VAAPI VVC decoder #714
VAAPI VVC decoder #714
Conversation
feiwan1
force-pushed
the
vvcd_github
branch
3 times, most recently
from
f71956d
to
f235f81
Compare
| pic->param_allocated, | ||
| sizeof(*pic->param_buffers)); | ||
| if (!pic->param_buffers) | ||
| return AVERROR(ENOMEM); |
There was a problem hiding this comment.
two issues if av_realloc_array failed
- pic->param_buffers is not released.
- you have updated pic->param_allocated, next "if (pic->nb_param_buffers == pic->param_allocated)" check is wrong
There was a problem hiding this comment.
- If realloc fail, ff_vaapi_decode_cancel() will be called to destory all va buffer and release param_buffers handle.
| @@ -138,6 +140,10 @@ static VVCFrame *alloc_frame(VVCContext *s, VVCFrameContext *fc) | |||
| if (!frame->progress) | |||
| goto fail; | |||
|
|
|||
| ret = ff_hwaccel_frame_priv_alloc(s->avctx, &frame->hwaccel_picture_private); | |||
There was a problem hiding this comment.
all software buffer allocation can be skipped, like rpl to progress
also all software delay frames can be removed. we do not need a delay for hw. we can output it immediately after vaEndPicture.
There was a problem hiding this comment.
Maybe better to keep software buffers there which can help check illegal syntax. And RPL may insert new ref frame to DPB in some special case, if skipped it, maybe cause unexpected error to HW. I will check how to remove defay frames.
There was a problem hiding this comment.
Delay logic in vvc native decoder may provide sync_depth function to HW like VAAPI encoder/vpp. And there is option that can turn OFF the delay. So I think it will be better to keep delay for HW too.
| typedef struct VAAPIDecodePicture { | ||
| VASurfaceID output_surface; | ||
|
|
||
| int nb_param_buffers; | ||
| VABufferID param_buffers[MAX_PARAM_BUFFERS]; | ||
| VABufferID *param_buffers; | ||
| int param_allocated; |
There was a problem hiding this comment.
s/param_allocated/nb_allocated_buffers ?
There was a problem hiding this comment.
I'd prefer param_allocated which is samilar with slices_allocated.
Signed-off-by: Fei Wang <[email protected]>
So that hardware decoder can use the flags too. Signed-off-by: Fei Wang <[email protected]>
Signed-off-by: Fei Wang <[email protected]>
Signed-off-by: Fei Wang <[email protected]>
Signed-off-by: Fei Wang <[email protected]>
Signed-off-by: Fei Wang <[email protected]>
Signed-off-by: Fei Wang <[email protected]>
No description provided.