xref: /linux/drivers/media/platform/verisilicon/hantro_g1_h264_dec.c (revision 79790b6818e96c58fe2bffee1b418c16e64e7b80)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Rockchip RK3288 VPU codec driver
4  *
5  * Copyright (c) 2014 Rockchip Electronics Co., Ltd.
6  *	Hertz Wong <hertz.wong@rock-chips.com>
7  *	Herman Chen <herman.chen@rock-chips.com>
8  *
9  * Copyright (C) 2014 Google, Inc.
10  *	Tomasz Figa <tfiga@chromium.org>
11  */
12 
13 #include <linux/types.h>
14 #include <linux/sort.h>
15 
16 #include <media/v4l2-mem2mem.h>
17 
18 #include "hantro_g1_regs.h"
19 #include "hantro_hw.h"
20 #include "hantro_v4l2.h"
21 
set_params(struct hantro_ctx * ctx,struct vb2_v4l2_buffer * src_buf)22 static void set_params(struct hantro_ctx *ctx, struct vb2_v4l2_buffer *src_buf)
23 {
24 	const struct hantro_h264_dec_ctrls *ctrls = &ctx->h264_dec.ctrls;
25 	const struct v4l2_ctrl_h264_decode_params *dec_param = ctrls->decode;
26 	const struct v4l2_ctrl_h264_sps *sps = ctrls->sps;
27 	const struct v4l2_ctrl_h264_pps *pps = ctrls->pps;
28 	struct hantro_dev *vpu = ctx->dev;
29 	u32 reg;
30 
31 	/* Decoder control register 0. */
32 	reg = G1_REG_DEC_CTRL0_DEC_AXI_AUTO;
33 	if (sps->flags & V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD)
34 		reg |= G1_REG_DEC_CTRL0_SEQ_MBAFF_E;
35 	if (sps->profile_idc > 66) {
36 		reg |= G1_REG_DEC_CTRL0_PICORD_COUNT_E;
37 		if (dec_param->nal_ref_idc)
38 			reg |= G1_REG_DEC_CTRL0_WRITE_MVS_E;
39 	}
40 
41 	if (!(sps->flags & V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY) &&
42 	    (sps->flags & V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD ||
43 	     dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC))
44 		reg |= G1_REG_DEC_CTRL0_PIC_INTERLACE_E;
45 	if (dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC)
46 		reg |= G1_REG_DEC_CTRL0_PIC_FIELDMODE_E;
47 	if (!(dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD))
48 		reg |= G1_REG_DEC_CTRL0_PIC_TOPFIELD_E;
49 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL0);
50 
51 	/* Decoder control register 1. */
52 	reg = G1_REG_DEC_CTRL1_PIC_MB_WIDTH(MB_WIDTH(ctx->src_fmt.width)) |
53 	      G1_REG_DEC_CTRL1_PIC_MB_HEIGHT_P(MB_HEIGHT(ctx->src_fmt.height)) |
54 	      G1_REG_DEC_CTRL1_REF_FRAMES(sps->max_num_ref_frames);
55 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL1);
56 
57 	/* Decoder control register 2. */
58 	reg = G1_REG_DEC_CTRL2_CH_QP_OFFSET(pps->chroma_qp_index_offset) |
59 	      G1_REG_DEC_CTRL2_CH_QP_OFFSET2(pps->second_chroma_qp_index_offset);
60 
61 	if (pps->flags & V4L2_H264_PPS_FLAG_SCALING_MATRIX_PRESENT)
62 		reg |= G1_REG_DEC_CTRL2_TYPE1_QUANT_E;
63 	if (!(sps->flags & V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY))
64 		reg |= G1_REG_DEC_CTRL2_FIELDPIC_FLAG_E;
65 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL2);
66 
67 	/* Decoder control register 3. */
68 	reg = G1_REG_DEC_CTRL3_START_CODE_E |
69 	      G1_REG_DEC_CTRL3_INIT_QP(pps->pic_init_qp_minus26 + 26) |
70 	      G1_REG_DEC_CTRL3_STREAM_LEN(vb2_get_plane_payload(&src_buf->vb2_buf, 0));
71 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL3);
72 
73 	/* Decoder control register 4. */
74 	reg = G1_REG_DEC_CTRL4_FRAMENUM_LEN(sps->log2_max_frame_num_minus4 + 4) |
75 	      G1_REG_DEC_CTRL4_FRAMENUM(dec_param->frame_num) |
76 	      G1_REG_DEC_CTRL4_WEIGHT_BIPR_IDC(pps->weighted_bipred_idc);
77 	if (pps->flags & V4L2_H264_PPS_FLAG_ENTROPY_CODING_MODE)
78 		reg |= G1_REG_DEC_CTRL4_CABAC_E;
79 	if (sps->flags & V4L2_H264_SPS_FLAG_DIRECT_8X8_INFERENCE)
80 		reg |= G1_REG_DEC_CTRL4_DIR_8X8_INFER_E;
81 	if (sps->profile_idc >= 100 && sps->chroma_format_idc == 0)
82 		reg |= G1_REG_DEC_CTRL4_BLACKWHITE_E;
83 	if (pps->flags & V4L2_H264_PPS_FLAG_WEIGHTED_PRED)
84 		reg |= G1_REG_DEC_CTRL4_WEIGHT_PRED_E;
85 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL4);
86 
87 	/* Decoder control register 5. */
88 	reg = G1_REG_DEC_CTRL5_REFPIC_MK_LEN(dec_param->dec_ref_pic_marking_bit_size) |
89 	      G1_REG_DEC_CTRL5_IDR_PIC_ID(dec_param->idr_pic_id);
90 	if (pps->flags & V4L2_H264_PPS_FLAG_CONSTRAINED_INTRA_PRED)
91 		reg |= G1_REG_DEC_CTRL5_CONST_INTRA_E;
92 	if (pps->flags & V4L2_H264_PPS_FLAG_DEBLOCKING_FILTER_CONTROL_PRESENT)
93 		reg |= G1_REG_DEC_CTRL5_FILT_CTRL_PRES;
94 	if (pps->flags & V4L2_H264_PPS_FLAG_REDUNDANT_PIC_CNT_PRESENT)
95 		reg |= G1_REG_DEC_CTRL5_RDPIC_CNT_PRES;
96 	if (pps->flags & V4L2_H264_PPS_FLAG_TRANSFORM_8X8_MODE)
97 		reg |= G1_REG_DEC_CTRL5_8X8TRANS_FLAG_E;
98 	if (dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_IDR_PIC)
99 		reg |= G1_REG_DEC_CTRL5_IDR_PIC_E;
100 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL5);
101 
102 	/* Decoder control register 6. */
103 	reg = G1_REG_DEC_CTRL6_PPS_ID(pps->pic_parameter_set_id) |
104 	      G1_REG_DEC_CTRL6_REFIDX0_ACTIVE(pps->num_ref_idx_l0_default_active_minus1 + 1) |
105 	      G1_REG_DEC_CTRL6_REFIDX1_ACTIVE(pps->num_ref_idx_l1_default_active_minus1 + 1) |
106 	      G1_REG_DEC_CTRL6_POC_LENGTH(dec_param->pic_order_cnt_bit_size);
107 	vdpu_write_relaxed(vpu, reg, G1_REG_DEC_CTRL6);
108 
109 	/* Error concealment register. */
110 	vdpu_write_relaxed(vpu, 0, G1_REG_ERR_CONC);
111 
112 	/* Prediction filter tap register. */
113 	vdpu_write_relaxed(vpu,
114 			   G1_REG_PRED_FLT_PRED_BC_TAP_0_0(1) |
115 			   G1_REG_PRED_FLT_PRED_BC_TAP_0_1(-5 & 0x3ff) |
116 			   G1_REG_PRED_FLT_PRED_BC_TAP_0_2(20),
117 			   G1_REG_PRED_FLT);
118 
119 	/* Reference picture buffer control register. */
120 	vdpu_write_relaxed(vpu, 0, G1_REG_REF_BUF_CTRL);
121 
122 	/* Reference picture buffer control register 2. */
123 	vdpu_write_relaxed(vpu, G1_REG_REF_BUF_CTRL2_APF_THRESHOLD(8),
124 			   G1_REG_REF_BUF_CTRL2);
125 }
126 
set_ref(struct hantro_ctx * ctx)127 static void set_ref(struct hantro_ctx *ctx)
128 {
129 	const struct v4l2_h264_reference *b0_reflist, *b1_reflist, *p_reflist;
130 	struct hantro_dev *vpu = ctx->dev;
131 	int reg_num;
132 	u32 reg;
133 	int i;
134 
135 	vdpu_write_relaxed(vpu, ctx->h264_dec.dpb_valid, G1_REG_VALID_REF);
136 	vdpu_write_relaxed(vpu, ctx->h264_dec.dpb_longterm, G1_REG_LT_REF);
137 
138 	/*
139 	 * Set up reference frame picture numbers.
140 	 *
141 	 * Each G1_REG_REF_PIC(x) register contains numbers of two
142 	 * subsequential reference pictures.
143 	 */
144 	for (i = 0; i < HANTRO_H264_DPB_SIZE; i += 2) {
145 		reg = G1_REG_REF_PIC_REFER0_NBR(hantro_h264_get_ref_nbr(ctx, i)) |
146 		      G1_REG_REF_PIC_REFER1_NBR(hantro_h264_get_ref_nbr(ctx, i + 1));
147 		vdpu_write_relaxed(vpu, reg, G1_REG_REF_PIC(i / 2));
148 	}
149 
150 	b0_reflist = ctx->h264_dec.reflists.b0;
151 	b1_reflist = ctx->h264_dec.reflists.b1;
152 	p_reflist = ctx->h264_dec.reflists.p;
153 
154 	/*
155 	 * Each G1_REG_BD_REF_PIC(x) register contains three entries
156 	 * of each forward and backward picture list.
157 	 */
158 	reg_num = 0;
159 	for (i = 0; i < 15; i += 3) {
160 		reg = G1_REG_BD_REF_PIC_BINIT_RLIST_F0(b0_reflist[i].index) |
161 		      G1_REG_BD_REF_PIC_BINIT_RLIST_F1(b0_reflist[i + 1].index) |
162 		      G1_REG_BD_REF_PIC_BINIT_RLIST_F2(b0_reflist[i + 2].index) |
163 		      G1_REG_BD_REF_PIC_BINIT_RLIST_B0(b1_reflist[i].index) |
164 		      G1_REG_BD_REF_PIC_BINIT_RLIST_B1(b1_reflist[i + 1].index) |
165 		      G1_REG_BD_REF_PIC_BINIT_RLIST_B2(b1_reflist[i + 2].index);
166 		vdpu_write_relaxed(vpu, reg, G1_REG_BD_REF_PIC(reg_num++));
167 	}
168 
169 	/*
170 	 * G1_REG_BD_P_REF_PIC register contains last entries (index 15)
171 	 * of forward and backward reference picture lists and first 4 entries
172 	 * of P forward picture list.
173 	 */
174 	reg = G1_REG_BD_P_REF_PIC_BINIT_RLIST_F15(b0_reflist[15].index) |
175 	      G1_REG_BD_P_REF_PIC_BINIT_RLIST_B15(b1_reflist[15].index) |
176 	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F0(p_reflist[0].index) |
177 	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F1(p_reflist[1].index) |
178 	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F2(p_reflist[2].index) |
179 	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F3(p_reflist[3].index);
180 	vdpu_write_relaxed(vpu, reg, G1_REG_BD_P_REF_PIC);
181 
182 	/*
183 	 * Each G1_REG_FWD_PIC(x) register contains six consecutive
184 	 * entries of P forward picture list, starting from index 4.
185 	 */
186 	reg_num = 0;
187 	for (i = 4; i < HANTRO_H264_DPB_SIZE; i += 6) {
188 		reg = G1_REG_FWD_PIC_PINIT_RLIST_F0(p_reflist[i].index) |
189 		      G1_REG_FWD_PIC_PINIT_RLIST_F1(p_reflist[i + 1].index) |
190 		      G1_REG_FWD_PIC_PINIT_RLIST_F2(p_reflist[i + 2].index) |
191 		      G1_REG_FWD_PIC_PINIT_RLIST_F3(p_reflist[i + 3].index) |
192 		      G1_REG_FWD_PIC_PINIT_RLIST_F4(p_reflist[i + 4].index) |
193 		      G1_REG_FWD_PIC_PINIT_RLIST_F5(p_reflist[i + 5].index);
194 		vdpu_write_relaxed(vpu, reg, G1_REG_FWD_PIC(reg_num++));
195 	}
196 
197 	/* Set up addresses of DPB buffers. */
198 	for (i = 0; i < HANTRO_H264_DPB_SIZE; i++) {
199 		dma_addr_t dma_addr = hantro_h264_get_ref_buf(ctx, i);
200 
201 		vdpu_write_relaxed(vpu, dma_addr, G1_REG_ADDR_REF(i));
202 	}
203 }
204 
set_buffers(struct hantro_ctx * ctx,struct vb2_v4l2_buffer * src_buf)205 static void set_buffers(struct hantro_ctx *ctx, struct vb2_v4l2_buffer *src_buf)
206 {
207 	const struct hantro_h264_dec_ctrls *ctrls = &ctx->h264_dec.ctrls;
208 	struct vb2_v4l2_buffer *dst_buf;
209 	struct hantro_dev *vpu = ctx->dev;
210 	dma_addr_t src_dma, dst_dma;
211 	size_t offset = 0;
212 
213 	/* Source (stream) buffer. */
214 	src_dma = vb2_dma_contig_plane_dma_addr(&src_buf->vb2_buf, 0);
215 	vdpu_write_relaxed(vpu, src_dma, G1_REG_ADDR_STR);
216 
217 	/* Destination (decoded frame) buffer. */
218 	dst_buf = hantro_get_dst_buf(ctx);
219 	dst_dma = hantro_get_dec_buf_addr(ctx, &dst_buf->vb2_buf);
220 	/* Adjust dma addr to start at second line for bottom field */
221 	if (ctrls->decode->flags & V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD)
222 		offset = ALIGN(ctx->src_fmt.width, MB_DIM);
223 	vdpu_write_relaxed(vpu, dst_dma + offset, G1_REG_ADDR_DST);
224 
225 	/* Higher profiles require DMV buffer appended to reference frames. */
226 	if (ctrls->sps->profile_idc > 66 && ctrls->decode->nal_ref_idc) {
227 		unsigned int bytes_per_mb = 384;
228 
229 		/* DMV buffer for monochrome start directly after Y-plane */
230 		if (ctrls->sps->profile_idc >= 100 &&
231 		    ctrls->sps->chroma_format_idc == 0)
232 			bytes_per_mb = 256;
233 		offset = bytes_per_mb * MB_WIDTH(ctx->src_fmt.width) *
234 			 MB_HEIGHT(ctx->src_fmt.height);
235 
236 		/*
237 		 * DMV buffer is split in two for field encoded frames,
238 		 * adjust offset for bottom field
239 		 */
240 		if (ctrls->decode->flags & V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD)
241 			offset += 32 * MB_WIDTH(ctx->src_fmt.width) *
242 				  MB_HEIGHT(ctx->src_fmt.height);
243 		vdpu_write_relaxed(vpu, dst_dma + offset, G1_REG_ADDR_DIR_MV);
244 	}
245 
246 	/* Auxiliary buffer prepared in hantro_h264_dec_init(). */
247 	vdpu_write_relaxed(vpu, ctx->h264_dec.priv.dma, G1_REG_ADDR_QTABLE);
248 }
249 
hantro_g1_h264_dec_run(struct hantro_ctx * ctx)250 int hantro_g1_h264_dec_run(struct hantro_ctx *ctx)
251 {
252 	struct hantro_dev *vpu = ctx->dev;
253 	struct vb2_v4l2_buffer *src_buf;
254 	int ret;
255 
256 	/* Prepare the H264 decoder context. */
257 	ret = hantro_h264_dec_prepare_run(ctx);
258 	if (ret)
259 		return ret;
260 
261 	/* Configure hardware registers. */
262 	src_buf = hantro_get_src_buf(ctx);
263 	set_params(ctx, src_buf);
264 	set_ref(ctx);
265 	set_buffers(ctx, src_buf);
266 
267 	hantro_end_prepare_run(ctx);
268 
269 	/* Start decoding! */
270 	vdpu_write_relaxed(vpu,
271 			   G1_REG_CONFIG_DEC_AXI_RD_ID(0xffu) |
272 			   G1_REG_CONFIG_DEC_TIMEOUT_E |
273 			   G1_REG_CONFIG_DEC_OUT_ENDIAN |
274 			   G1_REG_CONFIG_DEC_STRENDIAN_E |
275 			   G1_REG_CONFIG_DEC_MAX_BURST(16) |
276 			   G1_REG_CONFIG_DEC_OUTSWAP32_E |
277 			   G1_REG_CONFIG_DEC_INSWAP32_E |
278 			   G1_REG_CONFIG_DEC_STRSWAP32_E |
279 			   G1_REG_CONFIG_DEC_CLK_GATE_E,
280 			   G1_REG_CONFIG);
281 	vdpu_write(vpu, G1_REG_INTERRUPT_DEC_E, G1_REG_INTERRUPT);
282 
283 	return 0;
284 }
285