Re: [PATCH v2 2/3] media: cedrus: Fix H264 default reference index count

[Jernej Åkrabec]

Dne Äetrtek, 03. oktober 2019 ob 22:58:57 CEST je Paul Kocialkowski 
napisal(a):
> Hi,
> 
> On Thu 03 Oct 19, 22:44, Jernej Åkrabec wrote:
> > Dne Äetrtek, 03. oktober 2019 ob 22:28:46 CEST je Paul Kocialkowski
> > 
> > napisal(a):
> > > Hi,
> > > 
> > > On Thu 03 Oct 19, 07:16, Jernej Åkrabec wrote:
> > > > Dne Äetrtek, 03. oktober 2019 ob 00:06:50 CEST je Paul Kocialkowski
> > > > 
> > > > napisal(a):
> > > > > Hi,
> > > > > 
> > > > > On Wed 02 Oct 19, 21:35, Jernej Skrabec wrote:
> > > > > > Reference index count in VE_H264_PPS should come from PPS control.
> > > > > > However, this is not really important, because reference index
> > > > > > count
> > > > > > is
> > > > > > in our case always overridden by that from slice header.
> > > > > 
> > > > > Thanks for the fixup!
> > > > > 
> > > > > Our libva userspace and v4l2-request testing tool currently don't
> > > > > provide
> > > > > this, but I have a pending merge request adding it for the hantro so
> > > > > it's
> > > > > good to go.
> > > > 
> > > > Actually, I think this is just cosmetic and it would work even if it
> > > > would
> > > > be always 0. We always override this number in SHS2 register with
> > > > VE_H264_SHS2_NUM_REF_IDX_ACTIVE_OVRD flag and recently there was a
> > > > patch
> > > > merged to clarify that value in slice parameters should be the one
> > > > that's
> > > > set on default value if override flag is not set in bitstream:
> > > > https://git.linuxtv.org/media_tree.git/commit/?
> > > > id=187ef7c5c78153acdce8c8714e5918b1018c710b
> > > > 
> > > > Well, we could always compare default and value in slice parameters,
> > > > but I
> > > > really don't see the benefit of doing that extra work.
> > > 
> > > Thanks for the detailed explanation! So I just realized that for HEVC, I
> > > didn't even include the default value in PPS and only went for the
> > > per-slice value. The HEVC hardware block apparently only needs the
> > > fields
> > > once at slice level, and by looking at the spec, only one of the two set
> > > of
> > > fields will be used.
> > > 
> > > So perhaps we could do the same for H.264 and only have the set of
> > > fields
> > > once in the slice params, so that both codecs are consistent. Userspace
> > > can
> > > just check the flag to know whether it should put the PPS default or
> > > slice-specific value in the slice-specific control.
> > > 
> > > What do you think?
> > 
> > I think that there would be less confusion if only value in slice params
> > would exists. But since Philipp rather made clarification in
> > documentation, maybe he sees benefit having both values?
> 
> Actually I just caught up with the discussion from thread:
> media: uapi: h264: Add num_ref_idx_active_override_flag
> 
> which explains that we need to pass the default fields for hardware that
> parses the slice header itself and we need the non-default fields and flag
> for other cases.
> 
> To cover the case of hardware that does slice header parsing, I guess it
> would also work to use the slice-specific values in place of the pps
> default values in the hardware register for that. But it feels quite
> confusing and a lot less straightforward than having all the fields and the
> override flag exposed.

I wasn't aware of that patch and related discussion. Ok, so it make sense to 
have both values. However, does it make sense to use default values in Cedrus?

> 
> So I think I should fix HEVC support accordingly, just in case the same
> situation arises for HEVC.

Seems reasonable. Does that mean there will be another revision of HEVC 
patches?  If so, I think slice_segment_addr should also be included in slice 
params, so multi-slice frames can be supported at later time.

Best regards,
Jernej 

> 
> Cheers,
> 
> Paul
> 
> > Best regards,
> > Jernej
> > 
> > > Cheers,
> > > 
> > > Paul
> > > 
> > > > Best regards,
> > > > Jernej
> > > > 
> > > > > Acked-by: Paul Kocialkowski <paul.kocialkowski@xxxxxxxxxxx>
> > > > > 
> > > > > Cheers,
> > > > > 
> > > > > Paul
> > > > > 
> > > > > > Signed-off-by: Jernej Skrabec <jernej.skrabec@xxxxxxxx>
> > > > > > ---
> > > > > > 
> > > > > >  drivers/staging/media/sunxi/cedrus/cedrus_h264.c | 8 ++------
> > > > > >  1 file changed, 2 insertions(+), 6 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/staging/media/sunxi/cedrus/cedrus_h264.c
> > > > > > b/drivers/staging/media/sunxi/cedrus/cedrus_h264.c index
> > > > > > bd848146eada..4a0e69855c7f 100644
> > > > > > --- a/drivers/staging/media/sunxi/cedrus/cedrus_h264.c
> > > > > > +++ b/drivers/staging/media/sunxi/cedrus/cedrus_h264.c
> > > > > > @@ -364,12 +364,8 @@ static void cedrus_set_params(struct
> > > > > > cedrus_ctx
> > > > > > *ctx,
> > > > > > 
> > > > > >  	// picture parameters
> > > > > >  	reg = 0;
> > > > > > 
> > > > > > -	/*
> > > > > > -	 * FIXME: the kernel headers are allowing the default value to
> > > > > > -	 * be passed, but the libva doesn't give us that.
> > > > > > -	 */
> > > > > > -	reg |= (slice->num_ref_idx_l0_active_minus1 & 0x1f) << 10;
> > > > > > -	reg |= (slice->num_ref_idx_l1_active_minus1 & 0x1f) << 5;
> > > > > > +	reg |= (pps->num_ref_idx_l0_default_active_minus1 & 0x1f) << 10;
> > > > > > +	reg |= (pps->num_ref_idx_l1_default_active_minus1 & 0x1f) << 5;
> > > > > > 
> > > > > >  	reg |= (pps->weighted_bipred_idc & 0x3) << 2;
> > > > > >  	if (pps->flags & V4L2_H264_PPS_FLAG_ENTROPY_CODING_MODE)
> > > > > >  	
> > > > > >  		reg |= VE_H264_PPS_ENTROPY_CODING_MODE;