On Tue, Aug 16, 2011 at 2:41 PM, Ronald S. Bultje <rsbul...@gmail.com> wrote:
> Hi,
>
> On Tue, Aug 16, 2011 at 8:05 AM, Diego Biurrun <di...@biurrun.de> wrote:
>> From: Jeff Downs <heydo...@somuchpressure.net>
>>
>> Correct computation of implicit weight tables when referencing pictures
>> that are marked for long reference.
>>
>> Signed-off-by: Diego Biurrun <di...@biurrun.de>
>> ---
>> libavcodec/h264.c | 20 +++++++++++---------
>> 1 files changed, 11 insertions(+), 9 deletions(-)
>>
>> diff --git a/libavcodec/h264.c b/libavcodec/h264.c
>> index a9a6c8c..5d42ed3 100644
>> --- a/libavcodec/h264.c
>> +++ b/libavcodec/h264.c
>> @@ -2174,15 +2174,17 @@ static void implicit_weight_table(H264Context *h,
>> int field){
>> for(ref0=ref_start; ref0 < ref_count0; ref0++){
>> int poc0 = h->ref_list[0][ref0].poc;
>> for(ref1=ref_start; ref1 < ref_count1; ref1++){
>> - int poc1 = h->ref_list[1][ref1].poc;
>> - int td = av_clip(poc1 - poc0, -128, 127);
>> - int w= 32;
>> - if(td){
>> - int tb = av_clip(cur_poc - poc0, -128, 127);
>> - int tx = (16384 + (FFABS(td) >> 1)) / td;
>> - int dist_scale_factor = (tb*tx + 32) >> 8;
>> - if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
>> - w = 64 - dist_scale_factor;
>> + int w = 32;
>> + if (!h->ref_list[0][ref0].long_ref &&
>> !h->ref_list[1][ref1].long_ref) {
>> + int poc1 = h->ref_list[1][ref1].poc;
>> + int td = av_clip(poc1 - poc0, -128, 127);
>> + if(td){
>> + int tb = av_clip(cur_poc - poc0, -128, 127);
>> + int tx = (16384 + (FFABS(td) >> 1)) / td;
>> + int dist_scale_factor = (tb*tx + 32) >> 8;
>> + if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
>> + w = 64 - dist_scale_factor;
>> + }
>
> Which sample/bug does this fix? I really have no idea if this is
> correct without spending some significant time on the spec right
> now... Jason, Loren, do you guys know?
if( td == 0 /* || pic0 is a long-term ref */ )
dist_scale_factor = 256;
Going by x264 code, this seems correct.
Jason
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
