h264: Lower max_num_ref_frames
GPU decoding seems to be more picky when it comes to the maximum number of reference frames.
This commit is contained in:
parent
cd016d3cb5
commit
b384129c63
@ -95,7 +95,8 @@ const std::vector<u8>& H264::ComposeFrameHeader(const NvdecCommon::NvdecRegister
|
|||||||
const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units /
|
const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units /
|
||||||
(context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
|
(context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
|
||||||
|
|
||||||
writer.WriteUe(16);
|
// TODO (ameerj): Where do we get this number, it seems to be particular for each stream
|
||||||
|
writer.WriteUe(6); // Max number of reference frames
|
||||||
writer.WriteBit(false);
|
writer.WriteBit(false);
|
||||||
writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
|
writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
|
||||||
writer.WriteUe(pic_height - 1);
|
writer.WriteUe(pic_height - 1);
|
||||||
|
Loading…
Reference in New Issue
Block a user