1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Hantro VPU HEVC codec driver
4 *
5 * Copyright (C) 2020 Safran Passenger Innovations LLC
6 */
7
8 #include "hantro_hw.h"
9 #include "hantro_g2_regs.h"
10
prepare_tile_info_buffer(struct hantro_ctx * ctx)11 static void prepare_tile_info_buffer(struct hantro_ctx *ctx)
12 {
13 struct hantro_dev *vpu = ctx->dev;
14 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
15 const struct v4l2_ctrl_hevc_pps *pps = ctrls->pps;
16 const struct v4l2_ctrl_hevc_sps *sps = ctrls->sps;
17 u16 *p = (u16 *)((u8 *)ctx->hevc_dec.tile_sizes.cpu);
18 unsigned int num_tile_rows = pps->num_tile_rows_minus1 + 1;
19 unsigned int num_tile_cols = pps->num_tile_columns_minus1 + 1;
20 unsigned int pic_width_in_ctbs, pic_height_in_ctbs;
21 unsigned int max_log2_ctb_size, ctb_size;
22 bool tiles_enabled, uniform_spacing;
23 u32 no_chroma = 0;
24
25 tiles_enabled = !!(pps->flags & V4L2_HEVC_PPS_FLAG_TILES_ENABLED);
26 uniform_spacing = !!(pps->flags & V4L2_HEVC_PPS_FLAG_UNIFORM_SPACING);
27
28 hantro_reg_write(vpu, &g2_tile_e, tiles_enabled);
29
30 max_log2_ctb_size = sps->log2_min_luma_coding_block_size_minus3 + 3 +
31 sps->log2_diff_max_min_luma_coding_block_size;
32 pic_width_in_ctbs = (sps->pic_width_in_luma_samples +
33 (1 << max_log2_ctb_size) - 1) >> max_log2_ctb_size;
34 pic_height_in_ctbs = (sps->pic_height_in_luma_samples + (1 << max_log2_ctb_size) - 1)
35 >> max_log2_ctb_size;
36 ctb_size = 1 << max_log2_ctb_size;
37
38 vpu_debug(1, "Preparing tile sizes buffer for %dx%d CTBs (CTB size %d)\n",
39 pic_width_in_ctbs, pic_height_in_ctbs, ctb_size);
40
41 if (tiles_enabled) {
42 unsigned int i, j, h;
43
44 vpu_debug(1, "Tiles enabled! %dx%d\n", num_tile_cols, num_tile_rows);
45
46 hantro_reg_write(vpu, &g2_num_tile_rows, num_tile_rows);
47 hantro_reg_write(vpu, &g2_num_tile_cols, num_tile_cols);
48
49 /* write width + height for each tile in pic */
50 if (!uniform_spacing) {
51 u32 tmp_w = 0, tmp_h = 0;
52
53 for (i = 0; i < num_tile_rows; i++) {
54 if (i == num_tile_rows - 1)
55 h = pic_height_in_ctbs - tmp_h;
56 else
57 h = pps->row_height_minus1[i] + 1;
58 tmp_h += h;
59 if (i == 0 && h == 1 && ctb_size == 16)
60 no_chroma = 1;
61 for (j = 0, tmp_w = 0; j < num_tile_cols - 1; j++) {
62 tmp_w += pps->column_width_minus1[j] + 1;
63 *p++ = pps->column_width_minus1[j] + 1;
64 *p++ = h;
65 if (i == 0 && h == 1 && ctb_size == 16)
66 no_chroma = 1;
67 }
68 /* last column */
69 *p++ = pic_width_in_ctbs - tmp_w;
70 *p++ = h;
71 }
72 } else { /* uniform spacing */
73 u32 tmp, prev_h, prev_w;
74
75 for (i = 0, prev_h = 0; i < num_tile_rows; i++) {
76 tmp = (i + 1) * pic_height_in_ctbs / num_tile_rows;
77 h = tmp - prev_h;
78 prev_h = tmp;
79 if (i == 0 && h == 1 && ctb_size == 16)
80 no_chroma = 1;
81 for (j = 0, prev_w = 0; j < num_tile_cols; j++) {
82 tmp = (j + 1) * pic_width_in_ctbs / num_tile_cols;
83 *p++ = tmp - prev_w;
84 *p++ = h;
85 if (j == 0 &&
86 (pps->column_width_minus1[0] + 1) == 1 &&
87 ctb_size == 16)
88 no_chroma = 1;
89 prev_w = tmp;
90 }
91 }
92 }
93 } else {
94 hantro_reg_write(vpu, &g2_num_tile_rows, 1);
95 hantro_reg_write(vpu, &g2_num_tile_cols, 1);
96
97 /* There's one tile, with dimensions equal to pic size. */
98 p[0] = pic_width_in_ctbs;
99 p[1] = pic_height_in_ctbs;
100 }
101
102 if (no_chroma)
103 vpu_debug(1, "%s: no chroma!\n", __func__);
104 }
105
compute_header_skip_length(struct hantro_ctx * ctx)106 static int compute_header_skip_length(struct hantro_ctx *ctx)
107 {
108 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
109 const struct v4l2_ctrl_hevc_decode_params *decode_params = ctrls->decode_params;
110 const struct v4l2_ctrl_hevc_sps *sps = ctrls->sps;
111 const struct v4l2_ctrl_hevc_pps *pps = ctrls->pps;
112 int skip = 0;
113
114 if (pps->flags & V4L2_HEVC_PPS_FLAG_OUTPUT_FLAG_PRESENT)
115 /* size of pic_output_flag */
116 skip++;
117
118 if (sps->flags & V4L2_HEVC_SPS_FLAG_SEPARATE_COLOUR_PLANE)
119 /* size of pic_order_cnt_lsb */
120 skip += 2;
121
122 if (!(decode_params->flags & V4L2_HEVC_DECODE_PARAM_FLAG_IDR_PIC)) {
123 /* size of pic_order_cnt_lsb */
124 skip += sps->log2_max_pic_order_cnt_lsb_minus4 + 4;
125
126 /* size of short_term_ref_pic_set_sps_flag */
127 skip++;
128
129 if (decode_params->short_term_ref_pic_set_size)
130 /* size of st_ref_pic_set( num_short_term_ref_pic_sets ) */
131 skip += decode_params->short_term_ref_pic_set_size;
132 else if (sps->num_short_term_ref_pic_sets > 1)
133 skip += fls(sps->num_short_term_ref_pic_sets - 1);
134
135 skip += decode_params->long_term_ref_pic_set_size;
136 }
137
138 return skip;
139 }
140
set_params(struct hantro_ctx * ctx)141 static void set_params(struct hantro_ctx *ctx)
142 {
143 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
144 const struct v4l2_ctrl_hevc_sps *sps = ctrls->sps;
145 const struct v4l2_ctrl_hevc_pps *pps = ctrls->pps;
146 const struct v4l2_ctrl_hevc_decode_params *decode_params = ctrls->decode_params;
147 struct hantro_dev *vpu = ctx->dev;
148 u32 min_log2_cb_size, max_log2_ctb_size, min_cb_size, max_ctb_size;
149 u32 pic_width_in_min_cbs, pic_height_in_min_cbs;
150 u32 pic_width_aligned, pic_height_aligned;
151 u32 partial_ctb_x, partial_ctb_y;
152
153 hantro_reg_write(vpu, &g2_bit_depth_y_minus8, sps->bit_depth_luma_minus8);
154 hantro_reg_write(vpu, &g2_bit_depth_c_minus8, sps->bit_depth_chroma_minus8);
155
156 hantro_reg_write(vpu, &g2_hdr_skip_length, compute_header_skip_length(ctx));
157
158 min_log2_cb_size = sps->log2_min_luma_coding_block_size_minus3 + 3;
159 max_log2_ctb_size = min_log2_cb_size + sps->log2_diff_max_min_luma_coding_block_size;
160
161 hantro_reg_write(vpu, &g2_min_cb_size, min_log2_cb_size);
162 hantro_reg_write(vpu, &g2_max_cb_size, max_log2_ctb_size);
163
164 min_cb_size = 1 << min_log2_cb_size;
165 max_ctb_size = 1 << max_log2_ctb_size;
166
167 pic_width_in_min_cbs = sps->pic_width_in_luma_samples / min_cb_size;
168 pic_height_in_min_cbs = sps->pic_height_in_luma_samples / min_cb_size;
169 pic_width_aligned = ALIGN(sps->pic_width_in_luma_samples, max_ctb_size);
170 pic_height_aligned = ALIGN(sps->pic_height_in_luma_samples, max_ctb_size);
171
172 partial_ctb_x = !!(sps->pic_width_in_luma_samples != pic_width_aligned);
173 partial_ctb_y = !!(sps->pic_height_in_luma_samples != pic_height_aligned);
174
175 hantro_reg_write(vpu, &g2_partial_ctb_x, partial_ctb_x);
176 hantro_reg_write(vpu, &g2_partial_ctb_y, partial_ctb_y);
177
178 hantro_reg_write(vpu, &g2_pic_width_in_cbs, pic_width_in_min_cbs);
179 hantro_reg_write(vpu, &g2_pic_height_in_cbs, pic_height_in_min_cbs);
180
181 hantro_reg_write(vpu, &g2_pic_width_4x4,
182 (pic_width_in_min_cbs * min_cb_size) / 4);
183 hantro_reg_write(vpu, &g2_pic_height_4x4,
184 (pic_height_in_min_cbs * min_cb_size) / 4);
185
186 hantro_reg_write(vpu, &hevc_max_inter_hierdepth,
187 sps->max_transform_hierarchy_depth_inter);
188 hantro_reg_write(vpu, &hevc_max_intra_hierdepth,
189 sps->max_transform_hierarchy_depth_intra);
190 hantro_reg_write(vpu, &hevc_min_trb_size,
191 sps->log2_min_luma_transform_block_size_minus2 + 2);
192 hantro_reg_write(vpu, &hevc_max_trb_size,
193 sps->log2_min_luma_transform_block_size_minus2 + 2 +
194 sps->log2_diff_max_min_luma_transform_block_size);
195
196 hantro_reg_write(vpu, &g2_tempor_mvp_e,
197 !!(sps->flags & V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED) &&
198 !(decode_params->flags & V4L2_HEVC_DECODE_PARAM_FLAG_IDR_PIC));
199 hantro_reg_write(vpu, &g2_strong_smooth_e,
200 !!(sps->flags & V4L2_HEVC_SPS_FLAG_STRONG_INTRA_SMOOTHING_ENABLED));
201 hantro_reg_write(vpu, &g2_asym_pred_e,
202 !!(sps->flags & V4L2_HEVC_SPS_FLAG_AMP_ENABLED));
203 hantro_reg_write(vpu, &g2_sao_e,
204 !!(sps->flags & V4L2_HEVC_SPS_FLAG_SAMPLE_ADAPTIVE_OFFSET));
205 hantro_reg_write(vpu, &g2_sign_data_hide,
206 !!(pps->flags & V4L2_HEVC_PPS_FLAG_SIGN_DATA_HIDING_ENABLED));
207
208 if (pps->flags & V4L2_HEVC_PPS_FLAG_CU_QP_DELTA_ENABLED) {
209 hantro_reg_write(vpu, &g2_cu_qpd_e, 1);
210 hantro_reg_write(vpu, &g2_max_cu_qpd_depth, pps->diff_cu_qp_delta_depth);
211 } else {
212 hantro_reg_write(vpu, &g2_cu_qpd_e, 0);
213 hantro_reg_write(vpu, &g2_max_cu_qpd_depth, 0);
214 }
215
216 hantro_reg_write(vpu, &g2_cb_qp_offset, pps->pps_cb_qp_offset);
217 hantro_reg_write(vpu, &g2_cr_qp_offset, pps->pps_cr_qp_offset);
218
219 hantro_reg_write(vpu, &g2_filt_offset_beta, pps->pps_beta_offset_div2);
220 hantro_reg_write(vpu, &g2_filt_offset_tc, pps->pps_tc_offset_div2);
221 hantro_reg_write(vpu, &g2_slice_hdr_ext_e,
222 !!(pps->flags & V4L2_HEVC_PPS_FLAG_SLICE_SEGMENT_HEADER_EXTENSION_PRESENT));
223 hantro_reg_write(vpu, &g2_slice_hdr_ext_bits, pps->num_extra_slice_header_bits);
224 hantro_reg_write(vpu, &g2_slice_chqp_present,
225 !!(pps->flags & V4L2_HEVC_PPS_FLAG_PPS_SLICE_CHROMA_QP_OFFSETS_PRESENT));
226 hantro_reg_write(vpu, &g2_weight_bipr_idc,
227 !!(pps->flags & V4L2_HEVC_PPS_FLAG_WEIGHTED_BIPRED));
228 hantro_reg_write(vpu, &g2_transq_bypass,
229 !!(pps->flags & V4L2_HEVC_PPS_FLAG_TRANSQUANT_BYPASS_ENABLED));
230 hantro_reg_write(vpu, &g2_list_mod_e,
231 !!(pps->flags & V4L2_HEVC_PPS_FLAG_LISTS_MODIFICATION_PRESENT));
232 hantro_reg_write(vpu, &g2_entropy_sync_e,
233 !!(pps->flags & V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED));
234 hantro_reg_write(vpu, &g2_cabac_init_present,
235 !!(pps->flags & V4L2_HEVC_PPS_FLAG_CABAC_INIT_PRESENT));
236 hantro_reg_write(vpu, &g2_idr_pic_e,
237 !!(decode_params->flags & V4L2_HEVC_DECODE_PARAM_FLAG_IRAP_PIC));
238 hantro_reg_write(vpu, &hevc_parallel_merge,
239 pps->log2_parallel_merge_level_minus2 + 2);
240 hantro_reg_write(vpu, &g2_pcm_filt_d,
241 !!(sps->flags & V4L2_HEVC_SPS_FLAG_PCM_LOOP_FILTER_DISABLED));
242 hantro_reg_write(vpu, &g2_pcm_e,
243 !!(sps->flags & V4L2_HEVC_SPS_FLAG_PCM_ENABLED));
244 if (sps->flags & V4L2_HEVC_SPS_FLAG_PCM_ENABLED) {
245 hantro_reg_write(vpu, &g2_max_pcm_size,
246 sps->log2_diff_max_min_pcm_luma_coding_block_size +
247 sps->log2_min_pcm_luma_coding_block_size_minus3 + 3);
248 hantro_reg_write(vpu, &g2_min_pcm_size,
249 sps->log2_min_pcm_luma_coding_block_size_minus3 + 3);
250 hantro_reg_write(vpu, &g2_bit_depth_pcm_y,
251 sps->pcm_sample_bit_depth_luma_minus1 + 1);
252 hantro_reg_write(vpu, &g2_bit_depth_pcm_c,
253 sps->pcm_sample_bit_depth_chroma_minus1 + 1);
254 } else {
255 hantro_reg_write(vpu, &g2_max_pcm_size, 0);
256 hantro_reg_write(vpu, &g2_min_pcm_size, 0);
257 hantro_reg_write(vpu, &g2_bit_depth_pcm_y, 0);
258 hantro_reg_write(vpu, &g2_bit_depth_pcm_c, 0);
259 }
260
261 hantro_reg_write(vpu, &g2_start_code_e, 1);
262 hantro_reg_write(vpu, &g2_init_qp, pps->init_qp_minus26 + 26);
263 hantro_reg_write(vpu, &g2_weight_pred_e,
264 !!(pps->flags & V4L2_HEVC_PPS_FLAG_WEIGHTED_PRED));
265 hantro_reg_write(vpu, &g2_cabac_init_present,
266 !!(pps->flags & V4L2_HEVC_PPS_FLAG_CABAC_INIT_PRESENT));
267 hantro_reg_write(vpu, &g2_const_intra_e,
268 !!(pps->flags & V4L2_HEVC_PPS_FLAG_CONSTRAINED_INTRA_PRED));
269 hantro_reg_write(vpu, &g2_transform_skip,
270 !!(pps->flags & V4L2_HEVC_PPS_FLAG_TRANSFORM_SKIP_ENABLED));
271 hantro_reg_write(vpu, &g2_out_filtering_dis,
272 !!(pps->flags & V4L2_HEVC_PPS_FLAG_PPS_DISABLE_DEBLOCKING_FILTER));
273 hantro_reg_write(vpu, &g2_filt_ctrl_pres,
274 !!(pps->flags & V4L2_HEVC_PPS_FLAG_DEBLOCKING_FILTER_CONTROL_PRESENT));
275 hantro_reg_write(vpu, &g2_dependent_slice,
276 !!(pps->flags & V4L2_HEVC_PPS_FLAG_DEPENDENT_SLICE_SEGMENT_ENABLED));
277 hantro_reg_write(vpu, &g2_filter_override,
278 !!(pps->flags & V4L2_HEVC_PPS_FLAG_DEBLOCKING_FILTER_OVERRIDE_ENABLED));
279 hantro_reg_write(vpu, &g2_refidx0_active,
280 pps->num_ref_idx_l0_default_active_minus1 + 1);
281 hantro_reg_write(vpu, &g2_refidx1_active,
282 pps->num_ref_idx_l1_default_active_minus1 + 1);
283 hantro_reg_write(vpu, &g2_apf_threshold, 8);
284 }
285
get_dpb_index(const struct v4l2_ctrl_hevc_decode_params * decode_params,const u32 index)286 static u32 get_dpb_index(const struct v4l2_ctrl_hevc_decode_params *decode_params,
287 const u32 index)
288 {
289 if (index > decode_params->num_active_dpb_entries)
290 return 0;
291
292 return index;
293 }
294
set_ref_pic_list(struct hantro_ctx * ctx)295 static void set_ref_pic_list(struct hantro_ctx *ctx)
296 {
297 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
298 struct hantro_dev *vpu = ctx->dev;
299 const struct v4l2_ctrl_hevc_decode_params *decode_params = ctrls->decode_params;
300 u32 list0[V4L2_HEVC_DPB_ENTRIES_NUM_MAX] = {};
301 u32 list1[V4L2_HEVC_DPB_ENTRIES_NUM_MAX] = {};
302 static const struct hantro_reg ref_pic_regs0[] = {
303 hevc_rlist_f0,
304 hevc_rlist_f1,
305 hevc_rlist_f2,
306 hevc_rlist_f3,
307 hevc_rlist_f4,
308 hevc_rlist_f5,
309 hevc_rlist_f6,
310 hevc_rlist_f7,
311 hevc_rlist_f8,
312 hevc_rlist_f9,
313 hevc_rlist_f10,
314 hevc_rlist_f11,
315 hevc_rlist_f12,
316 hevc_rlist_f13,
317 hevc_rlist_f14,
318 hevc_rlist_f15,
319 };
320 static const struct hantro_reg ref_pic_regs1[] = {
321 hevc_rlist_b0,
322 hevc_rlist_b1,
323 hevc_rlist_b2,
324 hevc_rlist_b3,
325 hevc_rlist_b4,
326 hevc_rlist_b5,
327 hevc_rlist_b6,
328 hevc_rlist_b7,
329 hevc_rlist_b8,
330 hevc_rlist_b9,
331 hevc_rlist_b10,
332 hevc_rlist_b11,
333 hevc_rlist_b12,
334 hevc_rlist_b13,
335 hevc_rlist_b14,
336 hevc_rlist_b15,
337 };
338 unsigned int i, j;
339
340 /* List 0 contains: short term before, short term after and long term */
341 j = 0;
342 for (i = 0; i < decode_params->num_poc_st_curr_before && j < ARRAY_SIZE(list0); i++)
343 list0[j++] = decode_params->poc_st_curr_before[i];
344 for (i = 0; i < decode_params->num_poc_st_curr_after && j < ARRAY_SIZE(list0); i++)
345 list0[j++] = decode_params->poc_st_curr_after[i];
346 for (i = 0; i < decode_params->num_poc_lt_curr && j < ARRAY_SIZE(list0); i++)
347 list0[j++] = decode_params->poc_lt_curr[i];
348
349 /* Fill the list, copying over and over */
350 i = 0;
351 while (j < ARRAY_SIZE(list0))
352 list0[j++] = list0[i++];
353
354 j = 0;
355 for (i = 0; i < decode_params->num_poc_st_curr_after && j < ARRAY_SIZE(list1); i++)
356 list1[j++] = decode_params->poc_st_curr_after[i];
357 for (i = 0; i < decode_params->num_poc_st_curr_before && j < ARRAY_SIZE(list1); i++)
358 list1[j++] = decode_params->poc_st_curr_before[i];
359 for (i = 0; i < decode_params->num_poc_lt_curr && j < ARRAY_SIZE(list1); i++)
360 list1[j++] = decode_params->poc_lt_curr[i];
361
362 i = 0;
363 while (j < ARRAY_SIZE(list1))
364 list1[j++] = list1[i++];
365
366 for (i = 0; i < V4L2_HEVC_DPB_ENTRIES_NUM_MAX; i++) {
367 hantro_reg_write(vpu, &ref_pic_regs0[i],
368 get_dpb_index(decode_params, list0[i]));
369 hantro_reg_write(vpu, &ref_pic_regs1[i],
370 get_dpb_index(decode_params, list1[i]));
371 }
372 }
373
set_ref(struct hantro_ctx * ctx)374 static int set_ref(struct hantro_ctx *ctx)
375 {
376 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
377 const struct v4l2_ctrl_hevc_pps *pps = ctrls->pps;
378 const struct v4l2_ctrl_hevc_decode_params *decode_params = ctrls->decode_params;
379 const struct v4l2_hevc_dpb_entry *dpb = decode_params->dpb;
380 dma_addr_t luma_addr, chroma_addr, mv_addr = 0;
381 dma_addr_t compress_luma_addr, compress_chroma_addr = 0;
382 struct hantro_dev *vpu = ctx->dev;
383 struct vb2_v4l2_buffer *vb2_dst;
384 struct hantro_decoded_buffer *dst;
385 size_t cr_offset = hantro_g2_chroma_offset(ctx);
386 size_t mv_offset = hantro_g2_motion_vectors_offset(ctx);
387 size_t compress_luma_offset = hantro_g2_luma_compress_offset(ctx);
388 size_t compress_chroma_offset = hantro_g2_chroma_compress_offset(ctx);
389 u32 max_ref_frames;
390 u16 dpb_longterm_e;
391 static const struct hantro_reg cur_poc[] = {
392 hevc_cur_poc_00,
393 hevc_cur_poc_01,
394 hevc_cur_poc_02,
395 hevc_cur_poc_03,
396 hevc_cur_poc_04,
397 hevc_cur_poc_05,
398 hevc_cur_poc_06,
399 hevc_cur_poc_07,
400 hevc_cur_poc_08,
401 hevc_cur_poc_09,
402 hevc_cur_poc_10,
403 hevc_cur_poc_11,
404 hevc_cur_poc_12,
405 hevc_cur_poc_13,
406 hevc_cur_poc_14,
407 hevc_cur_poc_15,
408 };
409 unsigned int i;
410
411 max_ref_frames = decode_params->num_poc_lt_curr +
412 decode_params->num_poc_st_curr_before +
413 decode_params->num_poc_st_curr_after;
414 /*
415 * Set max_ref_frames to non-zero to avoid HW hang when decoding
416 * badly marked I-frames.
417 */
418 max_ref_frames = max_ref_frames ? max_ref_frames : 1;
419 hantro_reg_write(vpu, &g2_num_ref_frames, max_ref_frames);
420 hantro_reg_write(vpu, &g2_filter_over_slices,
421 !!(pps->flags & V4L2_HEVC_PPS_FLAG_PPS_LOOP_FILTER_ACROSS_SLICES_ENABLED));
422 hantro_reg_write(vpu, &g2_filter_over_tiles,
423 !!(pps->flags & V4L2_HEVC_PPS_FLAG_LOOP_FILTER_ACROSS_TILES_ENABLED));
424
425 /*
426 * Write POC count diff from current pic.
427 */
428 for (i = 0; i < decode_params->num_active_dpb_entries && i < ARRAY_SIZE(cur_poc); i++) {
429 char poc_diff = decode_params->pic_order_cnt_val - dpb[i].pic_order_cnt_val;
430
431 hantro_reg_write(vpu, &cur_poc[i], poc_diff);
432 }
433
434 if (i < ARRAY_SIZE(cur_poc)) {
435 /*
436 * After the references, fill one entry pointing to itself,
437 * i.e. difference is zero.
438 */
439 hantro_reg_write(vpu, &cur_poc[i], 0);
440 i++;
441 }
442
443 /* Fill the rest with the current picture */
444 for (; i < ARRAY_SIZE(cur_poc); i++)
445 hantro_reg_write(vpu, &cur_poc[i], decode_params->pic_order_cnt_val);
446
447 set_ref_pic_list(ctx);
448
449 /* We will only keep the reference pictures that are still used */
450 hantro_hevc_ref_init(ctx);
451
452 /* Set up addresses of DPB buffers */
453 dpb_longterm_e = 0;
454 for (i = 0; i < decode_params->num_active_dpb_entries &&
455 i < (V4L2_HEVC_DPB_ENTRIES_NUM_MAX - 1); i++) {
456 luma_addr = hantro_hevc_get_ref_buf(ctx, dpb[i].pic_order_cnt_val);
457 if (!luma_addr)
458 return -ENOMEM;
459
460 chroma_addr = luma_addr + cr_offset;
461 mv_addr = luma_addr + mv_offset;
462 compress_luma_addr = luma_addr + compress_luma_offset;
463 compress_chroma_addr = luma_addr + compress_chroma_offset;
464
465 if (dpb[i].flags & V4L2_HEVC_DPB_ENTRY_LONG_TERM_REFERENCE)
466 dpb_longterm_e |= BIT(V4L2_HEVC_DPB_ENTRIES_NUM_MAX - 1 - i);
467
468 hantro_write_addr(vpu, G2_REF_LUMA_ADDR(i), luma_addr);
469 hantro_write_addr(vpu, G2_REF_CHROMA_ADDR(i), chroma_addr);
470 hantro_write_addr(vpu, G2_REF_MV_ADDR(i), mv_addr);
471 hantro_write_addr(vpu, G2_REF_COMP_LUMA_ADDR(i), compress_luma_addr);
472 hantro_write_addr(vpu, G2_REF_COMP_CHROMA_ADDR(i), compress_chroma_addr);
473 }
474
475 vb2_dst = hantro_get_dst_buf(ctx);
476 dst = vb2_to_hantro_decoded_buf(&vb2_dst->vb2_buf);
477 luma_addr = hantro_get_dec_buf_addr(ctx, &dst->base.vb.vb2_buf);
478 if (!luma_addr)
479 return -ENOMEM;
480
481 if (hantro_hevc_add_ref_buf(ctx, decode_params->pic_order_cnt_val, luma_addr))
482 return -EINVAL;
483
484 chroma_addr = luma_addr + cr_offset;
485 mv_addr = luma_addr + mv_offset;
486 compress_luma_addr = luma_addr + compress_luma_offset;
487 compress_chroma_addr = luma_addr + compress_chroma_offset;
488
489 hantro_write_addr(vpu, G2_REF_LUMA_ADDR(i), luma_addr);
490 hantro_write_addr(vpu, G2_REF_CHROMA_ADDR(i), chroma_addr);
491 hantro_write_addr(vpu, G2_REF_MV_ADDR(i), mv_addr);
492 hantro_write_addr(vpu, G2_REF_COMP_LUMA_ADDR(i), compress_luma_addr);
493 hantro_write_addr(vpu, G2_REF_COMP_CHROMA_ADDR(i++), compress_chroma_addr);
494
495 hantro_write_addr(vpu, G2_OUT_LUMA_ADDR, luma_addr);
496 hantro_write_addr(vpu, G2_OUT_CHROMA_ADDR, chroma_addr);
497 hantro_write_addr(vpu, G2_OUT_MV_ADDR, mv_addr);
498 hantro_write_addr(vpu, G2_OUT_COMP_LUMA_ADDR, compress_luma_addr);
499 hantro_write_addr(vpu, G2_OUT_COMP_CHROMA_ADDR, compress_chroma_addr);
500
501 for (; i < V4L2_HEVC_DPB_ENTRIES_NUM_MAX; i++) {
502 hantro_write_addr(vpu, G2_REF_LUMA_ADDR(i), 0);
503 hantro_write_addr(vpu, G2_REF_CHROMA_ADDR(i), 0);
504 hantro_write_addr(vpu, G2_REF_MV_ADDR(i), 0);
505 hantro_write_addr(vpu, G2_REF_COMP_LUMA_ADDR(i), 0);
506 hantro_write_addr(vpu, G2_REF_COMP_CHROMA_ADDR(i), 0);
507 }
508
509 hantro_reg_write(vpu, &g2_refer_lterm_e, dpb_longterm_e);
510
511 return 0;
512 }
513
set_buffers(struct hantro_ctx * ctx)514 static void set_buffers(struct hantro_ctx *ctx)
515 {
516 struct vb2_v4l2_buffer *src_buf;
517 struct hantro_dev *vpu = ctx->dev;
518 dma_addr_t src_dma;
519 u32 src_len, src_buf_len;
520
521 src_buf = hantro_get_src_buf(ctx);
522
523 /* Source (stream) buffer. */
524 src_dma = vb2_dma_contig_plane_dma_addr(&src_buf->vb2_buf, 0);
525 src_len = vb2_get_plane_payload(&src_buf->vb2_buf, 0);
526 src_buf_len = vb2_plane_size(&src_buf->vb2_buf, 0);
527
528 hantro_write_addr(vpu, G2_STREAM_ADDR, src_dma);
529 hantro_reg_write(vpu, &g2_stream_len, src_len);
530 hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
531 hantro_reg_write(vpu, &g2_strm_start_offset, 0);
532 hantro_reg_write(vpu, &g2_start_bit, 0);
533 hantro_reg_write(vpu, &g2_write_mvs_e, 1);
534
535 hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
536 hantro_write_addr(vpu, G2_TILE_FILTER_ADDR, ctx->hevc_dec.tile_filter.dma);
537 hantro_write_addr(vpu, G2_TILE_SAO_ADDR, ctx->hevc_dec.tile_sao.dma);
538 hantro_write_addr(vpu, G2_TILE_BSD_ADDR, ctx->hevc_dec.tile_bsd.dma);
539 }
540
prepare_scaling_list_buffer(struct hantro_ctx * ctx)541 static void prepare_scaling_list_buffer(struct hantro_ctx *ctx)
542 {
543 struct hantro_dev *vpu = ctx->dev;
544 const struct hantro_hevc_dec_ctrls *ctrls = &ctx->hevc_dec.ctrls;
545 const struct v4l2_ctrl_hevc_scaling_matrix *sc = ctrls->scaling;
546 const struct v4l2_ctrl_hevc_sps *sps = ctrls->sps;
547 u8 *p = ((u8 *)ctx->hevc_dec.scaling_lists.cpu);
548 unsigned int scaling_list_enabled;
549 unsigned int i, j, k;
550
551 scaling_list_enabled = !!(sps->flags & V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED);
552 hantro_reg_write(vpu, &g2_scaling_list_e, scaling_list_enabled);
553
554 if (!scaling_list_enabled)
555 return;
556
557 for (i = 0; i < ARRAY_SIZE(sc->scaling_list_dc_coef_16x16); i++)
558 *p++ = sc->scaling_list_dc_coef_16x16[i];
559
560 for (i = 0; i < ARRAY_SIZE(sc->scaling_list_dc_coef_32x32); i++)
561 *p++ = sc->scaling_list_dc_coef_32x32[i];
562
563 /* 128-bit boundary */
564 p += 8;
565
566 /* write scaling lists column by column */
567
568 for (i = 0; i < 6; i++)
569 for (j = 0; j < 4; j++)
570 for (k = 0; k < 4; k++)
571 *p++ = sc->scaling_list_4x4[i][4 * k + j];
572
573 for (i = 0; i < 6; i++)
574 for (j = 0; j < 8; j++)
575 for (k = 0; k < 8; k++)
576 *p++ = sc->scaling_list_8x8[i][8 * k + j];
577
578 for (i = 0; i < 6; i++)
579 for (j = 0; j < 8; j++)
580 for (k = 0; k < 8; k++)
581 *p++ = sc->scaling_list_16x16[i][8 * k + j];
582
583 for (i = 0; i < 2; i++)
584 for (j = 0; j < 8; j++)
585 for (k = 0; k < 8; k++)
586 *p++ = sc->scaling_list_32x32[i][8 * k + j];
587
588 hantro_write_addr(vpu, G2_HEVC_SCALING_LIST_ADDR, ctx->hevc_dec.scaling_lists.dma);
589 }
590
hantro_g2_hevc_dec_run(struct hantro_ctx * ctx)591 int hantro_g2_hevc_dec_run(struct hantro_ctx *ctx)
592 {
593 struct hantro_dev *vpu = ctx->dev;
594 int ret;
595
596 /* Prepare HEVC decoder context. */
597 ret = hantro_hevc_dec_prepare_run(ctx);
598 if (ret)
599 return ret;
600
601 /* Configure hardware registers. */
602 set_params(ctx);
603
604 /* set reference pictures */
605 ret = set_ref(ctx);
606 if (ret)
607 return ret;
608
609 set_buffers(ctx);
610 prepare_tile_info_buffer(ctx);
611
612 prepare_scaling_list_buffer(ctx);
613
614 hantro_end_prepare_run(ctx);
615
616 hantro_reg_write(vpu, &g2_mode, HEVC_DEC_MODE);
617 hantro_reg_write(vpu, &g2_clk_gate_e, 1);
618
619 /* Don't disable output */
620 hantro_reg_write(vpu, &g2_out_dis, 0);
621
622 hantro_reg_write(vpu, &g2_ref_compress_bypass, !ctx->hevc_dec.use_compression);
623
624 /* Bus width and max burst */
625 hantro_reg_write(vpu, &g2_buswidth, BUS_WIDTH_128);
626 hantro_reg_write(vpu, &g2_max_burst, 16);
627
628 /* Swap */
629 hantro_reg_write(vpu, &g2_strm_swap, 0xf);
630 hantro_reg_write(vpu, &g2_dirmv_swap, 0xf);
631 hantro_reg_write(vpu, &g2_compress_swap, 0xf);
632
633 /* Start decoding! */
634 vdpu_write(vpu, G2_REG_INTERRUPT_DEC_E, G2_REG_INTERRUPT);
635
636 return 0;
637 }
638