[PATCH] media: verisilicon: AV1: Fix tx mode bit setting

[Nicolas Dufresne]

Hi,

Le lundi 08 décembre 2025 à 10:52 +0100, Benjamin Gaignard a écrit :
> If tx_mode field is non-zero then write it value + 2 else

it -> its



> write 0.


Please don't sent a an In-reply-to: email for a patch that isn't part of a
series.


> 
> Signed-off-by: Benjamin Gaignard <benjamin.gaignard at collabora.com>
> Fixes: 727a400686a2c ("media: verisilicon: Add Rockchip AV1 decoder")
> ---
>  drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c
> b/drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c
> index f4f7cb45b1f1..fccdece51b1b 100644
> --- a/drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c
> +++ b/drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c
> @@ -2005,7 +2005,7 @@ static void
> rockchip_vpu981_av1_dec_set_parameters(struct hantro_ctx *ctx)
>  			 !!(ctrls->frame->flags &
> V4L2_AV1_FRAME_FLAG_ALLOW_HIGH_PRECISION_MV));
>  	hantro_reg_write(vpu, &av1_comp_pred_mode,
>  			 (ctrls->frame->flags &
> V4L2_AV1_FRAME_FLAG_REFERENCE_SELECT) ? 2 : 0);
> -	hantro_reg_write(vpu, &av1_transform_mode, (ctrls->frame->tx_mode ==
> 1) ? 3 : 4);
> +	hantro_reg_write(vpu, &av1_transform_mode, ctrls->frame->tx_mode ?
> ctrls->frame->tx_mode + 2 : 0);

That all seem very hacky. Let's step back, and use that as inspiration for a
useful commit message. From bitstream we have:

tx_mode:
  0 == 4x4 only
  1 == SELECT
  2 == LARGEST

And the HW have:
  0 == 4x4 only
  1 == 8x8 and less
  2 == 16x16 and less
  3 == 32x32 and less
  4 == Select

Since the two enums have no relation (except for value 0), I'd suggest to
translate this in a switch (with nice enums so its readable).

  0 -> 0
  1 -> 4
  2 -> 3

Would be good to check if we can refine with some other params to reach HW value
1 and 2, but otherwise, that mapping is sufficient. Tha hacky +2 is really
obscure to me and I'd rather not do that upstream.

Nicolas

>  	hantro_reg_write(vpu, &av1_max_cb_size,
>  			 (ctrls->sequence->flags
>  			  & V4L2_AV1_SEQUENCE_FLAG_USE_128X128_SUPERBLOCK) ?
> 7 : 6);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: This is a digitally signed message part
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20251208/6965e3c9/attachment.sig>