media: hantro: Add support for VP8 decoding on rk3288

Introduce VP8 decoding support in RK3288.
Signed-off-by: ZhiChao Yu <zhichao.yu@rock-chips.com>
Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

drivers/staging/media/hantro/Makefile

@@ -5,10 +5,12 @@ hantro-vpu-y += \
 		hantro_v4l2.o \
 		hantro_h1_jpeg_enc.o \
 		hantro_g1_mpeg2_dec.o \
+		hantro_g1_vp8_dec.o \
 		rk3399_vpu_hw_jpeg_enc.o \
 		rk3399_vpu_hw_mpeg2_dec.o \
 		hantro_jpeg.o \
-		hantro_mpeg2.o
+		hantro_mpeg2.o \
+		hantro_vp8.o
 
 hantro-vpu-$(CONFIG_VIDEO_HANTRO_ROCKCHIP) += \
 		rk3288_vpu_hw.o \

drivers/staging/media/hantro/hantro.h

@@ -25,6 +25,10 @@
 
 #include "hantro_hw.h"
 
+#define VP8_MB_DIM			16
+#define VP8_MB_WIDTH(w)			DIV_ROUND_UP(w, VP8_MB_DIM)
+#define VP8_MB_HEIGHT(h)		DIV_ROUND_UP(h, VP8_MB_DIM)
+
 #define MPEG2_MB_DIM			16
 #define MPEG2_MB_WIDTH(w)		DIV_ROUND_UP(w, MPEG2_MB_DIM)
 #define MPEG2_MB_HEIGHT(h)		DIV_ROUND_UP(h, MPEG2_MB_DIM)
@@ -40,6 +44,7 @@ struct hantro_codec_ops;
 #define HANTRO_ENCODERS		0x0000ffff
 
 #define HANTRO_MPEG2_DECODER	BIT(16)
+#define HANTRO_VP8_DECODER	BIT(17)
 #define HANTRO_DECODERS		0xffff0000
 
 /**
@@ -97,11 +102,13 @@ struct hantro_variant {
  * @HANTRO_MODE_NONE:  No operating mode. Used for RAW video formats.
  * @HANTRO_MODE_JPEG_ENC: JPEG encoder.
  * @HANTRO_MODE_MPEG2_DEC: MPEG-2 decoder.
+ * @HANTRO_MODE_VP8_DEC: VP8 decoder.
  */
 enum hantro_codec_mode {
 	HANTRO_MODE_NONE = -1,
 	HANTRO_MODE_JPEG_ENC,
 	HANTRO_MODE_MPEG2_DEC,
+	HANTRO_MODE_VP8_DEC,
 };
 
 /*
@@ -215,6 +222,7 @@ struct hantro_dev {
  * @codec_ops:		Set of operations related to codec mode.
  * @jpeg_enc:		JPEG-encoding context.
  * @mpeg2_dec:		MPEG-2-decoding context.
+ * @vp8_dec:		VP8-decoding context.
  */
 struct hantro_ctx {
 	struct hantro_dev *dev;
@@ -241,6 +249,7 @@ struct hantro_ctx {
 	union {
 		struct hantro_jpeg_enc_hw_ctx jpeg_enc;
 		struct hantro_mpeg2_dec_hw_ctx mpeg2_dec;
+		struct hantro_vp8_dec_hw_ctx vp8_dec;
 	};
 };
 
@@ -265,6 +274,12 @@ struct hantro_fmt {
 	struct v4l2_frmsize_stepwise frmsize;
 };
 
+struct hantro_reg {
+	u32 base;
+	u32 shift;
+	u32 mask;
+};
+
 /* Logging helpers */
 
 /**
@@ -343,6 +358,18 @@ static inline u32 vdpu_read(struct hantro_dev *vpu, u32 reg)
 	return val;
 }
 
+static inline void hantro_reg_write(struct hantro_dev *vpu,
+				    const struct hantro_reg *reg,
+				    u32 val)
+{
+	u32 v;
+
+	v = vdpu_read(vpu, reg->base);
+	v &= ~(reg->mask << reg->shift);
+	v |= ((val & reg->mask) << reg->shift);
+	vdpu_write_relaxed(vpu, v, reg->base);
+}
+
 bool hantro_is_encoder_ctx(const struct hantro_ctx *ctx);
 
 void *hantro_get_ctrl(struct hantro_ctx *ctx, u32 id);

drivers/staging/media/hantro/hantro_drv.c

@@ -284,6 +284,12 @@ static struct hantro_ctrl controls[] = {
 		.cfg = {
 			.elem_size = sizeof(struct v4l2_ctrl_mpeg2_quantization),
 		},
+	}, {
+		.id = V4L2_CID_MPEG_VIDEO_VP8_FRAME_HEADER,
+		.codec = HANTRO_VP8_DECODER,
+		.cfg = {
+			.elem_size = sizeof(struct v4l2_ctrl_vp8_frame_header),
+		},
 	},
 };
 

drivers/staging/media/hantro/hantro_hw.h

@@ -12,6 +12,7 @@
 #include <linux/interrupt.h>
 #include <linux/v4l2-controls.h>
 #include <media/mpeg2-ctrls.h>
+#include <media/vp8-ctrls.h>
 #include <media/videobuf2-core.h>
 
 struct hantro_dev;
@@ -47,6 +48,16 @@ struct hantro_mpeg2_dec_hw_ctx {
 	struct hantro_aux_buf qtable;
 };
 
+/**
+ * struct hantro_vp8d_hw_ctx
+ * @segment_map:	Segment map buffer.
+ * @prob_tbl:		Probability table buffer.
+ */
+struct hantro_vp8_dec_hw_ctx {
+	struct hantro_aux_buf segment_map;
+	struct hantro_aux_buf prob_tbl;
+};
+
 /**
  * struct hantro_codec_ops - codec mode specific operations
  *
@@ -99,4 +110,10 @@ void hantro_mpeg2_dec_copy_qtable(u8 *qtable,
 int hantro_mpeg2_dec_init(struct hantro_ctx *ctx);
 void hantro_mpeg2_dec_exit(struct hantro_ctx *ctx);
 
+void hantro_g1_vp8_dec_run(struct hantro_ctx *ctx);
+int hantro_vp8_dec_init(struct hantro_ctx *ctx);
+void hantro_vp8_dec_exit(struct hantro_ctx *ctx);
+void hantro_vp8_prob_update(struct hantro_ctx *ctx,
+			    const struct v4l2_ctrl_vp8_frame_header *hdr);
+
 #endif /* HANTRO_HW_H_ */

drivers/staging/media/hantro/hantro_v4l2.c

@@ -344,6 +344,7 @@ hantro_update_requires_request(struct hantro_ctx *ctx, u32 fourcc)
 		ctx->fh.m2m_ctx->out_q_ctx.q.requires_requests = false;
 		break;
 	case V4L2_PIX_FMT_MPEG2_SLICE:
+	case V4L2_PIX_FMT_VP8_FRAME:
 		ctx->fh.m2m_ctx->out_q_ctx.q.requires_requests = true;
 		break;
 	default:

drivers/staging/media/hantro/hantro_vp8.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Hantro VPU codec driver
+ *
+ * Copyright (C) 2018 Rockchip Electronics Co., Ltd.
+ */
+
+#include "hantro.h"
+
+/*
+ * probs table with packed
+ */
+struct vp8_prob_tbl_packed {
+	u8 prob_mb_skip_false;
+	u8 prob_intra;
+	u8 prob_ref_last;
+	u8 prob_ref_golden;
+	u8 prob_segment[3];
+	u8 padding0;
+
+	u8 prob_luma_16x16_pred_mode[4];
+	u8 prob_chroma_pred_mode[3];
+	u8 padding1;
+
+	/* mv prob */
+	u8 prob_mv_context[2][19];
+	u8 padding2[2];
+
+	/* coeff probs */
+	u8 prob_coeffs[4][8][3][11];
+	u8 padding3[96];
+};
+
+void hantro_vp8_prob_update(struct hantro_ctx *ctx,
+			    const struct v4l2_ctrl_vp8_frame_header *hdr)
+{
+	const struct v4l2_vp8_entropy_header *entropy = &hdr->entropy_header;
+	u32 i, j, k;
+	u8 *dst;
+
+	/* first probs */
+	dst = ctx->vp8_dec.prob_tbl.cpu;
+
+	dst[0] = hdr->prob_skip_false;
+	dst[1] = hdr->prob_intra;
+	dst[2] = hdr->prob_last;
+	dst[3] = hdr->prob_gf;
+	dst[4] = hdr->segment_header.segment_probs[0];
+	dst[5] = hdr->segment_header.segment_probs[1];
+	dst[6] = hdr->segment_header.segment_probs[2];
+	dst[7] = 0;
+
+	dst += 8;
+	dst[0] = entropy->y_mode_probs[0];
+	dst[1] = entropy->y_mode_probs[1];
+	dst[2] = entropy->y_mode_probs[2];
+	dst[3] = entropy->y_mode_probs[3];
+	dst[4] = entropy->uv_mode_probs[0];
+	dst[5] = entropy->uv_mode_probs[1];
+	dst[6] = entropy->uv_mode_probs[2];
+	dst[7] = 0; /*unused */
+
+	/* mv probs */
+	dst += 8;
+	dst[0] = entropy->mv_probs[0][0]; /* is short */
+	dst[1] = entropy->mv_probs[1][0];
+	dst[2] = entropy->mv_probs[0][1]; /* sign */
+	dst[3] = entropy->mv_probs[1][1];
+	dst[4] = entropy->mv_probs[0][8 + 9];
+	dst[5] = entropy->mv_probs[0][9 + 9];
+	dst[6] = entropy->mv_probs[1][8 + 9];
+	dst[7] = entropy->mv_probs[1][9 + 9];
+	dst += 8;
+	for (i = 0; i < 2; ++i) {
+		for (j = 0; j < 8; j += 4) {
+			dst[0] = entropy->mv_probs[i][j + 9 + 0];
+			dst[1] = entropy->mv_probs[i][j + 9 + 1];
+			dst[2] = entropy->mv_probs[i][j + 9 + 2];
+			dst[3] = entropy->mv_probs[i][j + 9 + 3];
+			dst += 4;
+		}
+	}
+	for (i = 0; i < 2; ++i) {
+		dst[0] = entropy->mv_probs[i][0 + 2];
+		dst[1] = entropy->mv_probs[i][1 + 2];
+		dst[2] = entropy->mv_probs[i][2 + 2];
+		dst[3] = entropy->mv_probs[i][3 + 2];
+		dst[4] = entropy->mv_probs[i][4 + 2];
+		dst[5] = entropy->mv_probs[i][5 + 2];
+		dst[6] = entropy->mv_probs[i][6 + 2];
+		dst[7] = 0;	/*unused */
+		dst += 8;
+	}
+
+	/* coeff probs (header part) */
+	dst = ctx->vp8_dec.prob_tbl.cpu;
+	dst += (8 * 7);
+	for (i = 0; i < 4; ++i) {
+		for (j = 0; j < 8; ++j) {
+			for (k = 0; k < 3; ++k) {
+				dst[0] = entropy->coeff_probs[i][j][k][0];
+				dst[1] = entropy->coeff_probs[i][j][k][1];
+				dst[2] = entropy->coeff_probs[i][j][k][2];
+				dst[3] = entropy->coeff_probs[i][j][k][3];
+				dst += 4;
+			}
+		}
+	}
+
+	/* coeff probs (footer part) */
+	dst = ctx->vp8_dec.prob_tbl.cpu;
+	dst += (8 * 55);
+	for (i = 0; i < 4; ++i) {
+		for (j = 0; j < 8; ++j) {
+			for (k = 0; k < 3; ++k) {
+				dst[0] = entropy->coeff_probs[i][j][k][4];
+				dst[1] = entropy->coeff_probs[i][j][k][5];
+				dst[2] = entropy->coeff_probs[i][j][k][6];
+				dst[3] = entropy->coeff_probs[i][j][k][7];
+				dst[4] = entropy->coeff_probs[i][j][k][8];
+				dst[5] = entropy->coeff_probs[i][j][k][9];
+				dst[6] = entropy->coeff_probs[i][j][k][10];
+				dst[7] = 0;	/*unused */
+				dst += 8;
+			}
+		}
+	}
+}
+
+int hantro_vp8_dec_init(struct hantro_ctx *ctx)
+{
+	struct hantro_dev *vpu = ctx->dev;
+	struct hantro_aux_buf *aux_buf;
+	unsigned int mb_width, mb_height;
+	size_t segment_map_size;
+	int ret;
+
+	/* segment map table size calculation */
+	mb_width = DIV_ROUND_UP(ctx->dst_fmt.width, 16);
+	mb_height = DIV_ROUND_UP(ctx->dst_fmt.height, 16);
+	segment_map_size = round_up(DIV_ROUND_UP(mb_width * mb_height, 4), 64);
+
+	/*
+	 * In context init the dma buffer for segment map must be allocated.
+	 * And the data in segment map buffer must be set to all zero.
+	 */
+	aux_buf = &ctx->vp8_dec.segment_map;
+	aux_buf->size = segment_map_size;
+	aux_buf->cpu = dma_alloc_coherent(vpu->dev, aux_buf->size,
+					  &aux_buf->dma, GFP_KERNEL);
+	if (!aux_buf->cpu)
+		return -ENOMEM;
+
+	memset(aux_buf->cpu, 0, aux_buf->size);
+
+	/*
+	 * Allocate probability table buffer,
+	 * total 1208 bytes, 4K page is far enough.
+	 */
+	aux_buf = &ctx->vp8_dec.prob_tbl;
+	aux_buf->size = sizeof(struct vp8_prob_tbl_packed);
+	aux_buf->cpu = dma_alloc_coherent(vpu->dev, aux_buf->size,
+					  &aux_buf->dma, GFP_KERNEL);
+	if (!aux_buf->cpu) {
+		ret = -ENOMEM;
+		goto err_free_seg_map;
+	}
+
+	return 0;
+
+err_free_seg_map:
+	dma_free_coherent(vpu->dev, ctx->vp8_dec.segment_map.size,
+			  ctx->vp8_dec.segment_map.cpu,
+			  ctx->vp8_dec.segment_map.dma);
+
+	return ret;
+}
+
+void hantro_vp8_dec_exit(struct hantro_ctx *ctx)
+{
+	struct hantro_vp8_dec_hw_ctx *vp8_dec = &ctx->vp8_dec;
+	struct hantro_dev *vpu = ctx->dev;
+
+	dma_free_coherent(vpu->dev, vp8_dec->segment_map.size,
+			  vp8_dec->segment_map.cpu, vp8_dec->segment_map.dma);
+	dma_free_coherent(vpu->dev, vp8_dec->prob_tbl.size,
+			  vp8_dec->prob_tbl.cpu, vp8_dec->prob_tbl.dma);
+}

drivers/staging/media/hantro/rk3288_vpu_hw.c

@@ -74,6 +74,19 @@ static const struct hantro_fmt rk3288_vpu_dec_fmts[] = {
 			.step_height = MPEG2_MB_DIM,
 		},
 	},
+	{
+		.fourcc = V4L2_PIX_FMT_VP8_FRAME,
+		.codec_mode = HANTRO_MODE_VP8_DEC,
+		.max_depth = 2,
+		.frmsize = {
+			.min_width = 48,
+			.max_width = 3840,
+			.step_width = VP8_MB_DIM,
+			.min_height = 48,
+			.max_height = 2160,
+			.step_height = VP8_MB_DIM,
+		},
+	},
 };
 
 static irqreturn_t rk3288_vepu_irq(int irq, void *dev_id)
@@ -155,6 +168,12 @@ static const struct hantro_codec_ops rk3288_vpu_codec_ops[] = {
 		.init = hantro_mpeg2_dec_init,
 		.exit = hantro_mpeg2_dec_exit,
 	},
+	[HANTRO_MODE_VP8_DEC] = {
+		.run = hantro_g1_vp8_dec_run,
+		.reset = rk3288_vpu_dec_reset,
+		.init = hantro_vp8_dec_init,
+		.exit = hantro_vp8_dec_exit,
+	},
 };
 
 /*
@@ -177,7 +196,8 @@ const struct hantro_variant rk3288_vpu_variant = {
 	.dec_offset = 0x400,
 	.dec_fmts = rk3288_vpu_dec_fmts,
 	.num_dec_fmts = ARRAY_SIZE(rk3288_vpu_dec_fmts),
-	.codec = HANTRO_JPEG_ENCODER | HANTRO_MPEG2_DECODER,
+	.codec = HANTRO_JPEG_ENCODER | HANTRO_MPEG2_DECODER |
+		 HANTRO_VP8_DECODER,
 	.codec_ops = rk3288_vpu_codec_ops,
 	.irqs = rk3288_irqs,
 	.num_irqs = ARRAY_SIZE(rk3288_irqs),