Re: media: mtk-vcodec: reset segment data then trig decoder

[Alexandre Courbot]

On Wed, Jan 15, 2020 at 12:47 PM Alexandre Courbot
<acourbot@xxxxxxxxxxxx> wrote:
>
> On Tue, Jan 14, 2020 at 12:32 PM <gtk_ruiwang@xxxxxxxxxxxx> wrote:
> >
> > From: gtk_ruiwang <gtk_ruiwang@xxxxxxxxxxxx>
> >
> > VP9 bitstream specification indicate segment data should reset to
> > default when meet key frames, intra only frames or enable error
> > resilience mode. So memset segmentation map buffer before every
> > decode process is not appropriate.
> >
> > Reset segment data only when needed, then trig decoder hardware
> >
> > Signed-off-by: Rui Wang <gtk_ruiwang@xxxxxxxxxxxx>
> > ---
> >  .../platform/mtk-vcodec/vdec/vdec_vp9_if.c    | 19 +++++++++++++++----
> >  1 file changed, 15 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/media/platform/mtk-vcodec/vdec/vdec_vp9_if.c b/drivers/media/platform/mtk-vcodec/vdec/vdec_vp9_if.c
> > index 24c1f0bf2147..42c9c3c98076 100644
> > --- a/drivers/media/platform/mtk-vcodec/vdec/vdec_vp9_if.c
> > +++ b/drivers/media/platform/mtk-vcodec/vdec/vdec_vp9_if.c
> > @@ -110,7 +110,9 @@ struct vp9_sf_ref_fb {
> >   * @buf_len_sz_c : size used to store cbcr plane ufo info (AP-R, VPU-W)
> >
> >   * @profile : profile sparsed from vpu (AP-R, VPU-W)
> > - * @show_frame : display this frame or not (AP-R, VPU-W)
> > + * @show_frame : [BIT(0)] display this frame or not (AP-R, VPU-W)
> > + *     [BIT(14)] reset segment data or not (AP-R, VPU-W)
> > + *     [BIT(15)] trig decoder hardware or not (AP-R, VPU-W)
> >   * @show_existing_frame : inform this frame is show existing frame
> >   *     (AP-R, VPU-W)
> >   * @frm_to_show_idx : index to show frame (AP-R, VPU-W)
> > @@ -494,12 +496,12 @@ static void vp9_swap_frm_bufs(struct vdec_vp9_inst *inst)
> >                                         frm_to_show->fb->base_y.size);
> >                 }
> >                 if (!vp9_is_sf_ref_fb(inst, inst->cur_fb)) {
> > -                       if (vsi->show_frame)
> > +                       if (vsi->show_frame & BIT(0))
> >                                 vp9_add_to_fb_disp_list(inst, inst->cur_fb);
> >                 }
> >         } else {
> >                 if (!vp9_is_sf_ref_fb(inst, inst->cur_fb)) {
> > -                       if (vsi->show_frame)
> > +                       if (vsi->show_frame & BIT(0))
> >                                 vp9_add_to_fb_disp_list(inst, frm_to_show->fb);
> >                 }
> >         }
> > @@ -870,13 +872,22 @@ static int vdec_vp9_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
> >                                         vsi->sf_frm_sz[idx]);
> >                         }
> >                 }
> > -               memset(inst->seg_id_buf.va, 0, inst->seg_id_buf.size);
> >                 ret = vpu_dec_start(&inst->vpu, data, 3);
> >                 if (ret) {
> >                         mtk_vcodec_err(inst, "vpu_dec_start failed");
> >                         goto DECODE_ERROR;
> >                 }
> >
> > +               if ((vsi->show_frame & BIT(15)) &&
> > +                   (vsi->show_frame & BIT(14))) {
>
> Using the new bits in this manner means this patch is not compatible
> with the older firmware.
>
> On an older firmware, these bits will be 0, which means the decoder
> will never be started. To preserve compatibility, the behavior should
> be reversed: *do not* reset and/or start the decoder if the bits are
> set.
>
> Also both bits are only used together - we should either separate the
> data segment reset and decoder start, or rely on only one bit for
> this.

Ah, looks like I missed the fact that the decoder is still started
even when these bits are not set. So this indeed looks
backward-compatible. Please ignore my comment.