Re: [PATCH v2 13/13] media: verisilicon: Conditionnaly ignore native formats

From: Ezequiel Garcia
Date: Tue Jan 10 2023 - 14:32:36 EST


Hi Benjamin,

On Mon, Jan 9, 2023 at 6:05 AM Benjamin Gaignard
<benjamin.gaignard@xxxxxxxxxxxxx> wrote:
>
>
> Le 08/01/2023 à 22:12, Ezequiel Garcia a écrit :
> > On Tue, Jan 3, 2023 at 2:01 PM Benjamin Gaignard
> > <benjamin.gaignard@xxxxxxxxxxxxx> wrote:
> >> AV1 film grain feature requires to use the postprocessor to produce
> >> valid frames. In such case the driver shouldn't propose native pixels
> >> format but only post-processed pixels format.
> >> If a codec set need_postproc field in hantro_ctx structure to true
> >> native pixel formats will be ignored.
> >>
> >> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@xxxxxxxxxxxxx>
> >> ---
> >> drivers/media/platform/verisilicon/hantro.h | 3 ++
> >> .../media/platform/verisilicon/hantro_drv.c | 5 ++
> >> .../platform/verisilicon/hantro_postproc.c | 4 ++
> >> .../media/platform/verisilicon/hantro_v4l2.c | 46 +++++++++++++------
> >> 4 files changed, 45 insertions(+), 13 deletions(-)
> >>
> >> diff --git a/drivers/media/platform/verisilicon/hantro.h b/drivers/media/platform/verisilicon/hantro.h
> >> index a98cb40a8d3b..7a5357e810fb 100644
> >> --- a/drivers/media/platform/verisilicon/hantro.h
> >> +++ b/drivers/media/platform/verisilicon/hantro.h
> >> @@ -231,6 +231,8 @@ struct hantro_dev {
> >> * @ctrl_handler: Control handler used to register controls.
> >> * @jpeg_quality: User-specified JPEG compression quality.
> >> * @bit_depth: Bit depth of current frame
> >> + * @need_postproc: Set to true if the bitstream features require to
> >> + * use the post-processor.
> >> *
> >> * @codec_ops: Set of operations related to codec mode.
> >> * @postproc: Post-processing context.
> >> @@ -258,6 +260,7 @@ struct hantro_ctx {
> >> struct v4l2_ctrl_handler ctrl_handler;
> >> int jpeg_quality;
> >> int bit_depth;
> >> + bool need_postproc;
> >>
> >> const struct hantro_codec_ops *codec_ops;
> >> struct hantro_postproc_ctx postproc;
> >> diff --git a/drivers/media/platform/verisilicon/hantro_drv.c b/drivers/media/platform/verisilicon/hantro_drv.c
> >> index 4fc6dea16ae6..8d7055c0bf3b 100644
> >> --- a/drivers/media/platform/verisilicon/hantro_drv.c
> >> +++ b/drivers/media/platform/verisilicon/hantro_drv.c
> >> @@ -346,6 +346,11 @@ static int hantro_av1_s_ctrl(struct v4l2_ctrl *ctrl)
> >> return -EINVAL;
> >>
> >> ctx->bit_depth = bit_depth;
> >> +
> >> + if (ctrl->p_new.p_av1_sequence->flags
> >> + & V4L2_AV1_SEQUENCE_FLAG_FILM_GRAIN_PARAMS_PRESENT)
> >> + ctx->need_postproc = true;
> >> +

As discussed with Nicolas in IRC, I believe this should be handled differently.
This case is similar to
https://lore.kernel.org/all/f26407dbf3efc6cc046daaabdbe75c516743a187.camel@xxxxxxxxxxxxx/
and also similar to
https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/sunxi/sun8i-rotate/sun8i_rotate.c#L582.

Setting the SPS control changes the possible pixel formats, so the S_CTRL ioctl
should reset the format.

This means try_fmt should probably either check the (newly) SPS value
or some context fields (ctx->bit_depth, ctx->av1_film_grain, etc.).

Thanks,
Ezeqiuel

> >> break;
> >> default:
> >> return -EINVAL;
> >> diff --git a/drivers/media/platform/verisilicon/hantro_postproc.c b/drivers/media/platform/verisilicon/hantro_postproc.c
> >> index 7dc39519a2ee..293e5612e2ce 100644
> >> --- a/drivers/media/platform/verisilicon/hantro_postproc.c
> >> +++ b/drivers/media/platform/verisilicon/hantro_postproc.c
> >> @@ -57,6 +57,10 @@ bool hantro_needs_postproc(const struct hantro_ctx *ctx,
> >> {
> >> if (ctx->is_encoder)
> >> return false;
> >> +
> >> + if (ctx->need_postproc)
> >> + return true;
> >> +
> >> return fmt->postprocessed;
> >> }
> >>
> >> diff --git a/drivers/media/platform/verisilicon/hantro_v4l2.c b/drivers/media/platform/verisilicon/hantro_v4l2.c
> >> index bbe79dbd2cd9..5c381766cca3 100644
> >> --- a/drivers/media/platform/verisilicon/hantro_v4l2.c
> >> +++ b/drivers/media/platform/verisilicon/hantro_v4l2.c
> >> @@ -38,6 +38,11 @@ hantro_get_formats(const struct hantro_ctx *ctx, unsigned int *num_fmts)
> >> {
> >> const struct hantro_fmt *formats;
> >>
> >> + if (ctx->need_postproc) {
> >> + *num_fmts = 0;
> >> + return NULL;
> >> + }
> >> +
> >> if (ctx->is_encoder) {
> >> formats = ctx->dev->variant->enc_fmts;
> >> *num_fmts = ctx->dev->variant->num_enc_fmts;
> >> @@ -132,6 +137,15 @@ hantro_get_default_fmt(const struct hantro_ctx *ctx, bool bitstream)
> >> hantro_check_depth_match(ctx, &formats[i]))
> >> return &formats[i];
> >> }
> >> +
> >> + formats = hantro_get_postproc_formats(ctx, &num_fmts);
> >> + for (i = 0; i < num_fmts; i++) {
> >> + if (bitstream == (formats[i].codec_mode !=
> >> + HANTRO_MODE_NONE) &&
> >> + hantro_check_depth_match(ctx, &formats[i]))
> >> + return &formats[i];
> >> + }
> >> +
> >> return NULL;
> >> }
> >>
> >> @@ -261,19 +275,6 @@ static int vidioc_g_fmt_out_mplane(struct file *file, void *priv,
> >> return 0;
> >> }
> >>
> >> -static int vidioc_g_fmt_cap_mplane(struct file *file, void *priv,
> >> - struct v4l2_format *f)
> >> -{
> >> - struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
> >> - struct hantro_ctx *ctx = fh_to_ctx(priv);
> >> -
> >> - vpu_debug(4, "f->type = %d\n", f->type);
> >> -
> >> - *pix_mp = ctx->dst_fmt;
> >> -
> >> - return 0;
> >> -}
> >> -
> >> static int hantro_try_fmt(const struct hantro_ctx *ctx,
> >> struct v4l2_pix_format_mplane *pix_mp,
> >> enum v4l2_buf_type type)
> >> @@ -353,6 +354,25 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> >> return 0;
> >> }
> >>
> >> +static int vidioc_g_fmt_cap_mplane(struct file *file, void *priv,
> >> + struct v4l2_format *f)
> >> +{
> >> + struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
> >> + struct hantro_ctx *ctx = fh_to_ctx(priv);
> >> + int ret;
> >> +
> >> + vpu_debug(4, "f->type = %d\n", f->type);
> >> +
> >> + ret = hantro_try_fmt(ctx, pix_mp, f->type);
> >> + if (ret)
> >> + return ret;
> >> +
> >> + ctx->vpu_dst_fmt = hantro_find_format(ctx, pix_mp->pixelformat);
> >> + ctx->dst_fmt = *pix_mp;
> >> +
> > This looks like the g_fmt is setting some state in the context,
> > this looks incorrect.
>
> Indeed only a call to hantro_try_fmt() is needed here.
> I will fix that in v3.
>
> Benjamin
>
> >
> > Thanks,
> > Ezequiel