Le jeudi 31 mars 2022 à 10:48 +0800, Yunfei Dong a écrit :
> Add support for VP9 decoding using the stateless API,
> as supported by MT8192. And the drivers is lat and core architecture.
>
> Signed-off-by: George Sun <george.sun@xxxxxxxxxxxx>
> Signed-off-by: Xiaoyong Lu <xiaoyong.lu@xxxxxxxxxxxx>
> Signed-off-by: Yunfei Dong <yunfei.dong@xxxxxxxxxxxx>
> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@xxxxxxxxxxxxx>
> ---
> changed compare with v7:
> Using upstream interface to update vp9 prob tables.
> ---
> .../media/platform/mediatek/vcodec/Makefile | 1 +
> .../vcodec/mtk_vcodec_dec_stateless.c | 26 +-
> .../platform/mediatek/vcodec/mtk_vcodec_drv.h | 1 +
> .../vcodec/vdec/vdec_vp9_req_lat_if.c | 2072 +++++++++++++++++
> .../platform/mediatek/vcodec/vdec_drv_if.c | 4 +
> .../platform/mediatek/vcodec/vdec_drv_if.h | 1 +
> 6 files changed, 2102 insertions(+), 3 deletions(-)
> create mode 100644 drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c
>
> diff --git a/drivers/media/platform/mediatek/vcodec/Makefile b/drivers/media/platform/mediatek/vcodec/Makefile
> index b457daf2d196..93e7a343b5b0 100644
> --- a/drivers/media/platform/mediatek/vcodec/Makefile
> +++ b/drivers/media/platform/mediatek/vcodec/Makefile
> @@ -9,6 +9,7 @@ mtk-vcodec-dec-y := vdec/vdec_h264_if.o \
> vdec/vdec_vp8_if.o \
> vdec/vdec_vp8_req_if.o \
> vdec/vdec_vp9_if.o \
> + vdec/vdec_vp9_req_lat_if.o \
> vdec/vdec_h264_req_if.o \
> vdec/vdec_h264_req_common.o \
> vdec/vdec_h264_req_multi_if.o \
> diff --git a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec_stateless.c b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec_stateless.c
> index 3208f834ff80..a4735e67d39e 100644
> --- a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec_stateless.c
> +++ b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec_stateless.c
> @@ -91,13 +91,28 @@ static const struct mtk_stateless_control mtk_stateless_controls[] = {
> .max = V4L2_MPEG_VIDEO_VP8_PROFILE_3,
> },
> .codec_type = V4L2_PIX_FMT_VP8_FRAME,
> - }
> + },
> + {
> + .cfg = {
> + .id = V4L2_CID_STATELESS_VP9_FRAME,
> + },
> + .codec_type = V4L2_PIX_FMT_VP9_FRAME,
> + },
> + {
> + .cfg = {
> + .id = V4L2_CID_MPEG_VIDEO_VP9_PROFILE,
> + .min = V4L2_MPEG_VIDEO_VP9_PROFILE_0,
> + .def = V4L2_MPEG_VIDEO_VP9_PROFILE_0,
> + .max = V4L2_MPEG_VIDEO_VP9_PROFILE_3,
> + },
> + .codec_type = V4L2_PIX_FMT_VP9_FRAME,
> + },
> };
>
> #define NUM_CTRLS ARRAY_SIZE(mtk_stateless_controls)
>
> -static struct mtk_video_fmt mtk_video_formats[4];
> -static struct mtk_codec_framesizes mtk_vdec_framesizes[2];
> +static struct mtk_video_fmt mtk_video_formats[5];
> +static struct mtk_codec_framesizes mtk_vdec_framesizes[3];
>
> static struct mtk_video_fmt default_out_format;
> static struct mtk_video_fmt default_cap_format;
> @@ -338,6 +353,7 @@ static void mtk_vcodec_add_formats(unsigned int fourcc,
> switch (fourcc) {
> case V4L2_PIX_FMT_H264_SLICE:
> case V4L2_PIX_FMT_VP8_FRAME:
> + case V4L2_PIX_FMT_VP9_FRAME:
> mtk_video_formats[count_formats].fourcc = fourcc;
> mtk_video_formats[count_formats].type = MTK_FMT_DEC;
> mtk_video_formats[count_formats].num_planes = 1;
> @@ -385,6 +401,10 @@ static void mtk_vcodec_get_supported_formats(struct mtk_vcodec_ctx *ctx)
> mtk_vcodec_add_formats(V4L2_PIX_FMT_VP8_FRAME, ctx);
> out_format_count++;
> }
> + if (ctx->dev->dec_capability & MTK_VDEC_FORMAT_VP9_FRAME) {
> + mtk_vcodec_add_formats(V4L2_PIX_FMT_VP9_FRAME, ctx);
> + out_format_count++;
> + }
>
> if (cap_format_count)
> default_cap_format = mtk_video_formats[cap_format_count - 1];
> diff --git a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> index 2ba1c19f07b6..a29041a0b7e0 100644
> --- a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> +++ b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> @@ -355,6 +355,7 @@ enum mtk_vdec_format_types {
> MTK_VDEC_FORMAT_MT21C = 0x40,
> MTK_VDEC_FORMAT_H264_SLICE = 0x100,
> MTK_VDEC_FORMAT_VP8_FRAME = 0x200,
> + MTK_VDEC_FORMAT_VP9_FRAME = 0x400,
> };
>
> /**
> diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c
> new file mode 100644
> index 000000000000..d63399085b9b
> --- /dev/null
> +++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c
> @@ -0,0 +1,2072 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (c) 2021 MediaTek Inc.
> + * Author: George Sun <george.sun@xxxxxxxxxxxx>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/slab.h>
> +#include <media/videobuf2-dma-contig.h>
> +#include <media/v4l2-vp9.h>
> +
> +#include "../mtk_vcodec_util.h"
> +#include "../mtk_vcodec_dec.h"
> +#include "../mtk_vcodec_intr.h"
> +#include "../vdec_drv_base.h"
> +#include "../vdec_drv_if.h"
> +#include "../vdec_vpu_if.h"
> +
> +/* reset_frame_context defined in VP9 spec */
> +#define VP9_RESET_FRAME_CONTEXT_NONE0 0
> +#define VP9_RESET_FRAME_CONTEXT_NONE1 1
> +#define VP9_RESET_FRAME_CONTEXT_SPEC 2
> +#define VP9_RESET_FRAME_CONTEXT_ALL 3
> +
> +#define VP9_TILE_BUF_SIZE 4096
> +#define VP9_PROB_BUF_SIZE 2560
> +#define VP9_COUNTS_BUF_SIZE 16384
> +
> +#define HDR_FLAG(x) (!!((hdr)->flags & V4L2_VP9_FRAME_FLAG_##x))
> +#define LF_FLAG(x) (!!((lf)->flags & V4L2_VP9_LOOP_FILTER_FLAG_##x))
> +#define SEG_FLAG(x) (!!((seg)->flags & V4L2_VP9_SEGMENTATION_FLAG_##x))
> +#define VP9_BAND_6(band) ((band) == 0 ? 3 : 6)
> +
> +/*
> + * struct vdec_vp9_slice_frame_ctx - vp9 prob tables footprint
> + */
> +struct vdec_vp9_slice_frame_ctx {
> + struct {
> + u8 probs[6][3];
> + u8 padding[2];
> + } coef_probs[4][2][2][6];
> +
> + u8 y_mode_prob[4][16];
> + u8 switch_interp_prob[4][16];
> + u8 seg[32]; /* ignore */
> + u8 comp_inter_prob[16];
> + u8 comp_ref_prob[16];
> + u8 single_ref_prob[5][2];
> + u8 single_ref_prob_padding[6];
> +
> + u8 joint[3];
> + u8 joint_padding[13];
> + struct {
> + u8 sign;
> + u8 classes[10];
> + u8 padding[5];
> + } sign_classes[2];
> + struct {
> + u8 class0[1];
> + u8 bits[10];
> + u8 padding[5];
> + } class0_bits[2];
> + struct {
> + u8 class0_fp[2][3];
> + u8 fp[3];
> + u8 class0_hp;
> + u8 hp;
> + u8 padding[5];
> + } class0_fp_hp[2];
> +
> + u8 uv_mode_prob[10][16];
> + u8 uv_mode_prob_padding[2][16];
> +
> + u8 partition_prob[16][4];
> +
> + u8 inter_mode_probs[7][4];
> + u8 skip_probs[4];
> +
> + u8 tx_p8x8[2][4];
> + u8 tx_p16x16[2][4];
> + u8 tx_p32x32[2][4];
> + u8 intra_inter_prob[8];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_frame_counts - vp9 counts tables footprint
> + */
> +struct vdec_vp9_slice_frame_counts {
> + union {
> + struct {
> + u32 band_0[3];
> + u32 padding0[1];
> + u32 band_1_5[5][6];
> + u32 padding1[2];
> + } eob_branch[4][2][2];
> + u32 eob_branch_space[256 * 4];
> + };
> +
> + struct {
> + u32 band_0[3][4];
> + u32 band_1_5[5][6][4];
> + } coef_probs[4][2][2];
> +
> + u32 intra_inter[4][2];
> + u32 comp_inter[5][2];
> + u32 comp_inter_padding[2];
> + u32 comp_ref[5][2];
> + u32 comp_ref_padding[2];
> + u32 single_ref[5][2][2];
> + u32 inter_mode[7][4];
> + u32 y_mode[4][12];
> + u32 uv_mode[10][10];
> + u32 partition[16][4];
> + u32 switchable_interp[4][4];
> +
> + u32 tx_p8x8[2][2];
> + u32 tx_p16x16[2][4];
> + u32 tx_p32x32[2][4];
> +
> + u32 skip[3][4];
> +
> + u32 joint[4];
> +
> + struct {
> + u32 sign[2];
> + u32 class0[2];
> + u32 classes[12];
> + u32 bits[10][2];
> + u32 padding[4];
> + u32 class0_fp[2][4];
> + u32 fp[4];
> + u32 class0_hp[2];
> + u32 hp[2];
> + } mvcomp[2];
> +
> + u32 reserved[126][4];
> +};
> +
> +/**
> + * struct vdec_vp9_slice_counts_map - vp9 counts tables to map
> + * v4l2_vp9_frame_symbol_counts
> + * @skip: skip counts.
> + * @y_mode: Y prediction mode counts.
> + * @filter: interpolation filter counts.
> + * @mv_joint: motion vector joint counts.
> + * @sign: motion vector sign counts.
> + * @classes: motion vector class counts.
> + * @class0: motion vector class0 bit counts.
> + * @bits: motion vector bits counts.
> + * @class0_fp: motion vector class0 fractional bit counts.
> + * @fp: motion vector fractional bit counts.
> + * @class0_hp: motion vector class0 high precision fractional bit counts.
> + * @hp: motion vector high precision fractional bit counts.
> + */
> +struct vdec_vp9_slice_counts_map {
> + u32 skip[3][2];
> + u32 y_mode[4][10];
> + u32 filter[4][3];
> + u32 sign[2][2];
> + u32 classes[2][11];
> + u32 class0[2][2];
> + u32 bits[2][10][2];
> + u32 class0_fp[2][2][4];
> + u32 fp[2][4];
> + u32 class0_hp[2][2];
> + u32 hp[2][2];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_uncompressed_header - vp9 uncompressed header syntax
> + * used for decoding
> + */
> +struct vdec_vp9_slice_uncompressed_header {
> + u8 profile;
> + u8 last_frame_type;
> + u8 frame_type;
> +
> + u8 last_show_frame;
> + u8 show_frame;
> + u8 error_resilient_mode;
> +
> + u8 bit_depth;
> + u8 padding0[1];
> + u16 last_frame_width;
> + u16 last_frame_height;
> + u16 frame_width;
> + u16 frame_height;
> +
> + u8 intra_only;
> + u8 reset_frame_context;
> + u8 ref_frame_sign_bias[4];
> + u8 allow_high_precision_mv;
> + u8 interpolation_filter;
> +
> + u8 refresh_frame_context;
> + u8 frame_parallel_decoding_mode;
> + u8 frame_context_idx;
> +
> + /* loop_filter_params */
> + u8 loop_filter_level;
> + u8 loop_filter_sharpness;
> + u8 loop_filter_delta_enabled;
> + s8 loop_filter_ref_deltas[4];
> + s8 loop_filter_mode_deltas[2];
> +
> + /* quantization_params */
> + u8 base_q_idx;
> + s8 delta_q_y_dc;
> + s8 delta_q_uv_dc;
> + s8 delta_q_uv_ac;
> +
> + /* segmentation_params */
> + u8 segmentation_enabled;
> + u8 segmentation_update_map;
> + u8 segmentation_tree_probs[7];
> + u8 padding1[1];
> + u8 segmentation_temporal_udpate;
> + u8 segmentation_pred_prob[3];
> + u8 segmentation_update_data;
> + u8 segmentation_abs_or_delta_update;
> + u8 feature_enabled[8];
> + s16 feature_value[8][4];
> +
> + /* tile_info */
> + u8 tile_cols_log2;
> + u8 tile_rows_log2;
> + u8 padding2[2];
> +
> + u16 uncompressed_header_size;
> + u16 header_size_in_bytes;
> +
> + /* LAT OUT, CORE IN */
> + u32 dequant[8][4];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_compressed_header - vp9 compressed header syntax
> + * used for decoding.
> + */
> +struct vdec_vp9_slice_compressed_header {
> + u8 tx_mode;
> + u8 ref_mode;
> + u8 comp_fixed_ref;
> + u8 comp_var_ref[2];
> + u8 padding[3];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_tiles - vp9 tile syntax
> + */
> +struct vdec_vp9_slice_tiles {
> + u32 size[4][64];
> + u32 mi_rows[4];
> + u32 mi_cols[64];
> + u8 actual_rows;
> + u8 padding[7];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_reference - vp9 reference frame information
> + */
> +struct vdec_vp9_slice_reference {
> + u16 frame_width;
> + u16 frame_height;
> + u8 bit_depth;
> + u8 subsampling_x;
> + u8 subsampling_y;
> + u8 padding;
> +};
> +
> +/*
> + * struct vdec_vp9_slice_frame - vp9 syntax used for decoding
> + */
> +struct vdec_vp9_slice_frame {
> + struct vdec_vp9_slice_uncompressed_header uh;
> + struct vdec_vp9_slice_compressed_header ch;
> + struct vdec_vp9_slice_tiles tiles;
> + struct vdec_vp9_slice_reference ref[3];
> +};
> +
> +/*
> + * struct vdec_vp9_slice_init_vsi - VSI used to initialize instance
> + */
> +struct vdec_vp9_slice_init_vsi {
> + unsigned int architecture;
> + unsigned int reserved;
> + u64 core_vsi;
> + /* default frame context's position in MicroP */
> + u64 default_frame_ctx;
> +};
> +
> +/*
> + * struct vdec_vp9_slice_mem - memory address and size
> + */
> +struct vdec_vp9_slice_mem {
> + union {
> + u64 buf;
> + dma_addr_t dma_addr;
> + };
> + union {
> + size_t size;
> + dma_addr_t dma_addr_end;
> + u64 padding;
> + };
> +};
> +
> +/*
> + * struct vdec_vp9_slice_bs - input buffer for decoding
> + */
> +struct vdec_vp9_slice_bs {
> + struct vdec_vp9_slice_mem buf;
> + struct vdec_vp9_slice_mem frame;
> +};
> +
> +/*
> + * struct vdec_vp9_slice_fb - frame buffer for decoding
> + */
> +struct vdec_vp9_slice_fb {
> + struct vdec_vp9_slice_mem y;
> + struct vdec_vp9_slice_mem c;
> +};
> +
> +/*
> + * struct vdec_vp9_slice_state - decoding state
> + */
> +struct vdec_vp9_slice_state {
> + int err;
> + unsigned int full;
> + unsigned int timeout;
> + unsigned int perf;
> +
> + unsigned int crc[12];
> +};
> +
> +/**
> + * struct vdec_vp9_slice_vsi - exchange decoding information
> + * between Main CPU and MicroP
> + *
> + * @bs: input buffer
> + * @fb: output buffer
> + * @ref: 3 reference buffers
> + * @mv: mv working buffer
> + * @seg: segmentation working buffer
> + * @tile: tile buffer
> + * @prob: prob table buffer, used to set/update prob table
> + * @counts: counts table buffer, used to update prob table
> + * @ube: general buffer
> + * @trans: trans buffer position in general buffer
> + * @err_map: error buffer
> + * @row_info: row info buffer
> + * @frame: decoding syntax
> + * @state: decoding state
> + */
> +struct vdec_vp9_slice_vsi {
> + /* used in LAT stage */
> + struct vdec_vp9_slice_bs bs;
> + /* used in Core stage */
> + struct vdec_vp9_slice_fb fb;
> + struct vdec_vp9_slice_fb ref[3];
> +
> + struct vdec_vp9_slice_mem mv[2];
> + struct vdec_vp9_slice_mem seg[2];
> + struct vdec_vp9_slice_mem tile;
> + struct vdec_vp9_slice_mem prob;
> + struct vdec_vp9_slice_mem counts;
> +
> + /* LAT stage's output, Core stage's input */
> + struct vdec_vp9_slice_mem ube;
> + struct vdec_vp9_slice_mem trans;
> + struct vdec_vp9_slice_mem err_map;
> + struct vdec_vp9_slice_mem row_info;
> +
> + /* decoding parameters */
> + struct vdec_vp9_slice_frame frame;
> +
> + struct vdec_vp9_slice_state state;
> +};
> +
> +/**
> + * struct vdec_vp9_slice_pfc - per-frame context that contains a local vsi.
> + * pass it from lat to core
> + *
> + * @vsi: local vsi. copy to/from remote vsi before/after decoding
> + * @ref_idx: reference buffer index
> + * @seq: picture sequence
> + * @state: decoding state
> + */
> +struct vdec_vp9_slice_pfc {
> + struct vdec_vp9_slice_vsi vsi;
> +
> + u64 ref_idx[3];
> +
> + int seq;
> +
> + /* LAT/Core CRC */
> + struct vdec_vp9_slice_state state[2];
> +};
> +
> +/*
> + * enum vdec_vp9_slice_resolution_level
> + */
> +enum vdec_vp9_slice_resolution_level {
> + VP9_RES_NONE,
> + VP9_RES_FHD,
> + VP9_RES_4K,
> + VP9_RES_8K,
> +};
> +
> +/*
> + * struct vdec_vp9_slice_ref - picture's width & height should kept
> + * for later decoding as reference picture
> + */
> +struct vdec_vp9_slice_ref {
> + unsigned int width;
> + unsigned int height;
> +};
> +
> +/**
> + * struct vdec_vp9_slice_instance - represent one vp9 instance
> + *
> + * @ctx: pointer to codec's context
> + * @vpu: VPU instance
> + * @seq: global picture sequence
> + * @level: level of current resolution
> + * @width: width of last picture
> + * @height: height of last picture
> + * @frame_type: frame_type of last picture
> + * @irq: irq to Main CPU or MicroP
> + * @show_frame: show_frame of last picture
> + * @dpb: picture information (width/height) for reference
> + * @mv: mv working buffer
> + * @seg: segmentation working buffer
> + * @tile: tile buffer
> + * @prob: prob table buffer, used to set/update prob table
> + * @counts: counts table buffer, used to update prob table
> + * @frame_ctx: 4 frame context according to VP9 Spec
> + * @frame_ctx_helper: 4 frame context according to newest kernel spec
> + * @dirty: state of each frame context
> + * @init_vsi: vsi used for initialized VP9 instance
> + * @vsi: vsi used for decoding/flush ...
> + * @core_vsi: vsi used for Core stage
> + * @counts_map: used map to counts_helper
> + * &counts_helper: counts table according to newest kernel spec
Change & into @ to fix:
drivers/media/platform/mediatek/vcodec/vdec/vdec_vp9_req_lat_if.c:492: warning: Function parameter or member 'counts_helper' not described in 'vdec_vp9_slice_instance'
> + */
> +struct vdec_vp9_slice_instance {
> + struct mtk_vcodec_ctx *ctx;
> + struct vdec_vpu_inst vpu;
> +
> + int seq;
> +
> + enum vdec_vp9_slice_resolution_level level;
> +
> + /* for resolution change and get_pic_info */
> + unsigned int width;
> + unsigned int height;
> +
> + /* for last_frame_type */
> + unsigned int frame_type;
> + unsigned int irq;
> +
> + unsigned int show_frame;
> +
> + /* maintain vp9 reference frame state */
> + struct vdec_vp9_slice_ref dpb[VB2_MAX_FRAME];
> +
> + /*
> + * normal working buffers
> + * mv[0]/seg[0]/tile/prob/counts is used for LAT
> + * mv[1]/seg[1] is used for CORE
> + */
> + struct mtk_vcodec_mem mv[2];
> + struct mtk_vcodec_mem seg[2];
> + struct mtk_vcodec_mem tile;
> + struct mtk_vcodec_mem prob;
> + struct mtk_vcodec_mem counts;
> +
> + /* 4 prob tables */
> + struct vdec_vp9_slice_frame_ctx frame_ctx[4];
> + /*4 helper tables */
> + struct v4l2_vp9_frame_context frame_ctx_helper;
> + unsigned char dirty[4];
> +
> + /* MicroP vsi */
> + union {
> + struct vdec_vp9_slice_init_vsi *init_vsi;
> + struct vdec_vp9_slice_vsi *vsi;
> + };
> + struct vdec_vp9_slice_vsi *core_vsi;
> +
> + struct vdec_vp9_slice_counts_map counts_map;
> + struct v4l2_vp9_frame_symbol_counts counts_helper;
> +};
