1
/**************************************************************************
3
* Copyright 2013 Advanced Micro Devices, Inc.
6
* Permission is hereby granted, free of charge, to any person obtaining a
7
* copy of this software and associated documentation files (the
8
* "Software"), to deal in the Software without restriction, including
9
* without limitation the rights to use, copy, modify, merge, publish,
10
* distribute, sub license, and/or sell copies of the Software, and to
11
* permit persons to whom the Software is furnished to do so, subject to
12
* the following conditions:
14
* The above copyright notice and this permission notice (including the
15
* next paragraph) shall be included in all copies or substantial portions
18
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21
* IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR
22
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26
**************************************************************************/
31
#include "radeon_video.h"
32
#include "util/list.h"
34
#define RVCE_CS(value) (enc->cs.current.buf[enc->cs.current.cdw++] = (value))
35
#define RVCE_BEGIN(cmd) \
37
uint32_t *begin = &enc->cs.current.buf[enc->cs.current.cdw++]; \
39
#define RVCE_READ(buf, domain, off) \
40
si_vce_add_buffer(enc, (buf), RADEON_USAGE_READ, (domain), (off))
41
#define RVCE_WRITE(buf, domain, off) \
42
si_vce_add_buffer(enc, (buf), RADEON_USAGE_WRITE, (domain), (off))
43
#define RVCE_READWRITE(buf, domain, off) \
44
si_vce_add_buffer(enc, (buf), RADEON_USAGE_READWRITE, (domain), (off))
46
*begin = (&enc->cs.current.buf[enc->cs.current.cdw] - begin) * 4; \
49
#define RVCE_MAX_BITSTREAM_OUTPUT_ROW_SIZE (4096 * 16 * 2.5)
50
#define RVCE_MAX_AUX_BUFFER_NUM 4
54
/* driver dependent callback */
55
typedef void (*rvce_get_buffer)(struct pipe_resource *resource, struct pb_buffer **handle,
56
struct radeon_surf **surface);
58
/* Coded picture buffer slot */
59
struct rvce_cpb_slot {
60
struct list_head list;
63
enum pipe_h2645_enc_picture_type picture_type;
65
unsigned pic_order_cnt;
68
struct rvce_rate_control {
70
uint32_t target_bitrate;
71
uint32_t peak_bitrate;
72
uint32_t frame_rate_num;
74
uint32_t quant_i_frames;
75
uint32_t quant_p_frames;
76
uint32_t quant_b_frames;
77
uint32_t vbv_buffer_size;
78
uint32_t frame_rate_den;
81
uint32_t qp_initial_mode;
82
uint32_t target_bits_picture;
83
uint32_t peak_bits_picture_integer;
84
uint32_t peak_bits_picture_fraction;
87
uint32_t skip_frame_enable;
88
uint32_t fill_data_enable;
90
uint32_t b_pics_delta_qp;
91
uint32_t ref_b_pics_delta_qp;
92
uint32_t rc_reinit_disable;
93
uint32_t enc_lcvbr_init_qp_flag;
94
uint32_t lcvbrsatd_based_nonlinear_bit_budget_flag;
97
struct rvce_motion_estimation {
98
uint32_t enc_ime_decimation_search;
99
uint32_t motion_est_half_pixel;
100
uint32_t motion_est_quarter_pixel;
101
uint32_t disable_favor_pmv_point;
102
uint32_t force_zero_point_center;
104
uint32_t enc_search_range_x;
105
uint32_t enc_search_range_y;
106
uint32_t enc_search1_range_x;
107
uint32_t enc_search1_range_y;
108
uint32_t disable_16x16_frame1;
109
uint32_t disable_satd;
111
uint32_t enc_disable_sub_mode;
112
uint32_t enc_ime_skip_x;
113
uint32_t enc_ime_skip_y;
114
uint32_t enc_en_ime_overw_dis_subm;
115
uint32_t enc_ime_overw_dis_subm_no;
116
uint32_t enc_ime2_search_range_x;
117
uint32_t enc_ime2_search_range_y;
118
uint32_t parallel_mode_speedup_enable;
119
uint32_t fme0_enc_disable_sub_mode;
120
uint32_t fme1_enc_disable_sub_mode;
121
uint32_t ime_sw_speedup_enable;
124
struct rvce_pic_control {
125
uint32_t enc_use_constrained_intra_pred;
126
uint32_t enc_cabac_enable;
127
uint32_t enc_cabac_idc;
128
uint32_t enc_loop_filter_disable;
129
int32_t enc_lf_beta_offset;
130
int32_t enc_lf_alpha_c0_offset;
131
uint32_t enc_crop_left_offset;
132
uint32_t enc_crop_right_offset;
133
uint32_t enc_crop_top_offset;
134
uint32_t enc_crop_bottom_offset;
135
uint32_t enc_num_mbs_per_slice;
136
uint32_t enc_intra_refresh_num_mbs_per_slot;
137
uint32_t enc_force_intra_refresh;
138
uint32_t enc_force_imb_period;
139
uint32_t enc_pic_order_cnt_type;
140
uint32_t log2_max_pic_order_cnt_lsb_minus4;
143
uint32_t enc_constraint_set_flags;
144
uint32_t enc_b_pic_pattern;
145
uint32_t weight_pred_mode_b_picture;
146
uint32_t enc_number_of_reference_frames;
147
uint32_t enc_max_num_ref_frames;
148
uint32_t enc_num_default_active_ref_l0;
149
uint32_t enc_num_default_active_ref_l1;
150
uint32_t enc_slice_mode;
151
uint32_t enc_max_slice_size;
154
struct rvce_task_info {
155
uint32_t offset_of_next_task_info;
156
uint32_t task_operation;
157
uint32_t reference_picture_dependency;
158
uint32_t collocate_flag_dependency;
159
uint32_t feedback_index;
160
uint32_t video_bitstream_ring_index;
163
struct rvce_feedback_buf_pkg {
164
uint32_t feedback_ring_address_hi;
165
uint32_t feedback_ring_address_lo;
166
uint32_t feedback_ring_size;
170
uint32_t enc_disable_tbe_pred_i_frame;
171
uint32_t enc_disable_tbe_pred_p_frame;
172
uint32_t use_fme_interpol_y;
173
uint32_t use_fme_interpol_uv;
174
uint32_t use_fme_intrapol_y;
175
uint32_t use_fme_intrapol_uv;
176
uint32_t use_fme_interpol_y_1;
177
uint32_t use_fme_interpol_uv_1;
178
uint32_t use_fme_intrapol_y_1;
179
uint32_t use_fme_intrapol_uv_1;
180
uint32_t enc_16x16_cost_adj;
181
uint32_t enc_skip_cost_adj;
182
uint32_t enc_force_16x16_skip;
183
uint32_t enc_disable_threshold_calc_a;
184
uint32_t enc_luma_coeff_cost;
185
uint32_t enc_luma_mb_coeff_cost;
186
uint32_t enc_chroma_coeff_cost;
190
uint32_t aspect_ratio_info_present_flag;
191
uint32_t aspect_ratio_idc;
194
uint32_t overscan_info_present_flag;
195
uint32_t overscan_Approp_flag;
196
uint32_t video_signal_type_present_flag;
197
uint32_t video_format;
198
uint32_t video_full_range_flag;
199
uint32_t color_description_present_flag;
201
uint32_t transfer_char;
202
uint32_t matrix_coef;
203
uint32_t chroma_loc_info_present_flag;
204
uint32_t chroma_loc_top;
205
uint32_t chroma_loc_bottom;
206
uint32_t timing_info_present_flag;
207
uint32_t num_units_in_tick;
209
uint32_t fixed_frame_rate_flag;
210
uint32_t nal_hrd_parameters_present_flag;
211
uint32_t cpb_cnt_minus1;
212
uint32_t bit_rate_scale;
213
uint32_t cpb_size_scale;
214
uint32_t bit_rate_value_minus;
215
uint32_t cpb_size_value_minus;
217
uint32_t initial_cpb_removal_delay_length_minus1;
218
uint32_t cpb_removal_delay_length_minus1;
219
uint32_t dpb_output_delay_length_minus1;
220
uint32_t time_offset_length;
221
uint32_t low_delay_hrd_flag;
222
uint32_t pic_struct_present_flag;
223
uint32_t bitstream_restriction_present_flag;
224
uint32_t motion_vectors_over_pic_boundaries_flag;
225
uint32_t max_bytes_per_pic_denom;
226
uint32_t max_bits_per_mb_denom;
227
uint32_t log2_max_mv_length_hori;
228
uint32_t log2_max_mv_length_vert;
229
uint32_t num_reorder_frames;
230
uint32_t max_dec_frame_buffering;
233
struct rvce_enc_operation {
234
uint32_t insert_headers;
235
uint32_t picture_structure;
236
uint32_t allowed_max_bitstream_size;
237
uint32_t force_refresh_map;
239
uint32_t end_of_sequence;
240
uint32_t end_of_stream;
241
uint32_t input_picture_luma_address_hi;
242
uint32_t input_picture_luma_address_lo;
243
uint32_t input_picture_chroma_address_hi;
244
uint32_t input_picture_chroma_address_lo;
245
uint32_t enc_input_frame_y_pitch;
246
uint32_t enc_input_pic_luma_pitch;
247
uint32_t enc_input_pic_chroma_pitch;
249
uint32_t enc_input_pic_addr_array;
250
uint32_t enc_input_pic_addr_array_disable2pipe_disablemboffload;
251
uint32_t enc_input_pic_tile_config;
252
uint32_t enc_pic_type;
253
uint32_t enc_idr_flag;
254
uint32_t enc_idr_pic_id;
255
uint32_t enc_mgs_key_pic;
256
uint32_t enc_reference_flag;
257
uint32_t enc_temporal_layer_index;
258
uint32_t num_ref_idx_active_override_flag;
259
uint32_t num_ref_idx_l0_active_minus1;
260
uint32_t num_ref_idx_l1_active_minus1;
261
uint32_t enc_ref_list_modification_op;
262
uint32_t enc_ref_list_modification_num;
263
uint32_t enc_decoded_picture_marking_op;
264
uint32_t enc_decoded_picture_marking_num;
265
uint32_t enc_decoded_picture_marking_idx;
266
uint32_t enc_decoded_ref_base_picture_marking_op;
267
uint32_t enc_decoded_ref_base_picture_marking_num;
268
uint32_t l0_picture_structure;
269
uint32_t l0_enc_pic_type;
270
uint32_t l0_frame_number;
271
uint32_t l0_picture_order_count;
272
uint32_t l0_luma_offset;
273
uint32_t l0_chroma_offset;
274
uint32_t l1_picture_structure;
275
uint32_t l1_enc_pic_type;
276
uint32_t l1_frame_number;
277
uint32_t l1_picture_order_count;
278
uint32_t l1_luma_offset;
279
uint32_t l1_chroma_offset;
280
uint32_t enc_reconstructed_luma_offset;
281
uint32_t enc_reconstructed_chroma_offset;
283
uint32_t enc_coloc_buffer_offset;
284
uint32_t enc_reconstructed_ref_base_picture_luma_offset;
285
uint32_t enc_reconstructed_ref_base_picture_chroma_offset;
286
uint32_t enc_reference_ref_base_picture_luma_offset;
287
uint32_t enc_reference_ref_base_picture_chroma_offset;
288
uint32_t picture_count;
289
uint32_t frame_number;
290
uint32_t picture_order_count;
291
uint32_t num_i_pic_remain_in_rcgop;
292
uint32_t num_p_pic_remain_in_rcgop;
293
uint32_t num_b_pic_remain_in_rcgop;
294
uint32_t num_ir_pic_remain_in_rcgop;
295
uint32_t enable_intra_refresh;
296
uint32_t aq_variance_en;
297
uint32_t aq_block_size;
298
uint32_t aq_mb_variance_sel;
299
uint32_t aq_frame_variance_sel;
305
uint32_t context_in_sfb;
308
struct rvce_enc_create {
309
uint32_t enc_use_circular_buffer;
310
uint32_t enc_profile;
312
uint32_t enc_pic_struct_restriction;
313
uint32_t enc_image_width;
314
uint32_t enc_image_height;
315
uint32_t enc_ref_pic_luma_pitch;
316
uint32_t enc_ref_pic_chroma_pitch;
317
uint32_t enc_ref_y_height_in_qw;
318
uint32_t enc_ref_pic_addr_array_enc_pic_struct_restriction_disable_rdo;
319
uint32_t enc_pre_encode_context_buffer_offset;
320
uint32_t enc_pre_encode_input_luma_buffer_offset;
321
uint32_t enc_pre_encode_input_chroma_buffer_offset;
322
uint32_t enc_pre_encode_mode_chromaflag_vbaqmode_scenechangesensitivity;
325
struct rvce_config_ext {
326
uint32_t enc_enable_perf_logging;
329
struct rvce_h264_enc_pic {
330
struct rvce_rate_control rc;
331
struct rvce_motion_estimation me;
332
struct rvce_pic_control pc;
333
struct rvce_task_info ti;
334
struct rvce_feedback_buf_pkg fb;
337
struct rvce_enc_operation eo;
338
struct rvce_enc_create ec;
339
struct rvce_config_ext ce;
341
unsigned quant_i_frames;
342
unsigned quant_p_frames;
343
unsigned quant_b_frames;
345
enum pipe_h2645_enc_picture_type picture_type;
347
unsigned frame_num_cnt;
353
unsigned pic_order_cnt;
356
unsigned addrmode_arraymode_disrdo_distwoinstants;
360
bool has_ref_pic_list;
362
unsigned int ref_pic_list_0[32];
363
unsigned int ref_pic_list_1[32];
364
unsigned int frame_idx[32];
367
/* VCE encoder representation */
368
struct rvce_encoder {
369
struct pipe_video_codec base;
371
/* version specific packets */
372
void (*session)(struct rvce_encoder *enc);
373
void (*create)(struct rvce_encoder *enc);
374
void (*feedback)(struct rvce_encoder *enc);
375
void (*rate_control)(struct rvce_encoder *enc);
376
void (*config_extension)(struct rvce_encoder *enc);
377
void (*pic_control)(struct rvce_encoder *enc);
378
void (*motion_estimation)(struct rvce_encoder *enc);
379
void (*rdo)(struct rvce_encoder *enc);
380
void (*vui)(struct rvce_encoder *enc);
381
void (*config)(struct rvce_encoder *enc);
382
void (*encode)(struct rvce_encoder *enc);
383
void (*destroy)(struct rvce_encoder *enc);
384
void (*task_info)(struct rvce_encoder *enc, uint32_t op, uint32_t dep, uint32_t fb_idx,
386
void (*si_get_pic_param)(struct rvce_encoder *enc, struct pipe_h264_enc_picture_desc *pic);
388
unsigned stream_handle;
390
struct pipe_screen *screen;
391
struct radeon_winsys *ws;
392
struct radeon_cmdbuf cs;
394
rvce_get_buffer get_buffer;
396
struct pb_buffer *handle;
397
struct radeon_surf *luma;
398
struct radeon_surf *chroma;
400
struct pb_buffer *bs_handle;
403
struct rvce_cpb_slot *cpb_array;
404
struct list_head cpb_slots;
407
struct rvid_buffer *fb;
408
struct rvid_buffer cpb;
409
struct pipe_h264_enc_picture_desc pic;
410
struct rvce_h264_enc_pic enc_pic;
412
unsigned task_info_idx;
421
/* CPB handling functions */
422
struct rvce_cpb_slot *si_current_slot(struct rvce_encoder *enc);
423
struct rvce_cpb_slot *si_l0_slot(struct rvce_encoder *enc);
424
struct rvce_cpb_slot *si_l1_slot(struct rvce_encoder *enc);
425
void si_vce_frame_offset(struct rvce_encoder *enc, struct rvce_cpb_slot *slot, signed *luma_offset,
426
signed *chroma_offset);
428
struct pipe_video_codec *si_vce_create_encoder(struct pipe_context *context,
429
const struct pipe_video_codec *templat,
430
struct radeon_winsys *ws,
431
rvce_get_buffer get_buffer);
433
bool si_vce_is_fw_version_supported(struct si_screen *sscreen);
435
void si_vce_add_buffer(struct rvce_encoder *enc, struct pb_buffer *buf, unsigned usage,
436
enum radeon_bo_domain domain, signed offset);
438
/* init vce fw 40.2.2 specific callbacks */
439
void si_vce_40_2_2_init(struct rvce_encoder *enc);
441
/* init vce fw 50 specific callbacks */
442
void si_vce_50_init(struct rvce_encoder *enc);
444
/* init vce fw 52 specific callbacks */
445
void si_vce_52_init(struct rvce_encoder *enc);
447
/* get parameters for vce 40.2.2 */
448
void si_vce_40_2_2_get_param(struct rvce_encoder *enc, struct pipe_h264_enc_picture_desc *pic);
450
/* get parameters for vce 50 */
451
void si_vce_50_get_param(struct rvce_encoder *enc, struct pipe_h264_enc_picture_desc *pic);
453
/* get parameters for vce 52 */
454
void si_vce_52_get_param(struct rvce_encoder *enc, struct pipe_h264_enc_picture_desc *pic);