On 2022-10-05 22:58:48, Marijn Suijten wrote:
> On 2022-10-05 22:31:43, Dmitry Baryshkov wrote:
> > [..]
> > In fact, could you please take a look if we can switch to using this
> > function and drop our code?
>
> [..]
>
> Do you want me to do this in a v3, before applying this fractional-bits
> fix?  [..]

One thing to note:

        /* bpc 8 */
        dsc->flatness_min_qp = 3;
        dsc->flatness_max_qp = 12;
        dsc->rc_quant_incr_limit0 = 11;
        dsc->rc_quant_incr_limit1 = 11;
        dsc->mux_word_size = DSC_MUX_WORD_SIZE_8_10_BPC;

Here a bunch of properties are hardcoded, seemingly for bpc = 8.
mux_word_size is only ever read in drm_dsc_compute_rc_parameters() so
only becomes relevant _after_ the migration, and is currently dealt with
correctly by:

        mux_words_size = 48;            /* bpc == 8/10 */
        if (dsc->bits_per_component == 12)
                mux_words_size = 64;

Aside fixing that to assign these values (through the proper constants)
to dsc->mux_word_size, is it worth looking for the right parameters for
other bpc or return -EINVAL if bpc isn't 8, to uphold this comment until
support for other bpc is validated?

- Marijn

Reply via email to