On Fri, 13 Dec 2019 13:54:13 +0100
Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx> wrote:
> +static int rkvdec_try_capture_fmt(struct file *file, void *priv,
> + struct v4l2_format *f)
> +{
> + struct rkvdec_ctx *ctx = fh_to_rkvdec_ctx(priv);
> + const struct rkvdec_coded_fmt_desc *coded_desc;
> + u32 fourcc, width, height;
> + unsigned int i;
> +
> + /*
> + * The codec context should point to a coded format desc, if the format
> + * on the coded end has not been set yet, it should point to the
> + * default value.
> + */
> + coded_desc = ctx->coded_fmt_desc;
> + if (WARN_ON(!coded_desc))
> + return -EINVAL;
> +
> + fourcc = f->fmt.pix_mp.pixelformat;
> + for (i = 0; i < coded_desc->num_decoded_fmts; i++) {
> + if (coded_desc->decoded_fmts[i] == fourcc)
> + break;
> + }
> +
> + if (i == coded_desc->num_decoded_fmts)
> + return -EINVAL;
> +
> + /* Save the original width/height before aligning them. */
> + width = f->fmt.pix_mp.width;
> + height = f->fmt.pix_mp.height;
> +
> + /* Always apply the frmsize constraint of the coded end. */
> + v4l2_apply_frmsize_constraints(&f->fmt.pix_mp.width,
> + &f->fmt.pix_mp.height,
> + &coded_desc->frmsize);
> +
> + v4l2_fill_pixfmt_mp(&f->fmt.pix_mp, fourcc, f->fmt.pix_mp.width,
> + f->fmt.pix_mp.height);
> +
> + /*
> + * Now that we have computed sizeimage and bytesperline we can restore
> + * the original width/height (before macro block alignment).
> + */
> + f->fmt.pix_mp.width = width;
> + f->fmt.pix_mp.height = height;
Looks like restoring the original width/height was a bad idea (it
breaks the odd_size tests of the VP9 testsuite [1]). I'll adjust the
test in validate_dec_params() [2] to align the ctrl width/height params
on a superblock size (64 pixels) instead.
> +
> + f->fmt.pix_mp.field = V4L2_FIELD_NONE;
> +
> + return 0;
> +}
[1]https://storage.googleapis.com/downloads.webmproject.org/vp9/decoder-test-streams/Profile_0_8bit.zip
[2]https://patchwork.kernel.org/patch/11290603/
