1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) STMicroelectronics SA 2015
4 * Authors: Yannick Fertre <yannick.fertre@st.com>
5 * Hugues Fruchet <hugues.fruchet@st.com>
6 */
7
8 #include "hva.h"
9 #include "hva-hw.h"
10
11 #define MAX_SPS_PPS_SIZE 128
12
13 #define BITSTREAM_OFFSET_MASK 0x7F
14
15 /* video max size*/
16 #define H264_MAX_SIZE_W 1920
17 #define H264_MAX_SIZE_H 1920
18
19 /* macroBlocs number (width & height) */
20 #define MB_W(w) ((w + 0xF) / 0x10)
21 #define MB_H(h) ((h + 0xF) / 0x10)
22
23 /* formula to get temporal or spatial data size */
24 #define DATA_SIZE(w, h) (MB_W(w) * MB_H(h) * 16)
25
26 #define SEARCH_WINDOW_BUFFER_MAX_SIZE(w) ((4 * MB_W(w) + 42) * 256 * 3 / 2)
27 #define CABAC_CONTEXT_BUFFER_MAX_SIZE(w) (MB_W(w) * 16)
28 #define CTX_MB_BUFFER_MAX_SIZE(w) (MB_W(w) * 16 * 8)
29 #define SLICE_HEADER_SIZE (4 * 16)
30 #define BRC_DATA_SIZE (5 * 16)
31
32 /* source buffer copy in YUV 420 MB-tiled format with size=16*256*3/2 */
33 #define CURRENT_WINDOW_BUFFER_MAX_SIZE (16 * 256 * 3 / 2)
34
35 /*
36 * 4 lines of pixels (in Luma, Chroma blue and Chroma red) of top MB
37 * for deblocking with size=4*16*MBx*2
38 */
39 #define LOCAL_RECONSTRUCTED_BUFFER_MAX_SIZE(w) (4 * 16 * MB_W(w) * 2)
40
41 /* factor for bitrate and cpb buffer size max values if profile >= high */
42 #define H264_FACTOR_HIGH 1200
43
44 /* factor for bitrate and cpb buffer size max values if profile < high */
45 #define H264_FACTOR_BASELINE 1000
46
47 /* number of bytes for NALU_TYPE_FILLER_DATA header and footer */
48 #define H264_FILLER_DATA_SIZE 6
49
50 struct h264_profile {
51 enum v4l2_mpeg_video_h264_level level;
52 u32 max_mb_per_seconds;
53 u32 max_frame_size;
54 u32 max_bitrate;
55 u32 max_cpb_size;
56 u32 min_comp_ratio;
57 };
58
59 static const struct h264_profile h264_infos_list[] = {
60 {V4L2_MPEG_VIDEO_H264_LEVEL_1_0, 1485, 99, 64, 175, 2},
61 {V4L2_MPEG_VIDEO_H264_LEVEL_1B, 1485, 99, 128, 350, 2},
62 {V4L2_MPEG_VIDEO_H264_LEVEL_1_1, 3000, 396, 192, 500, 2},
63 {V4L2_MPEG_VIDEO_H264_LEVEL_1_2, 6000, 396, 384, 1000, 2},
64 {V4L2_MPEG_VIDEO_H264_LEVEL_1_3, 11880, 396, 768, 2000, 2},
65 {V4L2_MPEG_VIDEO_H264_LEVEL_2_0, 11880, 396, 2000, 2000, 2},
66 {V4L2_MPEG_VIDEO_H264_LEVEL_2_1, 19800, 792, 4000, 4000, 2},
67 {V4L2_MPEG_VIDEO_H264_LEVEL_2_2, 20250, 1620, 4000, 4000, 2},
68 {V4L2_MPEG_VIDEO_H264_LEVEL_3_0, 40500, 1620, 10000, 10000, 2},
69 {V4L2_MPEG_VIDEO_H264_LEVEL_3_1, 108000, 3600, 14000, 14000, 4},
70 {V4L2_MPEG_VIDEO_H264_LEVEL_3_2, 216000, 5120, 20000, 20000, 4},
71 {V4L2_MPEG_VIDEO_H264_LEVEL_4_0, 245760, 8192, 20000, 25000, 4},
72 {V4L2_MPEG_VIDEO_H264_LEVEL_4_1, 245760, 8192, 50000, 62500, 2},
73 {V4L2_MPEG_VIDEO_H264_LEVEL_4_2, 522240, 8704, 50000, 62500, 2},
74 {V4L2_MPEG_VIDEO_H264_LEVEL_5_0, 589824, 22080, 135000, 135000, 2},
75 {V4L2_MPEG_VIDEO_H264_LEVEL_5_1, 983040, 36864, 240000, 240000, 2}
76 };
77
78 enum hva_brc_type {
79 BRC_TYPE_NONE = 0,
80 BRC_TYPE_CBR = 1,
81 BRC_TYPE_VBR = 2,
82 BRC_TYPE_VBR_LOW_DELAY = 3
83 };
84
85 enum hva_entropy_coding_mode {
86 CAVLC = 0,
87 CABAC = 1
88 };
89
90 enum hva_picture_coding_type {
91 PICTURE_CODING_TYPE_I = 0,
92 PICTURE_CODING_TYPE_P = 1,
93 PICTURE_CODING_TYPE_B = 2
94 };
95
96 enum hva_h264_sampling_mode {
97 SAMPLING_MODE_NV12 = 0,
98 SAMPLING_MODE_UYVY = 1,
99 SAMPLING_MODE_RGB3 = 3,
100 SAMPLING_MODE_XRGB4 = 4,
101 SAMPLING_MODE_NV21 = 8,
102 SAMPLING_MODE_VYUY = 9,
103 SAMPLING_MODE_BGR3 = 11,
104 SAMPLING_MODE_XBGR4 = 12,
105 SAMPLING_MODE_RGBX4 = 20,
106 SAMPLING_MODE_BGRX4 = 28
107 };
108
109 enum hva_h264_nalu_type {
110 NALU_TYPE_UNKNOWN = 0,
111 NALU_TYPE_SLICE = 1,
112 NALU_TYPE_SLICE_DPA = 2,
113 NALU_TYPE_SLICE_DPB = 3,
114 NALU_TYPE_SLICE_DPC = 4,
115 NALU_TYPE_SLICE_IDR = 5,
116 NALU_TYPE_SEI = 6,
117 NALU_TYPE_SPS = 7,
118 NALU_TYPE_PPS = 8,
119 NALU_TYPE_AU_DELIMITER = 9,
120 NALU_TYPE_SEQ_END = 10,
121 NALU_TYPE_STREAM_END = 11,
122 NALU_TYPE_FILLER_DATA = 12,
123 NALU_TYPE_SPS_EXT = 13,
124 NALU_TYPE_PREFIX_UNIT = 14,
125 NALU_TYPE_SUBSET_SPS = 15,
126 NALU_TYPE_SLICE_AUX = 19,
127 NALU_TYPE_SLICE_EXT = 20
128 };
129
130 enum hva_h264_sei_payload_type {
131 SEI_BUFFERING_PERIOD = 0,
132 SEI_PICTURE_TIMING = 1,
133 SEI_STEREO_VIDEO_INFO = 21,
134 SEI_FRAME_PACKING_ARRANGEMENT = 45
135 };
136
137 /*
138 * stereo Video Info struct
139 */
140 struct hva_h264_stereo_video_sei {
141 u8 field_views_flag;
142 u8 top_field_is_left_view_flag;
143 u8 current_frame_is_left_view_flag;
144 u8 next_frame_is_second_view_flag;
145 u8 left_view_self_contained_flag;
146 u8 right_view_self_contained_flag;
147 };
148
149 /*
150 * struct hva_h264_td
151 *
152 * @frame_width: width in pixels of the buffer containing the input frame
153 * @frame_height: height in pixels of the buffer containing the input frame
154 * @frame_num: the parameter to be written in the slice header
155 * @picture_coding_type: type I, P or B
156 * @pic_order_cnt_type: POC mode, as defined in H264 std : can be 0,1,2
157 * @first_picture_in_sequence: flag telling to encoder that this is the
158 * first picture in a video sequence.
159 * Used for VBR
160 * @slice_size_type: 0 = no constraint to close the slice
161 * 1= a slice is closed as soon as the slice_mb_size limit
162 * is reached
163 * 2= a slice is closed as soon as the slice_byte_size limit
164 * is reached
165 * 3= a slice is closed as soon as either the slice_byte_size
166 * limit or the slice_mb_size limit is reached
167 * @slice_mb_size: defines the slice size in number of macroblocks
168 * (used when slice_size_type=1 or slice_size_type=3)
169 * @ir_param_option: defines the number of macroblocks per frame to be
170 * refreshed by AIR algorithm OR the refresh period
171 * by CIR algorithm
172 * @intra_refresh_type: enables the adaptive intra refresh algorithm.
173 * Disable=0 / Adaptative=1 and Cycle=2 as intra refresh
174 * @use_constrained_intra_flag: constrained_intra_pred_flag from PPS
175 * @transform_mode: controls the use of 4x4/8x8 transform mode
176 * @disable_deblocking_filter_idc:
177 * 0: specifies that all luma and chroma block edges of
178 * the slice are filtered.
179 * 1: specifies that deblocking is disabled for all block
180 * edges of the slice.
181 * 2: specifies that all luma and chroma block edges of
182 * the slice are filtered with exception of the block edges
183 * that coincide with slice boundaries
184 * @slice_alpha_c0_offset_div2: to be written in slice header,
185 * controls deblocking
186 * @slice_beta_offset_div2: to be written in slice header,
187 * controls deblocking
188 * @encoder_complexity: encoder complexity control (IME).
189 * 0 = I_16x16, P_16x16, Full ME Complexity
190 * 1 = I_16x16, I_NxN, P_16x16, Full ME Complexity
191 * 2 = I_16x16, I_NXN, P_16x16, P_WxH, Full ME Complexity
192 * 4 = I_16x16, P_16x16, Reduced ME Complexity
193 * 5 = I_16x16, I_NxN, P_16x16, Reduced ME Complexity
194 * 6 = I_16x16, I_NXN, P_16x16, P_WxH, Reduced ME Complexity
195 * @chroma_qp_index_offset: coming from picture parameter set
196 * (PPS see [H.264 STD] 7.4.2.2)
197 * @entropy_coding_mode: entropy coding mode.
198 * 0 = CAVLC
199 * 1 = CABAC
200 * @brc_type: selects the bit-rate control algorithm
201 * 0 = constant Qp, (no BRC)
202 * 1 = CBR
203 * 2 = VBR
204 * @quant: Quantization param used in case of fix QP encoding (no BRC)
205 * @non_VCL_NALU_Size: size of non-VCL NALUs (SPS, PPS, filler),
206 * used by BRC
207 * @cpb_buffer_size: size of Coded Picture Buffer, used by BRC
208 * @bit_rate: target bitrate, for BRC
209 * @qp_min: min QP threshold
210 * @qp_max: max QP threshold
211 * @framerate_num: target framerate numerator , used by BRC
212 * @framerate_den: target framerate denomurator , used by BRC
213 * @delay: End-to-End Initial Delay
214 * @strict_HRD_compliancy: flag for HDR compliancy (1)
215 * May impact quality encoding
216 * @addr_source_buffer: address of input frame buffer for current frame
217 * @addr_fwd_Ref_Buffer: address of reference frame buffer
218 * @addr_rec_buffer: address of reconstructed frame buffer
219 * @addr_output_bitstream_start: output bitstream start address
220 * @addr_output_bitstream_end: output bitstream end address
221 * @addr_external_sw : address of external search window
222 * @addr_lctx : address of context picture buffer
223 * @addr_local_rec_buffer: address of local reconstructed buffer
224 * @addr_spatial_context: address of spatial context buffer
225 * @bitstream_offset: offset in bits between aligned bitstream start
226 * address and first bit to be written by HVA.
227 * Range value is [0..63]
228 * @sampling_mode: Input picture format .
229 * 0: YUV420 semi_planar Interleaved
230 * 1: YUV422 raster Interleaved
231 * @addr_param_out: address of output parameters structure
232 * @addr_scaling_matrix: address to the coefficient of
233 * the inverse scaling matrix
234 * @addr_scaling_matrix_dir: address to the coefficient of
235 * the direct scaling matrix
236 * @addr_cabac_context_buffer: address of cabac context buffer
237 * @GmvX: Input information about the horizontal global displacement of
238 * the encoded frame versus the previous one
239 * @GmvY: Input information about the vertical global displacement of
240 * the encoded frame versus the previous one
241 * @window_width: width in pixels of the window to be encoded inside
242 * the input frame
243 * @window_height: width in pixels of the window to be encoded inside
244 * the input frame
245 * @window_horizontal_offset: horizontal offset in pels for input window
246 * within input frame
247 * @window_vertical_offset: vertical offset in pels for input window
248 * within input frame
249 * @addr_roi: Map of QP offset for the Region of Interest algorithm and
250 * also used for Error map.
251 * Bit 0-6 used for qp offset (value -64 to 63).
252 * Bit 7 used to force intra
253 * @addr_slice_header: address to slice header
254 * @slice_header_size_in_bits: size in bits of the Slice header
255 * @slice_header_offset0: Slice header offset where to insert
256 * first_Mb_in_slice
257 * @slice_header_offset1: Slice header offset where to insert
258 * slice_qp_delta
259 * @slice_header_offset2: Slice header offset where to insert
260 * num_MBs_in_slice
261 * @slice_synchro_enable: enable "slice ready" interrupt after each slice
262 * @max_slice_number: Maximum number of slice in a frame
263 * (0 is strictly forbidden)
264 * @rgb2_yuv_y_coeff: Four coefficients (C0C1C2C3) to convert from RGB to
265 * YUV for the Y component.
266 * Y = C0*R + C1*G + C2*B + C3 (C0 is on byte 0)
267 * @rgb2_yuv_u_coeff: four coefficients (C0C1C2C3) to convert from RGB to
268 * YUV for the Y component.
269 * Y = C0*R + C1*G + C2*B + C3 (C0 is on byte 0)
270 * @rgb2_yuv_v_coeff: Four coefficients (C0C1C2C3) to convert from RGB to
271 * YUV for the U (Cb) component.
272 * U = C0*R + C1*G + C2*B + C3 (C0 is on byte 0)
273 * @slice_byte_size: maximum slice size in bytes
274 * (used when slice_size_type=2 or slice_size_type=3)
275 * @max_air_intra_mb_nb: Maximum number of intra macroblock in a frame
276 * for the AIR algorithm
277 * @brc_no_skip: Disable skipping in the Bitrate Controller
278 * @addr_brc_in_out_parameter: address of static buffer for BRC parameters
279 */
280 struct hva_h264_td {
281 u16 frame_width;
282 u16 frame_height;
283 u32 frame_num;
284 u16 picture_coding_type;
285 u16 reserved1;
286 u16 pic_order_cnt_type;
287 u16 first_picture_in_sequence;
288 u16 slice_size_type;
289 u16 reserved2;
290 u32 slice_mb_size;
291 u16 ir_param_option;
292 u16 intra_refresh_type;
293 u16 use_constrained_intra_flag;
294 u16 transform_mode;
295 u16 disable_deblocking_filter_idc;
296 s16 slice_alpha_c0_offset_div2;
297 s16 slice_beta_offset_div2;
298 u16 encoder_complexity;
299 s16 chroma_qp_index_offset;
300 u16 entropy_coding_mode;
301 u16 brc_type;
302 u16 quant;
303 u32 non_vcl_nalu_size;
304 u32 cpb_buffer_size;
305 u32 bit_rate;
306 u16 qp_min;
307 u16 qp_max;
308 u16 framerate_num;
309 u16 framerate_den;
310 u16 delay;
311 u16 strict_hrd_compliancy;
312 u32 addr_source_buffer;
313 u32 addr_fwd_ref_buffer;
314 u32 addr_rec_buffer;
315 u32 addr_output_bitstream_start;
316 u32 addr_output_bitstream_end;
317 u32 addr_external_sw;
318 u32 addr_lctx;
319 u32 addr_local_rec_buffer;
320 u32 addr_spatial_context;
321 u16 bitstream_offset;
322 u16 sampling_mode;
323 u32 addr_param_out;
324 u32 addr_scaling_matrix;
325 u32 addr_scaling_matrix_dir;
326 u32 addr_cabac_context_buffer;
327 u32 reserved3;
328 u32 reserved4;
329 s16 gmv_x;
330 s16 gmv_y;
331 u16 window_width;
332 u16 window_height;
333 u16 window_horizontal_offset;
334 u16 window_vertical_offset;
335 u32 addr_roi;
336 u32 addr_slice_header;
337 u16 slice_header_size_in_bits;
338 u16 slice_header_offset0;
339 u16 slice_header_offset1;
340 u16 slice_header_offset2;
341 u32 reserved5;
342 u32 reserved6;
343 u16 reserved7;
344 u16 reserved8;
345 u16 slice_synchro_enable;
346 u16 max_slice_number;
347 u32 rgb2_yuv_y_coeff;
348 u32 rgb2_yuv_u_coeff;
349 u32 rgb2_yuv_v_coeff;
350 u32 slice_byte_size;
351 u16 max_air_intra_mb_nb;
352 u16 brc_no_skip;
353 u32 addr_temporal_context;
354 u32 addr_brc_in_out_parameter;
355 };
356
357 /*
358 * struct hva_h264_slice_po
359 *
360 * @ slice_size: slice size
361 * @ slice_start_time: start time
362 * @ slice_stop_time: stop time
363 * @ slice_num: slice number
364 */
365 struct hva_h264_slice_po {
366 u32 slice_size;
367 u32 slice_start_time;
368 u32 slice_end_time;
369 u32 slice_num;
370 };
371
372 /*
373 * struct hva_h264_po
374 *
375 * @ bitstream_size: bitstream size
376 * @ dct_bitstream_size: dtc bitstream size
377 * @ stuffing_bits: number of stuffing bits inserted by the encoder
378 * @ removal_time: removal time of current frame (nb of ticks 1/framerate)
379 * @ hvc_start_time: hvc start time
380 * @ hvc_stop_time: hvc stop time
381 * @ slice_count: slice count
382 */
383 struct hva_h264_po {
384 u32 bitstream_size;
385 u32 dct_bitstream_size;
386 u32 stuffing_bits;
387 u32 removal_time;
388 u32 hvc_start_time;
389 u32 hvc_stop_time;
390 u32 slice_count;
391 u32 reserved0;
392 struct hva_h264_slice_po slice_params[16];
393 };
394
395 struct hva_h264_task {
396 struct hva_h264_td td;
397 struct hva_h264_po po;
398 };
399
400 /*
401 * struct hva_h264_ctx
402 *
403 * @seq_info: sequence information buffer
404 * @ref_frame: reference frame buffer
405 * @rec_frame: reconstructed frame buffer
406 * @task: task descriptor
407 */
408 struct hva_h264_ctx {
409 struct hva_buffer *seq_info;
410 struct hva_buffer *ref_frame;
411 struct hva_buffer *rec_frame;
412 struct hva_buffer *task;
413 };
414
hva_h264_fill_slice_header(struct hva_ctx * pctx,u8 * slice_header_addr,struct hva_controls * ctrls,int frame_num,u16 * header_size,u16 * header_offset0,u16 * header_offset1,u16 * header_offset2)415 static int hva_h264_fill_slice_header(struct hva_ctx *pctx,
416 u8 *slice_header_addr,
417 struct hva_controls *ctrls,
418 int frame_num,
419 u16 *header_size,
420 u16 *header_offset0,
421 u16 *header_offset1,
422 u16 *header_offset2)
423 {
424 /*
425 * with this HVA hardware version, part of the slice header is computed
426 * on host and part by hardware.
427 * The part of host is precomputed and available through this array.
428 */
429 struct device *dev = ctx_to_dev(pctx);
430 int cabac = V4L2_MPEG_VIDEO_H264_ENTROPY_MODE_CABAC;
431 const unsigned char slice_header[] = { 0x00, 0x00, 0x00, 0x01,
432 0x41, 0x34, 0x07, 0x00};
433 int idr_pic_id = frame_num % 2;
434 enum hva_picture_coding_type type;
435 u32 frame_order = frame_num % ctrls->gop_size;
436
437 if (!(frame_num % ctrls->gop_size))
438 type = PICTURE_CODING_TYPE_I;
439 else
440 type = PICTURE_CODING_TYPE_P;
441
442 memcpy(slice_header_addr, slice_header, sizeof(slice_header));
443
444 *header_size = 56;
445 *header_offset0 = 40;
446 *header_offset1 = 13;
447 *header_offset2 = 0;
448
449 if (type == PICTURE_CODING_TYPE_I) {
450 slice_header_addr[4] = 0x65;
451 slice_header_addr[5] = 0x11;
452
453 /* toggle the I frame */
454 if ((frame_num / ctrls->gop_size) % 2) {
455 *header_size += 4;
456 *header_offset1 += 4;
457 slice_header_addr[6] = 0x04;
458 slice_header_addr[7] = 0x70;
459
460 } else {
461 *header_size += 2;
462 *header_offset1 += 2;
463 slice_header_addr[6] = 0x09;
464 slice_header_addr[7] = 0xC0;
465 }
466 } else {
467 if (ctrls->entropy_mode == cabac) {
468 *header_size += 1;
469 *header_offset1 += 1;
470 slice_header_addr[7] = 0x80;
471 }
472 /*
473 * update slice header with P frame order
474 * frame order is limited to 16 (coded on 4bits only)
475 */
476 slice_header_addr[5] += ((frame_order & 0x0C) >> 2);
477 slice_header_addr[6] += ((frame_order & 0x03) << 6);
478 }
479
480 dev_dbg(dev,
481 "%s %s slice header order %d idrPicId %d header size %d\n",
482 pctx->name, __func__, frame_order, idr_pic_id, *header_size);
483 return 0;
484 }
485
hva_h264_fill_data_nal(struct hva_ctx * pctx,unsigned int stuffing_bytes,u8 * addr,unsigned int stream_size,unsigned int * size)486 static int hva_h264_fill_data_nal(struct hva_ctx *pctx,
487 unsigned int stuffing_bytes, u8 *addr,
488 unsigned int stream_size, unsigned int *size)
489 {
490 struct device *dev = ctx_to_dev(pctx);
491 const u8 start[] = { 0x00, 0x00, 0x00, 0x01 };
492
493 dev_dbg(dev, "%s %s stuffing bytes %d\n", pctx->name, __func__,
494 stuffing_bytes);
495
496 if ((*size + stuffing_bytes + H264_FILLER_DATA_SIZE) > stream_size) {
497 dev_dbg(dev, "%s %s too many stuffing bytes %d\n",
498 pctx->name, __func__, stuffing_bytes);
499 return 0;
500 }
501
502 /* start code */
503 memcpy(addr + *size, start, sizeof(start));
504 *size += sizeof(start);
505
506 /* nal_unit_type */
507 addr[*size] = NALU_TYPE_FILLER_DATA;
508 *size += 1;
509
510 memset(addr + *size, 0xff, stuffing_bytes);
511 *size += stuffing_bytes;
512
513 addr[*size] = 0x80;
514 *size += 1;
515
516 return 0;
517 }
518
hva_h264_fill_sei_nal(struct hva_ctx * pctx,enum hva_h264_sei_payload_type type,u8 * addr,u32 * size)519 static int hva_h264_fill_sei_nal(struct hva_ctx *pctx,
520 enum hva_h264_sei_payload_type type,
521 u8 *addr, u32 *size)
522 {
523 struct device *dev = ctx_to_dev(pctx);
524 const u8 start[] = { 0x00, 0x00, 0x00, 0x01 };
525 struct hva_h264_stereo_video_sei info;
526 u8 offset = 7;
527 u8 msg = 0;
528
529 /* start code */
530 memcpy(addr + *size, start, sizeof(start));
531 *size += sizeof(start);
532
533 /* nal_unit_type */
534 addr[*size] = NALU_TYPE_SEI;
535 *size += 1;
536
537 /* payload type */
538 addr[*size] = type;
539 *size += 1;
540
541 switch (type) {
542 case SEI_STEREO_VIDEO_INFO:
543 memset(&info, 0, sizeof(info));
544
545 /* set to top/bottom frame packing arrangement */
546 info.field_views_flag = 1;
547 info.top_field_is_left_view_flag = 1;
548
549 /* payload size */
550 addr[*size] = 1;
551 *size += 1;
552
553 /* payload */
554 msg = info.field_views_flag << offset--;
555
556 if (info.field_views_flag) {
557 msg |= info.top_field_is_left_view_flag <<
558 offset--;
559 } else {
560 msg |= info.current_frame_is_left_view_flag <<
561 offset--;
562 msg |= info.next_frame_is_second_view_flag <<
563 offset--;
564 }
565 msg |= info.left_view_self_contained_flag << offset--;
566 msg |= info.right_view_self_contained_flag << offset--;
567
568 addr[*size] = msg;
569 *size += 1;
570
571 addr[*size] = 0x80;
572 *size += 1;
573
574 return 0;
575 case SEI_BUFFERING_PERIOD:
576 case SEI_PICTURE_TIMING:
577 case SEI_FRAME_PACKING_ARRANGEMENT:
578 default:
579 dev_err(dev, "%s sei nal type not supported %d\n",
580 pctx->name, type);
581 return -EINVAL;
582 }
583 }
584
hva_h264_prepare_task(struct hva_ctx * pctx,struct hva_h264_task * task,struct hva_frame * frame,struct hva_stream * stream)585 static int hva_h264_prepare_task(struct hva_ctx *pctx,
586 struct hva_h264_task *task,
587 struct hva_frame *frame,
588 struct hva_stream *stream)
589 {
590 struct hva_dev *hva = ctx_to_hdev(pctx);
591 struct device *dev = ctx_to_dev(pctx);
592 struct hva_h264_ctx *ctx = (struct hva_h264_ctx *)pctx->priv;
593 struct hva_buffer *seq_info = ctx->seq_info;
594 struct hva_buffer *fwd_ref_frame = ctx->ref_frame;
595 struct hva_buffer *loc_rec_frame = ctx->rec_frame;
596 struct hva_h264_td *td = &task->td;
597 struct hva_controls *ctrls = &pctx->ctrls;
598 struct v4l2_fract *time_per_frame = &pctx->ctrls.time_per_frame;
599 int cavlc = V4L2_MPEG_VIDEO_H264_ENTROPY_MODE_CAVLC;
600 u32 frame_num = pctx->stream_num;
601 u32 addr_esram = hva->esram_addr;
602 enum v4l2_mpeg_video_h264_level level;
603 dma_addr_t paddr = 0;
604 u8 *slice_header_vaddr;
605 u32 frame_width = frame->info.aligned_width;
606 u32 frame_height = frame->info.aligned_height;
607 u32 max_cpb_buffer_size;
608 unsigned int payload = stream->bytesused;
609 u32 max_bitrate;
610
611 /* check width and height parameters */
612 if ((frame_width > max(H264_MAX_SIZE_W, H264_MAX_SIZE_H)) ||
613 (frame_height > max(H264_MAX_SIZE_W, H264_MAX_SIZE_H))) {
614 dev_err(dev,
615 "%s width(%d) or height(%d) exceeds limits (%dx%d)\n",
616 pctx->name, frame_width, frame_height,
617 H264_MAX_SIZE_W, H264_MAX_SIZE_H);
618 pctx->frame_errors++;
619 return -EINVAL;
620 }
621
622 level = ctrls->level;
623
624 memset(td, 0, sizeof(struct hva_h264_td));
625
626 td->frame_width = frame_width;
627 td->frame_height = frame_height;
628
629 /* set frame alignment */
630 td->window_width = frame_width;
631 td->window_height = frame_height;
632 td->window_horizontal_offset = 0;
633 td->window_vertical_offset = 0;
634
635 td->first_picture_in_sequence = (!frame_num) ? 1 : 0;
636
637 /* pic_order_cnt_type hard coded to '2' as only I & P frames */
638 td->pic_order_cnt_type = 2;
639
640 /* useConstrainedIntraFlag set to false for better coding efficiency */
641 td->use_constrained_intra_flag = false;
642 td->brc_type = (ctrls->bitrate_mode == V4L2_MPEG_VIDEO_BITRATE_MODE_CBR)
643 ? BRC_TYPE_CBR : BRC_TYPE_VBR;
644
645 td->entropy_coding_mode = (ctrls->entropy_mode == cavlc) ? CAVLC :
646 CABAC;
647
648 td->bit_rate = ctrls->bitrate;
649
650 /* set framerate, framerate = 1 n/ time per frame */
651 if (time_per_frame->numerator >= 536) {
652 /*
653 * due to a hardware bug, framerate denominator can't exceed
654 * 536 (BRC overflow). Compute nearest framerate
655 */
656 td->framerate_den = 1;
657 td->framerate_num = (time_per_frame->denominator +
658 (time_per_frame->numerator >> 1) - 1) /
659 time_per_frame->numerator;
660
661 /*
662 * update bitrate to introduce a correction due to
663 * the new framerate
664 * new bitrate = (old bitrate * new framerate) / old framerate
665 */
666 td->bit_rate /= time_per_frame->numerator;
667 td->bit_rate *= time_per_frame->denominator;
668 td->bit_rate /= td->framerate_num;
669 } else {
670 td->framerate_den = time_per_frame->numerator;
671 td->framerate_num = time_per_frame->denominator;
672 }
673
674 /* compute maximum bitrate depending on profile */
675 if (ctrls->profile >= V4L2_MPEG_VIDEO_H264_PROFILE_HIGH)
676 max_bitrate = h264_infos_list[level].max_bitrate *
677 H264_FACTOR_HIGH;
678 else
679 max_bitrate = h264_infos_list[level].max_bitrate *
680 H264_FACTOR_BASELINE;
681
682 /* check if bitrate doesn't exceed max size */
683 if (td->bit_rate > max_bitrate) {
684 dev_dbg(dev,
685 "%s bitrate (%d) larger than level and profile allow, clip to %d\n",
686 pctx->name, td->bit_rate, max_bitrate);
687 td->bit_rate = max_bitrate;
688 }
689
690 /* convert cpb_buffer_size in bits */
691 td->cpb_buffer_size = ctrls->cpb_size * 8000;
692
693 /* compute maximum cpb buffer size depending on profile */
694 if (ctrls->profile >= V4L2_MPEG_VIDEO_H264_PROFILE_HIGH)
695 max_cpb_buffer_size =
696 h264_infos_list[level].max_cpb_size * H264_FACTOR_HIGH;
697 else
698 max_cpb_buffer_size =
699 h264_infos_list[level].max_cpb_size * H264_FACTOR_BASELINE;
700
701 /* check if cpb buffer size doesn't exceed max size */
702 if (td->cpb_buffer_size > max_cpb_buffer_size) {
703 dev_dbg(dev,
704 "%s cpb size larger than level %d allows, clip to %d\n",
705 pctx->name, td->cpb_buffer_size, max_cpb_buffer_size);
706 td->cpb_buffer_size = max_cpb_buffer_size;
707 }
708
709 /* enable skipping in the Bitrate Controller */
710 td->brc_no_skip = 0;
711
712 /* initial delay */
713 if ((ctrls->bitrate_mode == V4L2_MPEG_VIDEO_BITRATE_MODE_CBR) &&
714 td->bit_rate)
715 td->delay = 1000 * (td->cpb_buffer_size / td->bit_rate);
716 else
717 td->delay = 0;
718
719 switch (frame->info.pixelformat) {
720 case V4L2_PIX_FMT_NV12:
721 td->sampling_mode = SAMPLING_MODE_NV12;
722 break;
723 case V4L2_PIX_FMT_NV21:
724 td->sampling_mode = SAMPLING_MODE_NV21;
725 break;
726 default:
727 dev_err(dev, "%s invalid source pixel format\n",
728 pctx->name);
729 pctx->frame_errors++;
730 return -EINVAL;
731 }
732
733 /*
734 * fill matrix color converter (RGB to YUV)
735 * Y = 0,299 R + 0,587 G + 0,114 B
736 * Cb = -0,1687 R -0,3313 G + 0,5 B + 128
737 * Cr = 0,5 R - 0,4187 G - 0,0813 B + 128
738 */
739 td->rgb2_yuv_y_coeff = 0x12031008;
740 td->rgb2_yuv_u_coeff = 0x800EF7FB;
741 td->rgb2_yuv_v_coeff = 0x80FEF40E;
742
743 /* enable/disable transform mode */
744 td->transform_mode = ctrls->dct8x8;
745
746 /* encoder complexity fix to 2, ENCODE_I_16x16_I_NxN_P_16x16_P_WxH */
747 td->encoder_complexity = 2;
748
749 /* quant fix to 28, default VBR value */
750 td->quant = 28;
751
752 if (td->framerate_den == 0) {
753 dev_err(dev, "%s invalid framerate\n", pctx->name);
754 pctx->frame_errors++;
755 return -EINVAL;
756 }
757
758 /* if automatic framerate, deactivate bitrate controller */
759 if (td->framerate_num == 0)
760 td->brc_type = 0;
761
762 /* compliancy fix to true */
763 td->strict_hrd_compliancy = 1;
764
765 /* set minimum & maximum quantizers */
766 td->qp_min = clamp_val(ctrls->qpmin, 0, 51);
767 td->qp_max = clamp_val(ctrls->qpmax, 0, 51);
768
769 td->addr_source_buffer = frame->paddr;
770 td->addr_fwd_ref_buffer = fwd_ref_frame->paddr;
771 td->addr_rec_buffer = loc_rec_frame->paddr;
772
773 td->addr_output_bitstream_end = (u32)stream->paddr + stream->size;
774
775 td->addr_output_bitstream_start = (u32)stream->paddr;
776 td->bitstream_offset = (((u32)stream->paddr & 0xF) << 3) &
777 BITSTREAM_OFFSET_MASK;
778
779 td->addr_param_out = (u32)ctx->task->paddr +
780 offsetof(struct hva_h264_task, po);
781
782 /* swap spatial and temporal context */
783 if (frame_num % 2) {
784 paddr = seq_info->paddr;
785 td->addr_spatial_context = ALIGN(paddr, 0x100);
786 paddr = seq_info->paddr + DATA_SIZE(frame_width,
787 frame_height);
788 td->addr_temporal_context = ALIGN(paddr, 0x100);
789 } else {
790 paddr = seq_info->paddr;
791 td->addr_temporal_context = ALIGN(paddr, 0x100);
792 paddr = seq_info->paddr + DATA_SIZE(frame_width,
793 frame_height);
794 td->addr_spatial_context = ALIGN(paddr, 0x100);
795 }
796
797 paddr = seq_info->paddr + 2 * DATA_SIZE(frame_width, frame_height);
798
799 td->addr_brc_in_out_parameter = ALIGN(paddr, 0x100);
800
801 paddr = td->addr_brc_in_out_parameter + BRC_DATA_SIZE;
802 td->addr_slice_header = ALIGN(paddr, 0x100);
803 td->addr_external_sw = ALIGN(addr_esram, 0x100);
804
805 addr_esram += SEARCH_WINDOW_BUFFER_MAX_SIZE(frame_width);
806 td->addr_local_rec_buffer = ALIGN(addr_esram, 0x100);
807
808 addr_esram += LOCAL_RECONSTRUCTED_BUFFER_MAX_SIZE(frame_width);
809 td->addr_lctx = ALIGN(addr_esram, 0x100);
810
811 addr_esram += CTX_MB_BUFFER_MAX_SIZE(max(frame_width, frame_height));
812 td->addr_cabac_context_buffer = ALIGN(addr_esram, 0x100);
813
814 if (!(frame_num % ctrls->gop_size)) {
815 td->picture_coding_type = PICTURE_CODING_TYPE_I;
816 stream->vbuf.flags |= V4L2_BUF_FLAG_KEYFRAME;
817 } else {
818 td->picture_coding_type = PICTURE_CODING_TYPE_P;
819 stream->vbuf.flags &= ~V4L2_BUF_FLAG_KEYFRAME;
820 }
821
822 /* fill the slice header part */
823 slice_header_vaddr = seq_info->vaddr + (td->addr_slice_header -
824 seq_info->paddr);
825
826 hva_h264_fill_slice_header(pctx, slice_header_vaddr, ctrls, frame_num,
827 &td->slice_header_size_in_bits,
828 &td->slice_header_offset0,
829 &td->slice_header_offset1,
830 &td->slice_header_offset2);
831
832 td->chroma_qp_index_offset = 2;
833 td->slice_synchro_enable = 0;
834 td->max_slice_number = 1;
835
836 /*
837 * check the sps/pps header size for key frame only
838 * sps/pps header was previously fill by libv4l
839 * during qbuf of stream buffer
840 */
841 if ((stream->vbuf.flags == V4L2_BUF_FLAG_KEYFRAME) &&
842 (payload > MAX_SPS_PPS_SIZE)) {
843 dev_err(dev, "%s invalid sps/pps size %d\n", pctx->name,
844 payload);
845 pctx->frame_errors++;
846 return -EINVAL;
847 }
848
849 if (stream->vbuf.flags != V4L2_BUF_FLAG_KEYFRAME)
850 payload = 0;
851
852 /* add SEI nal (video stereo info) */
853 if (ctrls->sei_fp && hva_h264_fill_sei_nal(pctx, SEI_STEREO_VIDEO_INFO,
854 (u8 *)stream->vaddr,
855 &payload)) {
856 dev_err(dev, "%s fail to get SEI nal\n", pctx->name);
857 pctx->frame_errors++;
858 return -EINVAL;
859 }
860
861 /* fill size of non-VCL NAL units (SPS, PPS, filler and SEI) */
862 td->non_vcl_nalu_size = payload * 8;
863
864 /* compute bitstream offset & new start address of bitstream */
865 td->addr_output_bitstream_start += ((payload >> 4) << 4);
866 td->bitstream_offset += (payload - ((payload >> 4) << 4)) * 8;
867
868 stream->bytesused = payload;
869
870 return 0;
871 }
872
hva_h264_get_stream_size(struct hva_h264_task * task)873 static unsigned int hva_h264_get_stream_size(struct hva_h264_task *task)
874 {
875 struct hva_h264_po *po = &task->po;
876
877 return po->bitstream_size;
878 }
879
hva_h264_get_stuffing_bytes(struct hva_h264_task * task)880 static u32 hva_h264_get_stuffing_bytes(struct hva_h264_task *task)
881 {
882 struct hva_h264_po *po = &task->po;
883
884 return po->stuffing_bits >> 3;
885 }
886
hva_h264_open(struct hva_ctx * pctx)887 static int hva_h264_open(struct hva_ctx *pctx)
888 {
889 struct device *dev = ctx_to_dev(pctx);
890 struct hva_h264_ctx *ctx;
891 struct hva_dev *hva = ctx_to_hdev(pctx);
892 u32 frame_width = pctx->frameinfo.aligned_width;
893 u32 frame_height = pctx->frameinfo.aligned_height;
894 u32 size;
895 int ret;
896
897 /* check esram size necessary to encode a frame */
898 size = SEARCH_WINDOW_BUFFER_MAX_SIZE(frame_width) +
899 LOCAL_RECONSTRUCTED_BUFFER_MAX_SIZE(frame_width) +
900 CTX_MB_BUFFER_MAX_SIZE(max(frame_width, frame_height)) +
901 CABAC_CONTEXT_BUFFER_MAX_SIZE(frame_width);
902
903 if (hva->esram_size < size) {
904 dev_err(dev, "%s not enough esram (max:%d request:%d)\n",
905 pctx->name, hva->esram_size, size);
906 ret = -EINVAL;
907 goto err;
908 }
909
910 /* allocate context for codec */
911 ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL);
912 if (!ctx) {
913 ret = -ENOMEM;
914 goto err;
915 }
916
917 /* allocate sequence info buffer */
918 ret = hva_mem_alloc(pctx,
919 2 * DATA_SIZE(frame_width, frame_height) +
920 SLICE_HEADER_SIZE +
921 BRC_DATA_SIZE,
922 "hva sequence info",
923 &ctx->seq_info);
924 if (ret) {
925 dev_err(dev,
926 "%s failed to allocate sequence info buffer\n",
927 pctx->name);
928 goto err_ctx;
929 }
930
931 /* allocate reference frame buffer */
932 ret = hva_mem_alloc(pctx,
933 frame_width * frame_height * 3 / 2,
934 "hva reference frame",
935 &ctx->ref_frame);
936 if (ret) {
937 dev_err(dev, "%s failed to allocate reference frame buffer\n",
938 pctx->name);
939 goto err_seq_info;
940 }
941
942 /* allocate reconstructed frame buffer */
943 ret = hva_mem_alloc(pctx,
944 frame_width * frame_height * 3 / 2,
945 "hva reconstructed frame",
946 &ctx->rec_frame);
947 if (ret) {
948 dev_err(dev,
949 "%s failed to allocate reconstructed frame buffer\n",
950 pctx->name);
951 goto err_ref_frame;
952 }
953
954 /* allocate task descriptor */
955 ret = hva_mem_alloc(pctx,
956 sizeof(struct hva_h264_task),
957 "hva task descriptor",
958 &ctx->task);
959 if (ret) {
960 dev_err(dev,
961 "%s failed to allocate task descriptor\n",
962 pctx->name);
963 goto err_rec_frame;
964 }
965
966 pctx->priv = (void *)ctx;
967
968 return 0;
969
970 err_rec_frame:
971 hva_mem_free(pctx, ctx->rec_frame);
972 err_ref_frame:
973 hva_mem_free(pctx, ctx->ref_frame);
974 err_seq_info:
975 hva_mem_free(pctx, ctx->seq_info);
976 err_ctx:
977 devm_kfree(dev, ctx);
978 err:
979 pctx->sys_errors++;
980 return ret;
981 }
982
hva_h264_close(struct hva_ctx * pctx)983 static int hva_h264_close(struct hva_ctx *pctx)
984 {
985 struct hva_h264_ctx *ctx = (struct hva_h264_ctx *)pctx->priv;
986 struct device *dev = ctx_to_dev(pctx);
987
988 if (ctx->seq_info)
989 hva_mem_free(pctx, ctx->seq_info);
990
991 if (ctx->ref_frame)
992 hva_mem_free(pctx, ctx->ref_frame);
993
994 if (ctx->rec_frame)
995 hva_mem_free(pctx, ctx->rec_frame);
996
997 if (ctx->task)
998 hva_mem_free(pctx, ctx->task);
999
1000 devm_kfree(dev, ctx);
1001
1002 return 0;
1003 }
1004
hva_h264_encode(struct hva_ctx * pctx,struct hva_frame * frame,struct hva_stream * stream)1005 static int hva_h264_encode(struct hva_ctx *pctx, struct hva_frame *frame,
1006 struct hva_stream *stream)
1007 {
1008 struct hva_h264_ctx *ctx = (struct hva_h264_ctx *)pctx->priv;
1009 struct hva_h264_task *task = (struct hva_h264_task *)ctx->task->vaddr;
1010 u32 stuffing_bytes = 0;
1011 int ret = 0;
1012
1013 ret = hva_h264_prepare_task(pctx, task, frame, stream);
1014 if (ret)
1015 goto err;
1016
1017 ret = hva_hw_execute_task(pctx, H264_ENC, ctx->task);
1018 if (ret)
1019 goto err;
1020
1021 pctx->stream_num++;
1022 stream->bytesused += hva_h264_get_stream_size(task);
1023
1024 stuffing_bytes = hva_h264_get_stuffing_bytes(task);
1025
1026 if (stuffing_bytes)
1027 hva_h264_fill_data_nal(pctx, stuffing_bytes,
1028 (u8 *)stream->vaddr,
1029 stream->size,
1030 &stream->bytesused);
1031
1032 /* switch reference & reconstructed frame */
1033 swap(ctx->ref_frame, ctx->rec_frame);
1034
1035 return 0;
1036 err:
1037 stream->bytesused = 0;
1038 return ret;
1039 }
1040
1041 const struct hva_enc nv12h264enc = {
1042 .name = "H264(NV12)",
1043 .pixelformat = V4L2_PIX_FMT_NV12,
1044 .streamformat = V4L2_PIX_FMT_H264,
1045 .max_width = H264_MAX_SIZE_W,
1046 .max_height = H264_MAX_SIZE_H,
1047 .open = hva_h264_open,
1048 .close = hva_h264_close,
1049 .encode = hva_h264_encode,
1050 };
1051
1052 const struct hva_enc nv21h264enc = {
1053 .name = "H264(NV21)",
1054 .pixelformat = V4L2_PIX_FMT_NV21,
1055 .streamformat = V4L2_PIX_FMT_H264,
1056 .max_width = H264_MAX_SIZE_W,
1057 .max_height = H264_MAX_SIZE_H,
1058 .open = hva_h264_open,
1059 .close = hva_h264_close,
1060 .encode = hva_h264_encode,
1061 };
1062