Re: [PATCH v3 4/4] uapi: linux: add MT8188 AIE

["CK Hu (胡俊光)" <ck.hu@mediatek.com>]

On Wed, 2024-12-25 at 17:00 +0800, bo.kong wrote:
> From: Bo Kong <Bo.Kong@mediatek.com>
> 
> Add AIE control related definitions.
> 
> Signed-off-by: Bo Kong <Bo.Kong@mediatek.com>
> ---
> 
> Changes in v3:
> none
> 
> Changes in v2:
> 1. Fix coding style
> ---
>  include/uapi/linux/mtk_aie_v4l2_controls.h | 132 +++++++++++++++++++++
>  include/uapi/linux/videodev2.h             |   6 +
>  2 files changed, 138 insertions(+)
>  create mode 100644 include/uapi/linux/mtk_aie_v4l2_controls.h
> 
> diff --git a/include/uapi/linux/mtk_aie_v4l2_controls.h b/include/uapi/linux/mtk_aie_v4l2_controls.h
> new file mode 100644
> index 000000000000..e635548c2cdf
> --- /dev/null
> +++ b/include/uapi/linux/mtk_aie_v4l2_controls.h
> @@ -0,0 +1,132 @@
> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> +/*
> + * AIE Controls Header
> + *
> + * Copyright (c) 2020 MediaTek Inc.
> + * Author: Fish Wu <fish.wu@mediatek.com>
> + */
> +
> +#ifndef __MTK_AIE_V4L2_CONTROLS_H__
> +#define __MTK_AIE_V4L2_CONTROLS_H__
> +
> +#include <linux/types.h>
> +
> +/*
> + * The base for the mediatek Face Detection driver controls.
> + * We reserve 16 controls for this driver.
> + * Each CID represents different stages of AIE, with different structures and functions
> + * and cannot be reused
> + */
> +#define V4L2_CID_USER_MTK_FD_BASE (V4L2_CID_USER_BASE + 0x1fd0)
> +
> +#define V4L2_CID_MTK_AIE_INIT (V4L2_CID_USER_MTK_FD_BASE + 1)
> +#define V4L2_CID_MTK_AIE_PARAM (V4L2_CID_USER_MTK_FD_BASE + 2)
> +
> +#define V4L2_FLD_MAX_FRAME 15
> +
> +/**
> + * struct v4l2_ctrl_aie_init - aie init parameters.
> + *
> + * @max_img_width: maximum width of the source image.
> + * @max_img_height: maximum height of the source image.
> + * @pyramid_width: maximum width of the base pyramid.
> + * @pyramid_height: maximum height of the base pyramid.
> + * @feature_threshold: feature threshold for hareware.
> + */
> +struct v4l2_ctrl_aie_init {
> +	__u32 max_img_width;
> +	__u32 max_img_height;
> +	__u32 pyramid_width;
> +	__u32 pyramid_height;
> +	__s32 feature_threshold;
> +};
> +
> +/**
> + * struct v4l2_aie_roi - aie roi parameters.
> + *
> + * @x1: x1 of the roi coordinate.
> + * @y1: y1 of the roi coordinate.
> + * @x2: x2 of the roi coordinate.
> + * @y2: y2 of the roi coordinate.
> + */
> +struct v4l2_aie_roi {
> +	__u32 x1;
> +	__u32 y1;
> +	__u32 x2;
> +	__u32 y2;
> +};
> +
> +/**
> + * struct v4l2_aie_padding - aie padding parameters.
> + *
> + * @left: the size of padding left.
> + * @right: the size of padding right.
> + * @down: the size of padding below.
> + * @up: the size of padding above.
> + */
> +struct v4l2_aie_padding {
> +	__u32 left;
> +	__u32 right;
> +	__u32 down;
> +	__u32 up;
> +};
> +
> +/**
> + * struct v4l2_fld_crop_rip_rop - aie fld parameters.
> + *
> + * @fld_in_crop_x1: x1 of the crop coordinate.
> + * @fld_in_crop_y1: y1 of the crop coordinate.
> + * @fld_in_crop_x2: x2 of the crop coordinate.
> + * @fld_in_crop_y2: y2 of the crop coordinate.
> + * @fld_in_rip: fld in rip.
> + * @fld_in_rop: fld in rop.
> + */
> +struct v4l2_fld_crop_rip_rop {
> +	__u32 fld_in_crop_x1;
> +	__u32 fld_in_crop_y1;
> +	__u32 fld_in_crop_x2;
> +	__u32 fld_in_crop_y2;
> +	__u32 fld_in_rip;
> +	__u32 fld_in_rop;
> +};
> +
> +/**
> + * struct v4l2_fld_crop_rip_rop - aie fld parameters.
> + *
> + * @fd_mode: select a mode for current fd.

Where is the definition of mode?

> + * @src_img_fmt: source image format.
> + * @src_img_width: the width of the source image.
> + * @src_img_height: the height of the source image.
> + * @src_img_stride: the stride of the source image.

I'm not familiar with V4L2, but I think there is a standard interface to set source buffer format, width, height, stride.

> + * @pyramid_base_width: the width of the base pyramid.
> + * @pyramid_base_height: the width of the base pyramid.
> + * @number_of_pyramid: number of pyramid.

How does pyramid work?
Does pyramid width and height has any limitation? Larger than image width and height?
How many number of pyramid could be set?

> + * @rotate_degree: the rotate degree of the image.
> + * @en_roi: enable roi.
> + * @src_roi: roi params.

What is roi? Does roi means 'region of interest'?
If it is enable, how it work?

> + * @en_padding: enable padding.
> + * @src_padding: padding params.

In which case need padding?
For face detection, padding would not help any thing.

> + * @freq_level: frequency level.

What's freq_level?
How to assign it?

> + * @fld_face_num: the number of faces in fld.
> + * @fld_input: fld input params.

What is fld?
Why user space already know how many faces in image?
I think number of faces is detected by kernel.

> + */
> +struct v4l2_ctrl_aie_param {
> +	__u32 fd_mode;
> +	__u32 src_img_fmt;
> +	__u32 src_img_width;
> +	__u32 src_img_height;
> +	__u32 src_img_stride;
> +	__u32 pyramid_base_width;
> +	__u32 pyramid_base_height;
> +	__u32 number_of_pyramid;
> +	__u32 rotate_degree;
> +	__s32 en_roi;
> +	struct v4l2_aie_roi src_roi;
> +	__s32 en_padding;
> +	struct v4l2_aie_padding src_padding;
> +	__u32 freq_level;
> +	__u32 fld_face_num;
> +	struct v4l2_fld_crop_rip_rop fld_input[V4L2_FLD_MAX_FRAME];
> +};
> +
> +#endif /* __MTK_AIE_V4L2_CONTROLS_H__ */
> diff --git a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
> index e7c4dce39007..b23a9e99c835 100644
> --- a/include/uapi/linux/videodev2.h
> +++ b/include/uapi/linux/videodev2.h
> @@ -877,6 +877,9 @@ struct v4l2_pix_format {
>  #define V4L2_META_FMT_GENERIC_CSI2_24	v4l2_fourcc('M', 'C', '1', 'O') /* 24-bit CSI-2 packed 8-bit metadata */
>  #endif
>  
> +/* Vendor-specific definition: used for the MediaTek camera subsystem's face detection results */
> +#define V4L2_META_FMT_MTFD_RESULT v4l2_fourcc('M', 'T', 'f', 'd')
> +
>  /* priv field value to indicates that subsequent fields are valid. */
>  #define V4L2_PIX_FMT_PRIV_MAGIC		0xfeedcafe
>  
> @@ -1961,6 +1964,9 @@ enum v4l2_ctrl_type {
>  	V4L2_CTRL_TYPE_AV1_TILE_GROUP_ENTRY = 0x281,
>  	V4L2_CTRL_TYPE_AV1_FRAME	    = 0x282,
>  	V4L2_CTRL_TYPE_AV1_FILM_GRAIN	    = 0x283,
> +
> +	V4L2_CTRL_TYPE_AIE_INIT		= 0x0290,
> +	V4L2_CTRL_TYPE_AIE_PARAM	= 0x0291,
>  };
>  
>  /*  Used in the VIDIOC_QUERYCTRL ioctl for querying controls */