public inbox for linux-staging@lists.linux.dev
 help / color / mirror / Atom feed
From: Nicolas Dufresne <nicolas.dufresne@collabora.com>
To: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	kernel@collabora.com,
	Sebastian Fricke <sebastian.fricke@collabora.com>,
	linux-media@vger.kernel.org,  linux-rockchip@lists.infradead.org,
	linux-staging@lists.linux.dev,  linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 13/23] media: rkvdec: h264: Fix dpb_valid implementation
Date: Tue, 05 Apr 2022 11:10:57 -0400	[thread overview]
Message-ID: <92001085580ac1005596b34f79f1dadcc4b3f3c9.camel@collabora.com> (raw)
In-Reply-To: <YkgwmwTMa83RysKB@eze-laptop>

Le samedi 02 avril 2022 à 08:16 -0300, Ezequiel Garcia a écrit :
> On Thu, Mar 31, 2022 at 03:37:15PM -0400, Nicolas Dufresne wrote:
> > The ref builder only provided references that are marked as valid in the
> > dpb. Thus the current implementation of dpb_valid would always set the
> > flag to 1. This is not representing missing frames (this is called
> > 'non-existing' pictures in the spec). In some context, these non-existing
> > pictures still need to occupy a slot in the reference list according to
> > the spec.
> > 
> 
> Good catch! OOC, did you find this because it was failing a test vector?

The effect is complex, so I could not correlate to specific tests. Also, what I
wanted to fix isn't covered by the ITU conformance, its mostly resiliance
requirement. But this should remove some of the IOMMU fault on broken streams
and make it less likely to use references that don't exists or aren't set what
we expect. After this change, the driver was getting more stable, and results
was also more reproducible (specially in parallel decode case, which I use to
speed up testing).

> 
> > Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
> > Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
> 
> Fixes: cd33c830448ba ("media: rkvdec: Add the rkvdec driver")
> Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>

Thanks for the review.

> 
> Thanks,
> Ezequiel
> 
> > ---
> >  drivers/staging/media/rkvdec/rkvdec-h264.c | 33 ++++++++++++++++------
> >  1 file changed, 24 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
> > index dff89732ddd0..bcde37d72244 100644
> > --- a/drivers/staging/media/rkvdec/rkvdec-h264.c
> > +++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
> > @@ -112,6 +112,7 @@ struct rkvdec_h264_run {
> >  	const struct v4l2_ctrl_h264_sps *sps;
> >  	const struct v4l2_ctrl_h264_pps *pps;
> >  	const struct v4l2_ctrl_h264_scaling_matrix *scaling_matrix;
> > +	int ref_buf_idx[V4L2_H264_NUM_DPB_ENTRIES];
> >  };
> >  
> >  struct rkvdec_h264_ctx {
> > @@ -725,6 +726,26 @@ static void assemble_hw_pps(struct rkvdec_ctx *ctx,
> >  	}
> >  }
> >  
> > +static void lookup_ref_buf_idx(struct rkvdec_ctx *ctx,
> > +			       struct rkvdec_h264_run *run)
> > +{
> > +	const struct v4l2_ctrl_h264_decode_params *dec_params = run->decode_params;
> > +	u32 i;
> > +
> > +	for (i = 0; i < ARRAY_SIZE(dec_params->dpb); i++) {
> > +		struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
> > +		const struct v4l2_h264_dpb_entry *dpb = run->decode_params->dpb;
> > +		struct vb2_queue *cap_q = &m2m_ctx->cap_q_ctx.q;
> > +		int buf_idx = -1;
> > +
> > +		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
> > +			buf_idx = vb2_find_timestamp(cap_q,
> > +						     dpb[i].reference_ts, 0);
> > +
> > +		run->ref_buf_idx[i] = buf_idx;
> > +	}
> > +}
> > +
> >  static void assemble_hw_rps(struct rkvdec_ctx *ctx,
> >  			    struct rkvdec_h264_run *run)
> >  {
> > @@ -762,7 +783,7 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
> >  
> >  	for (j = 0; j < RKVDEC_NUM_REFLIST; j++) {
> >  		for (i = 0; i < h264_ctx->reflists.num_valid; i++) {
> > -			u8 dpb_valid = 0;
> > +			bool dpb_valid = run->ref_buf_idx[i] >= 0;
> >  			u8 idx = 0;
> >  
> >  			switch (j) {
> > @@ -779,8 +800,6 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
> >  
> >  			if (idx >= ARRAY_SIZE(dec_params->dpb))
> >  				continue;
> > -			dpb_valid = !!(dpb[idx].flags &
> > -				       V4L2_H264_DPB_ENTRY_FLAG_ACTIVE);
> >  
> >  			set_ps_field(hw_rps, DPB_INFO(i, j),
> >  				     idx | dpb_valid << 4);
> > @@ -859,13 +878,8 @@ get_ref_buf(struct rkvdec_ctx *ctx, struct rkvdec_h264_run *run,
> >  	    unsigned int dpb_idx)
> >  {
> >  	struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
> > -	const struct v4l2_h264_dpb_entry *dpb = run->decode_params->dpb;
> >  	struct vb2_queue *cap_q = &m2m_ctx->cap_q_ctx.q;
> > -	int buf_idx = -1;
> > -
> > -	if (dpb[dpb_idx].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
> > -		buf_idx = vb2_find_timestamp(cap_q,
> > -					     dpb[dpb_idx].reference_ts, 0);
> > +	int buf_idx = run->ref_buf_idx[dpb_idx];
> >  
> >  	/*
> >  	 * If a DPB entry is unused or invalid, address of current destination
> > @@ -1102,6 +1116,7 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
> >  
> >  	assemble_hw_scaling_list(ctx, &run);
> >  	assemble_hw_pps(ctx, &run);
> > +	lookup_ref_buf_idx(ctx, &run);
> >  	assemble_hw_rps(ctx, &run);
> >  	config_registers(ctx, &run);
> >  
> > -- 
> > 2.34.1
> > 


  reply	other threads:[~2022-04-05 15:11 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20220331193726.289559-1-nicolas.dufresne@collabora.com>
2022-03-31 19:37 ` [PATCH v2 05/23] media: h264: Use v4l2_h264_reference for reflist Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 06/23] media: h264: Increase reference lists size to 32 Nicolas Dufresne
2022-04-02 12:03   ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 12/23] media: rkvdec: Stop overclocking the decoder Nicolas Dufresne
2022-04-02 11:05   ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 13/23] media: rkvdec: h264: Fix dpb_valid implementation Nicolas Dufresne
2022-04-02 11:16   ` Ezequiel Garcia
2022-04-05 15:10     ` Nicolas Dufresne [this message]
2022-04-05 15:34       ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 14/23] media: rkvdec: h264: Fix bit depth wrap in pps packet Nicolas Dufresne
2022-04-02 11:20   ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 15/23] media: rkvdec: h264: Validate and use pic width and height in mbs Nicolas Dufresne
2022-04-02 11:32   ` Ezequiel Garcia
2022-04-05 15:44     ` Nicolas Dufresne
2022-04-05 16:04       ` Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 16/23] media: rkvdec: h264: Fix reference frame_num wrap for second field Nicolas Dufresne
2022-04-02 11:33   ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 17/23] media: rkvdec: Enable capture buffer holding for H264 Nicolas Dufresne
2022-04-02 11:35   ` Ezequiel Garcia
2022-03-31 19:37 ` [PATCH v2 18/23] media: rkvdec: Ensure decoded resolution fit coded resolution Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 19/23] media: rkvdec-h264: Add field decoding support Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 20/23] media: hantro: Enable HOLD_CAPTURE_BUF for H.264 Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 21/23] media: hantro: Stop using H.264 parameter pic_num Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 22/23] media: hantro: h264: Make dpb entry management more robust Nicolas Dufresne
2022-03-31 19:37 ` [PATCH v2 23/23] media: hantro: Add H.264 field decoding support Nicolas Dufresne

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=92001085580ac1005596b34f79f1dadcc4b3f3c9.camel@collabora.com \
    --to=nicolas.dufresne@collabora.com \
    --cc=ezequiel@vanguardiasur.com.ar \
    --cc=gregkh@linuxfoundation.org \
    --cc=kernel@collabora.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-media@vger.kernel.org \
    --cc=linux-rockchip@lists.infradead.org \
    --cc=linux-staging@lists.linux.dev \
    --cc=mchehab@kernel.org \
    --cc=sebastian.fricke@collabora.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox