[PATCH v4 05/10] media: verisilicon: Store chroma and motion vectors offset

Benjamin Gaignard benjamin.gaignard at collabora.com
Tue Aug 22 05:55:49 PDT 2023


Le 21/08/2023 à 16:41, Hans Verkuil a écrit :
> On 05/07/2023 14:10, Benjamin Gaignard wrote:
>> Store computed values of chroma and motion vectors offset because
>> they depends on width and height values which change if the resolution
>> change.
> Is this a bug fix? Does this patch belong in this series?
>
> Same actually for the next few verisilicon patches. Shouldn't they be
> part of a separate 'fixes' patch series? It's confusing to see them
> in this series.

They fix bugs that happens only when VP9 resolution change without doing stream off/on
that why they are in this series.

This one, for example, is useless if the resolution change on a keyframe because the
frame will have the same resolution than the current one but is need to store resolution
of each frames if the resize happens between keyframes.

Benjamin

>
> Regards,
>
> 	Hans
>
>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard at collabora.com>
>> ---
>>   drivers/media/platform/verisilicon/hantro.h            | 2 ++
>>   drivers/media/platform/verisilicon/hantro_g2_vp9_dec.c | 6 ++++--
>>   2 files changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/media/platform/verisilicon/hantro.h b/drivers/media/platform/verisilicon/hantro.h
>> index c8a3cf10cc64..53be00142473 100644
>> --- a/drivers/media/platform/verisilicon/hantro.h
>> +++ b/drivers/media/platform/verisilicon/hantro.h
>> @@ -320,6 +320,8 @@ struct hantro_vp9_decoded_buffer_info {
>>   	/* Info needed when the decoded frame serves as a reference frame. */
>>   	unsigned short width;
>>   	unsigned short height;
>> +	size_t chroma_offset;
>> +	size_t mv_offset;
>>   	u32 bit_depth : 4;
>>   };
>>   
>> diff --git a/drivers/media/platform/verisilicon/hantro_g2_vp9_dec.c b/drivers/media/platform/verisilicon/hantro_g2_vp9_dec.c
>> index 6fc4b555517f..6db1c32fce4d 100644
>> --- a/drivers/media/platform/verisilicon/hantro_g2_vp9_dec.c
>> +++ b/drivers/media/platform/verisilicon/hantro_g2_vp9_dec.c
>> @@ -158,9 +158,11 @@ static void config_output(struct hantro_ctx *ctx,
>>   
>>   	chroma_addr = luma_addr + chroma_offset(ctx, dec_params);
>>   	hantro_write_addr(ctx->dev, G2_OUT_CHROMA_ADDR, chroma_addr);
>> +	dst->vp9.chroma_offset = chroma_offset(ctx, dec_params);
>>   
>>   	mv_addr = luma_addr + mv_offset(ctx, dec_params);
>>   	hantro_write_addr(ctx->dev, G2_OUT_MV_ADDR, mv_addr);
>> +	dst->vp9.mv_offset = mv_offset(ctx, dec_params);
>>   }
>>   
>>   struct hantro_vp9_ref_reg {
>> @@ -195,7 +197,7 @@ static void config_ref(struct hantro_ctx *ctx,
>>   	luma_addr = hantro_get_dec_buf_addr(ctx, &buf->base.vb.vb2_buf);
>>   	hantro_write_addr(ctx->dev, ref_reg->y_base, luma_addr);
>>   
>> -	chroma_addr = luma_addr + chroma_offset(ctx, dec_params);
>> +	chroma_addr = luma_addr + buf->vp9.chroma_offset;
>>   	hantro_write_addr(ctx->dev, ref_reg->c_base, chroma_addr);
>>   }
>>   
>> @@ -238,7 +240,7 @@ static void config_ref_registers(struct hantro_ctx *ctx,
>>   	config_ref(ctx, dst, &ref_regs[2], dec_params, dec_params->alt_frame_ts);
>>   
>>   	mv_addr = hantro_get_dec_buf_addr(ctx, &mv_ref->base.vb.vb2_buf) +
>> -		  mv_offset(ctx, dec_params);
>> +		  mv_ref->vp9.mv_offset;
>>   	hantro_write_addr(ctx->dev, G2_REF_MV_ADDR(0), mv_addr);
>>   
>>   	hantro_reg_write(ctx->dev, &vp9_last_sign_bias,



More information about the Linux-rockchip mailing list