Re: [PATCH] Allow S/G_PARM for stateful decoder

From: Hans Verkuil
Date: Fri Sep 09 2022 - 08:19:49 EST


Hi Dikshita,

First I'll comment on this patch, then at the end I have more high-level
comments.

On 29/08/2022 09:54, Dikshita Agarwal wrote:
> Some stateful decoder supports S/G_PARM similar to a

decoder -> decoders

BTW, which decoder drivers support this in mainline? Or is this
for out-of-tree drivers?

> stateful encoder. S_PARM(OUTPUT) reserves hardware decoder
> resources, and G_PARM(CAPTURE) returns the embedded
> frame interval. Allow the same with this change.
>
> Signed-off-by: Dikshita Agarwal <quic_dikshita@xxxxxxxxxxx>
> ---
> utils/v4l2-compliance/v4l2-test-formats.cpp | 4 ----
> 1 file changed, 4 deletions(-)
>
> diff --git a/utils/v4l2-compliance/v4l2-test-formats.cpp b/utils/v4l2-compliance/v4l2-test-formats.cpp
> index 269a383..e996558 100644
> --- a/utils/v4l2-compliance/v4l2-test-formats.cpp
> +++ b/utils/v4l2-compliance/v4l2-test-formats.cpp
> @@ -64,8 +64,6 @@ static int testEnumFrameIntervals(struct node *node, __u32 pixfmt,
> ret = doioctl(node, VIDIOC_ENUM_FRAMEINTERVALS, &frmival);
> if (ret == ENOTTY)
> return ret;
> - // M2M devices don't support this, except for stateful encoders
> - fail_on_test(node->is_m2m && !(node->codec_mask & STATEFUL_ENCODER));

M2M devices still don't support this, except for stateful codecs. So this test
should still be there, just changed to:

!(node->codec_mask & (STATEFUL_ENCODER | STATEFUL_DECODER))

> if (f == 0 && ret == EINVAL) {
> if (type == V4L2_FRMSIZE_TYPE_DISCRETE)
> warn("found framesize %dx%d, but no frame intervals\n", w, h);
> @@ -1367,8 +1365,6 @@ static int testParmType(struct node *node, unsigned type)
> }
> if (ret == ENOTTY)
> return ret;
> - // M2M devices don't support this, except for stateful encoders
> - fail_on_test(node->is_m2m && !is_stateful_enc);

Similar.

> if (ret == EINVAL)
> return ENOTTY;
> if (ret)

But modifying the utility is something you only do if the spec has first been updated.
So the stateful decoder chapter needs to be modified so it explains how G/S_PARM
shall be used. And ideally have at least one driver that supports it.

Regards,

Hans