diff mbox series

[v1,5/9] media: verisilicon: Compute motion vectors size for AV1 frames

Message ID 20221219155616.848690-6-benjamin.gaignard@collabora.com
State New
Headers
Series AV1 stateless decoder for RK3588 |

Commit Message

Benjamin Gaignard Dec. 19, 2022, 3:56 p.m. UTC 
  Compute the additional required to store motion vectors at
the end of the frames buffers.


Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
---
 drivers/media/platform/verisilicon/hantro_hw.h      | 13 +++++++++++++
 .../media/platform/verisilicon/hantro_postproc.c    |  3 +++
 drivers/media/platform/verisilicon/hantro_v4l2.c    |  5 +++++
 3 files changed, 21 insertions(+)

Comments

Nicolas Dufresne Dec. 19, 2022, 8:42 p.m. UTC | #1 
Le lundi 19 décembre 2022 à 16:56 +0100, Benjamin Gaignard a écrit :
> Compute the additional required to store motion vectors at

requires *space*, requires *buffer* ? I think this is missing a word.

> the end of the frames buffers.
> 
> 
> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> ---
>  drivers/media/platform/verisilicon/hantro_hw.h      | 13 +++++++++++++
>  .../media/platform/verisilicon/hantro_postproc.c    |  3 +++
>  drivers/media/platform/verisilicon/hantro_v4l2.c    |  5 +++++
>  3 files changed, 21 insertions(+)
> 
> diff --git a/drivers/media/platform/verisilicon/hantro_hw.h b/drivers/media/platform/verisilicon/hantro_hw.h
> index e83f0c523a30..8b3bc7e31395 100644
> --- a/drivers/media/platform/verisilicon/hantro_hw.h
> +++ b/drivers/media/platform/verisilicon/hantro_hw.h
> @@ -417,6 +417,19 @@ hantro_hevc_mv_size(unsigned int width, unsigned int height)
>  	return width * height / 16;
>  }
>  
> +static inline unsigned short hantro_av1_num_sbs(unsigned short dimension)
> +{
> +	return DIV_ROUND_UP(dimension, 64) + 1;

Why plus one ? I've tested locally with the logical DIV_ROUND_UP(dimension, 64),
and didn't see any difference. It then match hantro_vp_num_sbs(), so can't this
be shared ?

> +}
> +
> +static inline size_t
> +hantro_av1_mv_size(unsigned int width, unsigned int height)
> +{
> +	size_t num_sbs = hantro_av1_num_sbs(width) * hantro_av1_num_sbs(height);
> +
> +	return ALIGN(num_sbs * 384, 16) + 512;

Shall the magic numbers be turned into defines ?

> +}
> +
>  int hantro_g1_mpeg2_dec_run(struct hantro_ctx *ctx);
>  int rockchip_vpu2_mpeg2_dec_run(struct hantro_ctx *ctx);
>  void hantro_mpeg2_dec_copy_qtable(u8 *qtable,
> diff --git a/drivers/media/platform/verisilicon/hantro_postproc.c b/drivers/media/platform/verisilicon/hantro_postproc.c
> index 09d8cf942689..7dc39519a2ee 100644
> --- a/drivers/media/platform/verisilicon/hantro_postproc.c
> +++ b/drivers/media/platform/verisilicon/hantro_postproc.c
> @@ -213,6 +213,9 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)
>  	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_HEVC_SLICE)
>  		buf_size += hantro_hevc_mv_size(pix_mp.width,
>  						pix_mp.height);
> +	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME)
> +		buf_size += hantro_av1_mv_size(pix_mp.width,
> +					       pix_mp.height);

nit: Time to turn into a switch or use an ops ?

>  
>  	for (i = 0; i < num_buffers; ++i) {
>  		struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
> diff --git a/drivers/media/platform/verisilicon/hantro_v4l2.c b/drivers/media/platform/verisilicon/hantro_v4l2.c
> index 2c7a805289e7..d41dcb108a6d 100644
> --- a/drivers/media/platform/verisilicon/hantro_v4l2.c
> +++ b/drivers/media/platform/verisilicon/hantro_v4l2.c
> @@ -334,6 +334,11 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
>  			pix_mp->plane_fmt[0].sizeimage +=
>  				hantro_hevc_mv_size(pix_mp->width,
>  						    pix_mp->height);
> +		else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME &&
> +			 !hantro_needs_postproc(ctx, fmt))
> +			pix_mp->plane_fmt[0].sizeimage +=
> +				hantro_av1_mv_size(pix_mp->width,
> +						   pix_mp->height);
>  	} else if (!pix_mp->plane_fmt[0].sizeimage) {
>  		/*
>  		 * For coded formats the application can specify
Benjamin Gaignard Dec. 20, 2022, 1:13 p.m. UTC | #2 
Le 19/12/2022 à 21:42, Nicolas Dufresne a écrit :
> Le lundi 19 décembre 2022 à 16:56 +0100, Benjamin Gaignard a écrit :
>> Compute the additional required to store motion vectors at
> requires *space*, requires *buffer* ? I think this is missing a word.
>
>> the end of the frames buffers.
>>
>>
>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
>> ---
>>   drivers/media/platform/verisilicon/hantro_hw.h      | 13 +++++++++++++
>>   .../media/platform/verisilicon/hantro_postproc.c    |  3 +++
>>   drivers/media/platform/verisilicon/hantro_v4l2.c    |  5 +++++
>>   3 files changed, 21 insertions(+)
>>
>> diff --git a/drivers/media/platform/verisilicon/hantro_hw.h b/drivers/media/platform/verisilicon/hantro_hw.h
>> index e83f0c523a30..8b3bc7e31395 100644
>> --- a/drivers/media/platform/verisilicon/hantro_hw.h
>> +++ b/drivers/media/platform/verisilicon/hantro_hw.h
>> @@ -417,6 +417,19 @@ hantro_hevc_mv_size(unsigned int width, unsigned int height)
>>   	return width * height / 16;
>>   }
>>   
>> +static inline unsigned short hantro_av1_num_sbs(unsigned short dimension)
>> +{
>> +	return DIV_ROUND_UP(dimension, 64) + 1;
> Why plus one ? I've tested locally with the logical DIV_ROUND_UP(dimension, 64),
> and didn't see any difference. It then match hantro_vp_num_sbs(), so can't this
> be shared ?

MPP code use plus one so I keep it like that.

>
>> +}
>> +
>> +static inline size_t
>> +hantro_av1_mv_size(unsigned int width, unsigned int height)
>> +{
>> +	size_t num_sbs = hantro_av1_num_sbs(width) * hantro_av1_num_sbs(height);
>> +
>> +	return ALIGN(num_sbs * 384, 16) + 512;
> Shall the magic numbers be turned into defines ?

MPP code is:
dir_mvs_size = MPP_ALIGN(num_sbs * 24 * 128 / 8, 16);
and 512 is added later by another piece of code.

I have no clue about the meaning of those values, sorry.

>
>> +}
>> +
>>   int hantro_g1_mpeg2_dec_run(struct hantro_ctx *ctx);
>>   int rockchip_vpu2_mpeg2_dec_run(struct hantro_ctx *ctx);
>>   void hantro_mpeg2_dec_copy_qtable(u8 *qtable,
>> diff --git a/drivers/media/platform/verisilicon/hantro_postproc.c b/drivers/media/platform/verisilicon/hantro_postproc.c
>> index 09d8cf942689..7dc39519a2ee 100644
>> --- a/drivers/media/platform/verisilicon/hantro_postproc.c
>> +++ b/drivers/media/platform/verisilicon/hantro_postproc.c
>> @@ -213,6 +213,9 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)
>>   	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_HEVC_SLICE)
>>   		buf_size += hantro_hevc_mv_size(pix_mp.width,
>>   						pix_mp.height);
>> +	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME)
>> +		buf_size += hantro_av1_mv_size(pix_mp.width,
>> +					       pix_mp.height);
> nit: Time to turn into a switch or use an ops ?
>
>>   
>>   	for (i = 0; i < num_buffers; ++i) {
>>   		struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
>> diff --git a/drivers/media/platform/verisilicon/hantro_v4l2.c b/drivers/media/platform/verisilicon/hantro_v4l2.c
>> index 2c7a805289e7..d41dcb108a6d 100644
>> --- a/drivers/media/platform/verisilicon/hantro_v4l2.c
>> +++ b/drivers/media/platform/verisilicon/hantro_v4l2.c
>> @@ -334,6 +334,11 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
>>   			pix_mp->plane_fmt[0].sizeimage +=
>>   				hantro_hevc_mv_size(pix_mp->width,
>>   						    pix_mp->height);
>> +		else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME &&
>> +			 !hantro_needs_postproc(ctx, fmt))
>> +			pix_mp->plane_fmt[0].sizeimage +=
>> +				hantro_av1_mv_size(pix_mp->width,
>> +						   pix_mp->height);
>>   	} else if (!pix_mp->plane_fmt[0].sizeimage) {
>>   		/*
>>   		 * For coded formats the application can specify
diff mbox series

Patch

diff --git a/drivers/media/platform/verisilicon/hantro_hw.h b/drivers/media/platform/verisilicon/hantro_hw.h
index e83f0c523a30..8b3bc7e31395 100644
--- a/drivers/media/platform/verisilicon/hantro_hw.h
+++ b/drivers/media/platform/verisilicon/hantro_hw.h
@@ -417,6 +417,19 @@  hantro_hevc_mv_size(unsigned int width, unsigned int height)
 	return width * height / 16;
 }
 
+static inline unsigned short hantro_av1_num_sbs(unsigned short dimension)
+{
+	return DIV_ROUND_UP(dimension, 64) + 1;
+}
+
+static inline size_t
+hantro_av1_mv_size(unsigned int width, unsigned int height)
+{
+	size_t num_sbs = hantro_av1_num_sbs(width) * hantro_av1_num_sbs(height);
+
+	return ALIGN(num_sbs * 384, 16) + 512;
+}
+
 int hantro_g1_mpeg2_dec_run(struct hantro_ctx *ctx);
 int rockchip_vpu2_mpeg2_dec_run(struct hantro_ctx *ctx);
 void hantro_mpeg2_dec_copy_qtable(u8 *qtable,
diff --git a/drivers/media/platform/verisilicon/hantro_postproc.c b/drivers/media/platform/verisilicon/hantro_postproc.c
index 09d8cf942689..7dc39519a2ee 100644
--- a/drivers/media/platform/verisilicon/hantro_postproc.c
+++ b/drivers/media/platform/verisilicon/hantro_postproc.c
@@ -213,6 +213,9 @@  int hantro_postproc_alloc(struct hantro_ctx *ctx)
 	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_HEVC_SLICE)
 		buf_size += hantro_hevc_mv_size(pix_mp.width,
 						pix_mp.height);
+	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME)
+		buf_size += hantro_av1_mv_size(pix_mp.width,
+					       pix_mp.height);
 
 	for (i = 0; i < num_buffers; ++i) {
 		struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
diff --git a/drivers/media/platform/verisilicon/hantro_v4l2.c b/drivers/media/platform/verisilicon/hantro_v4l2.c
index 2c7a805289e7..d41dcb108a6d 100644
--- a/drivers/media/platform/verisilicon/hantro_v4l2.c
+++ b/drivers/media/platform/verisilicon/hantro_v4l2.c
@@ -334,6 +334,11 @@  static int hantro_try_fmt(const struct hantro_ctx *ctx,
 			pix_mp->plane_fmt[0].sizeimage +=
 				hantro_hevc_mv_size(pix_mp->width,
 						    pix_mp->height);
+		else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME &&
+			 !hantro_needs_postproc(ctx, fmt))
+			pix_mp->plane_fmt[0].sizeimage +=
+				hantro_av1_mv_size(pix_mp->width,
+						   pix_mp->height);
 	} else if (!pix_mp->plane_fmt[0].sizeimage) {
 		/*
 		 * For coded formats the application can specify