM, "Janne Grunau" <[email protected]> wrote:
>> >> >
>> >> > Compute dist_scale_factor_field only for MBAFF since that is the only
>> >> > case in which it is used.
>> >>
>> >> The patch also increases an array size - why?
>> >
>> > to handle up to 32 references for field pictures. Alternatively both
>> > arrays can probably be merged to the usual 16 + 32 size.
>>
>> Check the code:
>
> omitting the important:
>   +    if (FRAME_MBAFF)
>
>> +            for (i = 0; i < 2 * h->ref_count[0]; i++)
>> +                h->dist_scale_factor_field[field][i^field] =
>> +                    get_scale_factor(h, poc, poc1, i+16);
>> +        }
>> +
>> +    for (i = 0; i < h->ref_count[0]; i++){
>>          h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
>>      }
>>
>> For arrays of size:
>>
>> -    int dist_scale_factor[16];
>> +    int dist_scale_factor[32];
>>      int dist_scale_factor_field[2][32];
>>
>> If i for setting dist_scale_factor_field is [0, ref_count * 2] and
>> that's [0, 32], them ref_count is 16, thus the second loop i [0,
>> ref_count] is [0, 16], thus dist_scale_factor[] only needs to be of
>> size 16.
>
> dist_scale_factor_field is only used for MBAFF. The max reference count
> for MBAFF is 16 (!field_pic_flag). For field_pic_flag maximal reference
> count is 32.
>
> The naming of the variables is unfortunate. In other cases we use arrays
> sizes like ...[16 + 32] to handle normal and field references for MBAFF
> in the same array. That one would be large enough to 32 field pic
> references too.

So, shall we commit this and add a FIXME to that effect so we can do that later?

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to