1 /**************************************************************************
2 *
3 * Copyright 2017 Advanced Micro Devices, Inc.
4 * All Rights Reserved.
5 *
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:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
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.
25 *
26 **************************************************************************/
27
28 #include <assert.h>
29 #include <stdio.h>
30
31 #include "pipe/p_video_codec.h"
32
33 #include "util/u_memory.h"
34 #include "util/u_video.h"
35
36 #include "vl/vl_mpeg12_decoder.h"
37
38 #include "radeonsi/si_pipe.h"
39 #include "radeon_video.h"
40 #include "radeon_vcn_dec.h"
41
42 #define FB_BUFFER_OFFSET 0x1000
43 #define FB_BUFFER_SIZE 2048
44 #define IT_SCALING_TABLE_SIZE 992
45 #define RDECODE_SESSION_CONTEXT_SIZE (128 * 1024)
46
47 #define RDECODE_GPCOM_VCPU_CMD 0x2070c
48 #define RDECODE_GPCOM_VCPU_DATA0 0x20710
49 #define RDECODE_GPCOM_VCPU_DATA1 0x20714
50 #define RDECODE_ENGINE_CNTL 0x20718
51
52 #define NUM_BUFFERS 4
53 #define NUM_MPEG2_REFS 6
54 #define NUM_H264_REFS 17
55 #define NUM_VC1_REFS 5
56
57 struct radeon_decoder {
58 struct pipe_video_codec base;
59
60 unsigned stream_handle;
61 unsigned stream_type;
62 unsigned frame_number;
63
64 struct pipe_screen *screen;
65 struct radeon_winsys *ws;
66 struct radeon_winsys_cs *cs;
67
68 void *msg;
69 uint32_t *fb;
70 uint8_t *it;
71 void *bs_ptr;
72
73 struct rvid_buffer msg_fb_it_buffers[NUM_BUFFERS];
74 struct rvid_buffer bs_buffers[NUM_BUFFERS];
75 struct rvid_buffer dpb;
76 struct rvid_buffer ctx;
77 struct rvid_buffer sessionctx;
78
79 unsigned bs_size;
80 unsigned cur_buffer;
81 void *render_pic_list[16];
82 };
83
get_h264_msg(struct radeon_decoder * dec,struct pipe_h264_picture_desc * pic)84 static rvcn_dec_message_avc_t get_h264_msg(struct radeon_decoder *dec,
85 struct pipe_h264_picture_desc *pic)
86 {
87 rvcn_dec_message_avc_t result;
88
89 memset(&result, 0, sizeof(result));
90 switch (pic->base.profile) {
91 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE:
92 result.profile = RDECODE_H264_PROFILE_BASELINE;
93 break;
94
95 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN:
96 result.profile = RDECODE_H264_PROFILE_MAIN;
97 break;
98
99 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH:
100 result.profile = RDECODE_H264_PROFILE_HIGH;
101 break;
102
103 default:
104 assert(0);
105 break;
106 }
107
108 result.level = dec->base.level;
109
110 result.sps_info_flags = 0;
111 result.sps_info_flags |= pic->pps->sps->direct_8x8_inference_flag << 0;
112 result.sps_info_flags |= pic->pps->sps->mb_adaptive_frame_field_flag << 1;
113 result.sps_info_flags |= pic->pps->sps->frame_mbs_only_flag << 2;
114 result.sps_info_flags |= pic->pps->sps->delta_pic_order_always_zero_flag << 3;
115 result.sps_info_flags |= 1 << RDECODE_SPS_INFO_H264_EXTENSION_SUPPORT_FLAG_SHIFT;
116
117 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8;
118 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8;
119 result.log2_max_frame_num_minus4 = pic->pps->sps->log2_max_frame_num_minus4;
120 result.pic_order_cnt_type = pic->pps->sps->pic_order_cnt_type;
121 result.log2_max_pic_order_cnt_lsb_minus4 =
122 pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4;
123
124 switch (dec->base.chroma_format) {
125 case PIPE_VIDEO_CHROMA_FORMAT_NONE:
126 break;
127 case PIPE_VIDEO_CHROMA_FORMAT_400:
128 result.chroma_format = 0;
129 break;
130 case PIPE_VIDEO_CHROMA_FORMAT_420:
131 result.chroma_format = 1;
132 break;
133 case PIPE_VIDEO_CHROMA_FORMAT_422:
134 result.chroma_format = 2;
135 break;
136 case PIPE_VIDEO_CHROMA_FORMAT_444:
137 result.chroma_format = 3;
138 break;
139 }
140
141 result.pps_info_flags = 0;
142 result.pps_info_flags |= pic->pps->transform_8x8_mode_flag << 0;
143 result.pps_info_flags |= pic->pps->redundant_pic_cnt_present_flag << 1;
144 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 2;
145 result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag << 3;
146 result.pps_info_flags |= pic->pps->weighted_bipred_idc << 4;
147 result.pps_info_flags |= pic->pps->weighted_pred_flag << 6;
148 result.pps_info_flags |= pic->pps->bottom_field_pic_order_in_frame_present_flag << 7;
149 result.pps_info_flags |= pic->pps->entropy_coding_mode_flag << 8;
150
151 result.num_slice_groups_minus1 = pic->pps->num_slice_groups_minus1;
152 result.slice_group_map_type = pic->pps->slice_group_map_type;
153 result.slice_group_change_rate_minus1 = pic->pps->slice_group_change_rate_minus1;
154 result.pic_init_qp_minus26 = pic->pps->pic_init_qp_minus26;
155 result.chroma_qp_index_offset = pic->pps->chroma_qp_index_offset;
156 result.second_chroma_qp_index_offset = pic->pps->second_chroma_qp_index_offset;
157
158 memcpy(result.scaling_list_4x4, pic->pps->ScalingList4x4, 6*16);
159 memcpy(result.scaling_list_8x8, pic->pps->ScalingList8x8, 2*64);
160
161 memcpy(dec->it, result.scaling_list_4x4, 6*16);
162 memcpy((dec->it + 96), result.scaling_list_8x8, 2*64);
163
164 result.num_ref_frames = pic->num_ref_frames;
165
166 result.num_ref_idx_l0_active_minus1 = pic->num_ref_idx_l0_active_minus1;
167 result.num_ref_idx_l1_active_minus1 = pic->num_ref_idx_l1_active_minus1;
168
169 result.frame_num = pic->frame_num;
170 memcpy(result.frame_num_list, pic->frame_num_list, 4*16);
171 result.curr_field_order_cnt_list[0] = pic->field_order_cnt[0];
172 result.curr_field_order_cnt_list[1] = pic->field_order_cnt[1];
173 memcpy(result.field_order_cnt_list, pic->field_order_cnt_list, 4*16*2);
174
175 result.decoded_pic_idx = pic->frame_num;
176
177 return result;
178 }
179
radeon_dec_destroy_associated_data(void * data)180 static void radeon_dec_destroy_associated_data(void *data)
181 {
182 /* NOOP, since we only use an intptr */
183 }
184
get_h265_msg(struct radeon_decoder * dec,struct pipe_video_buffer * target,struct pipe_h265_picture_desc * pic)185 static rvcn_dec_message_hevc_t get_h265_msg(struct radeon_decoder *dec,
186 struct pipe_video_buffer *target,
187 struct pipe_h265_picture_desc *pic)
188 {
189 rvcn_dec_message_hevc_t result;
190 unsigned i, j;
191
192 memset(&result, 0, sizeof(result));
193 result.sps_info_flags = 0;
194 result.sps_info_flags |= pic->pps->sps->scaling_list_enabled_flag << 0;
195 result.sps_info_flags |= pic->pps->sps->amp_enabled_flag << 1;
196 result.sps_info_flags |= pic->pps->sps->sample_adaptive_offset_enabled_flag << 2;
197 result.sps_info_flags |= pic->pps->sps->pcm_enabled_flag << 3;
198 result.sps_info_flags |= pic->pps->sps->pcm_loop_filter_disabled_flag << 4;
199 result.sps_info_flags |= pic->pps->sps->long_term_ref_pics_present_flag << 5;
200 result.sps_info_flags |= pic->pps->sps->sps_temporal_mvp_enabled_flag << 6;
201 result.sps_info_flags |= pic->pps->sps->strong_intra_smoothing_enabled_flag << 7;
202 result.sps_info_flags |= pic->pps->sps->separate_colour_plane_flag << 8;
203 if (((struct si_screen*)dec->screen)->info.family == CHIP_CARRIZO)
204 result.sps_info_flags |= 1 << 9;
205 if (pic->UseRefPicList == true)
206 result.sps_info_flags |= 1 << 10;
207
208 result.chroma_format = pic->pps->sps->chroma_format_idc;
209 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8;
210 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8;
211 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4;
212 result.sps_max_dec_pic_buffering_minus1 = pic->pps->sps->sps_max_dec_pic_buffering_minus1;
213 result.log2_min_luma_coding_block_size_minus3 =
214 pic->pps->sps->log2_min_luma_coding_block_size_minus3;
215 result.log2_diff_max_min_luma_coding_block_size =
216 pic->pps->sps->log2_diff_max_min_luma_coding_block_size;
217 result.log2_min_transform_block_size_minus2 =
218 pic->pps->sps->log2_min_transform_block_size_minus2;
219 result.log2_diff_max_min_transform_block_size =
220 pic->pps->sps->log2_diff_max_min_transform_block_size;
221 result.max_transform_hierarchy_depth_inter =
222 pic->pps->sps->max_transform_hierarchy_depth_inter;
223 result.max_transform_hierarchy_depth_intra =
224 pic->pps->sps->max_transform_hierarchy_depth_intra;
225 result.pcm_sample_bit_depth_luma_minus1 = pic->pps->sps->pcm_sample_bit_depth_luma_minus1;
226 result.pcm_sample_bit_depth_chroma_minus1 =
227 pic->pps->sps->pcm_sample_bit_depth_chroma_minus1;
228 result.log2_min_pcm_luma_coding_block_size_minus3 =
229 pic->pps->sps->log2_min_pcm_luma_coding_block_size_minus3;
230 result.log2_diff_max_min_pcm_luma_coding_block_size =
231 pic->pps->sps->log2_diff_max_min_pcm_luma_coding_block_size;
232 result.num_short_term_ref_pic_sets = pic->pps->sps->num_short_term_ref_pic_sets;
233
234 result.pps_info_flags = 0;
235 result.pps_info_flags |= pic->pps->dependent_slice_segments_enabled_flag << 0;
236 result.pps_info_flags |= pic->pps->output_flag_present_flag << 1;
237 result.pps_info_flags |= pic->pps->sign_data_hiding_enabled_flag << 2;
238 result.pps_info_flags |= pic->pps->cabac_init_present_flag << 3;
239 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 4;
240 result.pps_info_flags |= pic->pps->transform_skip_enabled_flag << 5;
241 result.pps_info_flags |= pic->pps->cu_qp_delta_enabled_flag << 6;
242 result.pps_info_flags |= pic->pps->pps_slice_chroma_qp_offsets_present_flag << 7;
243 result.pps_info_flags |= pic->pps->weighted_pred_flag << 8;
244 result.pps_info_flags |= pic->pps->weighted_bipred_flag << 9;
245 result.pps_info_flags |= pic->pps->transquant_bypass_enabled_flag << 10;
246 result.pps_info_flags |= pic->pps->tiles_enabled_flag << 11;
247 result.pps_info_flags |= pic->pps->entropy_coding_sync_enabled_flag << 12;
248 result.pps_info_flags |= pic->pps->uniform_spacing_flag << 13;
249 result.pps_info_flags |= pic->pps->loop_filter_across_tiles_enabled_flag << 14;
250 result.pps_info_flags |= pic->pps->pps_loop_filter_across_slices_enabled_flag << 15;
251 result.pps_info_flags |= pic->pps->deblocking_filter_override_enabled_flag << 16;
252 result.pps_info_flags |= pic->pps->pps_deblocking_filter_disabled_flag << 17;
253 result.pps_info_flags |= pic->pps->lists_modification_present_flag << 18;
254 result.pps_info_flags |= pic->pps->slice_segment_header_extension_present_flag << 19;
255
256 result.num_extra_slice_header_bits = pic->pps->num_extra_slice_header_bits;
257 result.num_long_term_ref_pic_sps = pic->pps->sps->num_long_term_ref_pics_sps;
258 result.num_ref_idx_l0_default_active_minus1 = pic->pps->num_ref_idx_l0_default_active_minus1;
259 result.num_ref_idx_l1_default_active_minus1 = pic->pps->num_ref_idx_l1_default_active_minus1;
260 result.pps_cb_qp_offset = pic->pps->pps_cb_qp_offset;
261 result.pps_cr_qp_offset = pic->pps->pps_cr_qp_offset;
262 result.pps_beta_offset_div2 = pic->pps->pps_beta_offset_div2;
263 result.pps_tc_offset_div2 = pic->pps->pps_tc_offset_div2;
264 result.diff_cu_qp_delta_depth = pic->pps->diff_cu_qp_delta_depth;
265 result.num_tile_columns_minus1 = pic->pps->num_tile_columns_minus1;
266 result.num_tile_rows_minus1 = pic->pps->num_tile_rows_minus1;
267 result.log2_parallel_merge_level_minus2 = pic->pps->log2_parallel_merge_level_minus2;
268 result.init_qp_minus26 = pic->pps->init_qp_minus26;
269
270 for (i = 0; i < 19; ++i)
271 result.column_width_minus1[i] = pic->pps->column_width_minus1[i];
272
273 for (i = 0; i < 21; ++i)
274 result.row_height_minus1[i] = pic->pps->row_height_minus1[i];
275
276 result.num_delta_pocs_ref_rps_idx = pic->NumDeltaPocsOfRefRpsIdx;
277 result.curr_poc = pic->CurrPicOrderCntVal;
278
279 for (i = 0 ; i < 16 ; i++) {
280 for (j = 0; (pic->ref[j] != NULL) && (j < 16) ; j++) {
281 if (dec->render_pic_list[i] == pic->ref[j])
282 break;
283 if (j == 15)
284 dec->render_pic_list[i] = NULL;
285 else if (pic->ref[j+1] == NULL)
286 dec->render_pic_list[i] = NULL;
287 }
288 }
289 for (i = 0 ; i < 16 ; i++) {
290 if (dec->render_pic_list[i] == NULL) {
291 dec->render_pic_list[i] = target;
292 result.curr_idx = i;
293 break;
294 }
295 }
296
297 vl_video_buffer_set_associated_data(target, &dec->base,
298 (void *)(uintptr_t)result.curr_idx,
299 &radeon_dec_destroy_associated_data);
300
301 for (i = 0; i < 16; ++i) {
302 struct pipe_video_buffer *ref = pic->ref[i];
303 uintptr_t ref_pic = 0;
304
305 result.poc_list[i] = pic->PicOrderCntVal[i];
306
307 if (ref)
308 ref_pic = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base);
309 else
310 ref_pic = 0x7F;
311 result.ref_pic_list[i] = ref_pic;
312 }
313
314 for (i = 0; i < 8; ++i) {
315 result.ref_pic_set_st_curr_before[i] = 0xFF;
316 result.ref_pic_set_st_curr_after[i] = 0xFF;
317 result.ref_pic_set_lt_curr[i] = 0xFF;
318 }
319
320 for (i = 0; i < pic->NumPocStCurrBefore; ++i)
321 result.ref_pic_set_st_curr_before[i] = pic->RefPicSetStCurrBefore[i];
322
323 for (i = 0; i < pic->NumPocStCurrAfter; ++i)
324 result.ref_pic_set_st_curr_after[i] = pic->RefPicSetStCurrAfter[i];
325
326 for (i = 0; i < pic->NumPocLtCurr; ++i)
327 result.ref_pic_set_lt_curr[i] = pic->RefPicSetLtCurr[i];
328
329 for (i = 0; i < 6; ++i)
330 result.ucScalingListDCCoefSizeID2[i] = pic->pps->sps->ScalingListDCCoeff16x16[i];
331
332 for (i = 0; i < 2; ++i)
333 result.ucScalingListDCCoefSizeID3[i] = pic->pps->sps->ScalingListDCCoeff32x32[i];
334
335 memcpy(dec->it, pic->pps->sps->ScalingList4x4, 6 * 16);
336 memcpy(dec->it + 96, pic->pps->sps->ScalingList8x8, 6 * 64);
337 memcpy(dec->it + 480, pic->pps->sps->ScalingList16x16, 6 * 64);
338 memcpy(dec->it + 864, pic->pps->sps->ScalingList32x32, 2 * 64);
339
340 for (i = 0 ; i < 2 ; i++) {
341 for (j = 0 ; j < 15 ; j++)
342 result.direct_reflist[i][j] = pic->RefPicList[i][j];
343 }
344
345 if (pic->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10) {
346 if (target->buffer_format == PIPE_FORMAT_P016) {
347 result.p010_mode = 1;
348 result.msb_mode = 1;
349 } else {
350 result.p010_mode = 0;
351 result.luma_10to8 = 5;
352 result.chroma_10to8 = 5;
353 result.hevc_reserved[0] = 4; /* sclr_luma10to8 */
354 result.hevc_reserved[1] = 4; /* sclr_chroma10to8 */
355 }
356 }
357
358 return result;
359 }
360
calc_ctx_size_h265_main(struct radeon_decoder * dec)361 static unsigned calc_ctx_size_h265_main(struct radeon_decoder *dec)
362 {
363 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
364 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
365
366 unsigned max_references = dec->base.max_references + 1;
367
368 if (dec->base.width * dec->base.height >= 4096*2000)
369 max_references = MAX2(max_references, 8);
370 else
371 max_references = MAX2(max_references, 17);
372
373 width = align (width, 16);
374 height = align (height, 16);
375 return ((width + 255) / 16) * ((height + 255) / 16) * 16 * max_references + 52 * 1024;
376 }
377
calc_ctx_size_h265_main10(struct radeon_decoder * dec,struct pipe_h265_picture_desc * pic)378 static unsigned calc_ctx_size_h265_main10(struct radeon_decoder *dec, struct pipe_h265_picture_desc *pic)
379 {
380 unsigned block_size, log2_ctb_size, width_in_ctb, height_in_ctb, num_16x16_block_per_ctb;
381 unsigned context_buffer_size_per_ctb_row, cm_buffer_size, max_mb_address, db_left_tile_pxl_size;
382 unsigned db_left_tile_ctx_size = 4096 / 16 * (32 + 16 * 4);
383
384 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
385 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
386 unsigned coeff_10bit = (pic->pps->sps->bit_depth_luma_minus8 ||
387 pic->pps->sps->bit_depth_chroma_minus8) ? 2 : 1;
388
389 unsigned max_references = dec->base.max_references + 1;
390
391 if (dec->base.width * dec->base.height >= 4096*2000)
392 max_references = MAX2(max_references, 8);
393 else
394 max_references = MAX2(max_references, 17);
395
396 block_size = (1 << (pic->pps->sps->log2_min_luma_coding_block_size_minus3 + 3));
397 log2_ctb_size = block_size + pic->pps->sps->log2_diff_max_min_luma_coding_block_size;
398
399 width_in_ctb = (width + ((1 << log2_ctb_size) - 1)) >> log2_ctb_size;
400 height_in_ctb = (height + ((1 << log2_ctb_size) - 1)) >> log2_ctb_size;
401
402 num_16x16_block_per_ctb = ((1 << log2_ctb_size) >> 4) * ((1 << log2_ctb_size) >> 4);
403 context_buffer_size_per_ctb_row = align(width_in_ctb * num_16x16_block_per_ctb * 16, 256);
404 max_mb_address = (unsigned) ceil(height * 8 / 2048.0);
405
406 cm_buffer_size = max_references * context_buffer_size_per_ctb_row * height_in_ctb;
407 db_left_tile_pxl_size = coeff_10bit * (max_mb_address * 2 * 2048 + 1024);
408
409 return cm_buffer_size + db_left_tile_ctx_size + db_left_tile_pxl_size;
410 }
411
get_vc1_msg(struct pipe_vc1_picture_desc * pic)412 static rvcn_dec_message_vc1_t get_vc1_msg(struct pipe_vc1_picture_desc *pic)
413 {
414 rvcn_dec_message_vc1_t result;
415
416 memset(&result, 0, sizeof(result));
417 switch(pic->base.profile) {
418 case PIPE_VIDEO_PROFILE_VC1_SIMPLE:
419 result.profile = RDECODE_VC1_PROFILE_SIMPLE;
420 result.level = 1;
421 break;
422
423 case PIPE_VIDEO_PROFILE_VC1_MAIN:
424 result.profile = RDECODE_VC1_PROFILE_MAIN;
425 result.level = 2;
426 break;
427
428 case PIPE_VIDEO_PROFILE_VC1_ADVANCED:
429 result.profile = RDECODE_VC1_PROFILE_ADVANCED;
430 result.level = 4;
431 break;
432
433 default:
434 assert(0);
435 }
436
437 result.sps_info_flags |= pic->postprocflag << 7;
438 result.sps_info_flags |= pic->pulldown << 6;
439 result.sps_info_flags |= pic->interlace << 5;
440 result.sps_info_flags |= pic->tfcntrflag << 4;
441 result.sps_info_flags |= pic->finterpflag << 3;
442 result.sps_info_flags |= pic->psf << 1;
443
444 result.pps_info_flags |= pic->range_mapy_flag << 31;
445 result.pps_info_flags |= pic->range_mapy << 28;
446 result.pps_info_flags |= pic->range_mapuv_flag << 27;
447 result.pps_info_flags |= pic->range_mapuv << 24;
448 result.pps_info_flags |= pic->multires << 21;
449 result.pps_info_flags |= pic->maxbframes << 16;
450 result.pps_info_flags |= pic->overlap << 11;
451 result.pps_info_flags |= pic->quantizer << 9;
452 result.pps_info_flags |= pic->panscan_flag << 7;
453 result.pps_info_flags |= pic->refdist_flag << 6;
454 result.pps_info_flags |= pic->vstransform << 0;
455
456 if (pic->base.profile != PIPE_VIDEO_PROFILE_VC1_SIMPLE) {
457 result.pps_info_flags |= pic->syncmarker << 20;
458 result.pps_info_flags |= pic->rangered << 19;
459 result.pps_info_flags |= pic->loopfilter << 5;
460 result.pps_info_flags |= pic->fastuvmc << 4;
461 result.pps_info_flags |= pic->extended_mv << 3;
462 result.pps_info_flags |= pic->extended_dmv << 8;
463 result.pps_info_flags |= pic->dquant << 1;
464 }
465
466 result.chroma_format = 1;
467
468 return result;
469 }
470
get_ref_pic_idx(struct radeon_decoder * dec,struct pipe_video_buffer * ref)471 static uint32_t get_ref_pic_idx(struct radeon_decoder *dec, struct pipe_video_buffer *ref)
472 {
473 uint32_t min = MAX2(dec->frame_number, NUM_MPEG2_REFS) - NUM_MPEG2_REFS;
474 uint32_t max = MAX2(dec->frame_number, 1) - 1;
475 uintptr_t frame;
476
477 /* seems to be the most sane fallback */
478 if (!ref)
479 return max;
480
481 /* get the frame number from the associated data */
482 frame = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base);
483
484 /* limit the frame number to a valid range */
485 return MAX2(MIN2(frame, max), min);
486 }
487
get_mpeg2_msg(struct radeon_decoder * dec,struct pipe_mpeg12_picture_desc * pic)488 static rvcn_dec_message_mpeg2_vld_t get_mpeg2_msg(struct radeon_decoder *dec,
489 struct pipe_mpeg12_picture_desc *pic)
490 {
491 const int *zscan = pic->alternate_scan ? vl_zscan_alternate : vl_zscan_normal;
492 rvcn_dec_message_mpeg2_vld_t result;
493 unsigned i;
494
495 memset(&result, 0, sizeof(result));
496 result.decoded_pic_idx = dec->frame_number;
497
498 result.forward_ref_pic_idx = get_ref_pic_idx(dec, pic->ref[0]);
499 result.backward_ref_pic_idx = get_ref_pic_idx(dec, pic->ref[1]);
500
501 if(pic->intra_matrix) {
502 result.load_intra_quantiser_matrix = 1;
503 for (i = 0; i < 64; ++i) {
504 result.intra_quantiser_matrix[i] = pic->intra_matrix[zscan[i]];
505 }
506 }
507 if(pic->non_intra_matrix) {
508 result.load_nonintra_quantiser_matrix = 1;
509 for (i = 0; i < 64; ++i) {
510 result.nonintra_quantiser_matrix[i] = pic->non_intra_matrix[zscan[i]];
511 }
512 }
513
514 result.profile_and_level_indication = 0;
515 result.chroma_format = 0x1;
516
517 result.picture_coding_type = pic->picture_coding_type;
518 result.f_code[0][0] = pic->f_code[0][0] + 1;
519 result.f_code[0][1] = pic->f_code[0][1] + 1;
520 result.f_code[1][0] = pic->f_code[1][0] + 1;
521 result.f_code[1][1] = pic->f_code[1][1] + 1;
522 result.intra_dc_precision = pic->intra_dc_precision;
523 result.pic_structure = pic->picture_structure;
524 result.top_field_first = pic->top_field_first;
525 result.frame_pred_frame_dct = pic->frame_pred_frame_dct;
526 result.concealment_motion_vectors = pic->concealment_motion_vectors;
527 result.q_scale_type = pic->q_scale_type;
528 result.intra_vlc_format = pic->intra_vlc_format;
529 result.alternate_scan = pic->alternate_scan;
530
531 return result;
532 }
533
get_mpeg4_msg(struct radeon_decoder * dec,struct pipe_mpeg4_picture_desc * pic)534 static rvcn_dec_message_mpeg4_asp_vld_t get_mpeg4_msg(struct radeon_decoder *dec,
535 struct pipe_mpeg4_picture_desc *pic)
536 {
537 rvcn_dec_message_mpeg4_asp_vld_t result;
538 unsigned i;
539
540 memset(&result, 0, sizeof(result));
541 result.decoded_pic_idx = dec->frame_number;
542
543 result.forward_ref_pic_idx = get_ref_pic_idx(dec, pic->ref[0]);
544 result.backward_ref_pic_idx = get_ref_pic_idx(dec, pic->ref[1]);
545
546 result.variant_type = 0;
547 result.profile_and_level_indication = 0xF0;
548
549 result.video_object_layer_verid = 0x5;
550 result.video_object_layer_shape = 0x0;
551
552 result.video_object_layer_width = dec->base.width;
553 result.video_object_layer_height = dec->base.height;
554
555 result.vop_time_increment_resolution = pic->vop_time_increment_resolution;
556
557 result.short_video_header = pic->short_video_header;
558 result.interlaced = pic->interlaced;
559 result.load_intra_quant_mat = 1;
560 result.load_nonintra_quant_mat = 1;
561 result.quarter_sample = pic->quarter_sample;
562 result.complexity_estimation_disable = 1;
563 result.resync_marker_disable = pic->resync_marker_disable;
564 result.newpred_enable = 0;
565 result.reduced_resolution_vop_enable = 0;
566
567 result.quant_type = pic->quant_type;
568
569 for (i = 0; i < 64; ++i) {
570 result.intra_quant_mat[i] = pic->intra_matrix[vl_zscan_normal[i]];
571 result.nonintra_quant_mat[i] = pic->non_intra_matrix[vl_zscan_normal[i]];
572 }
573
574 return result;
575 }
576
rvcn_dec_message_create(struct radeon_decoder * dec)577 static void rvcn_dec_message_create(struct radeon_decoder *dec)
578 {
579 rvcn_dec_message_header_t *header = dec->msg;
580 rvcn_dec_message_create_t *create = dec->msg + sizeof(rvcn_dec_message_header_t);
581 unsigned sizes = sizeof(rvcn_dec_message_header_t) + sizeof(rvcn_dec_message_create_t);
582
583 memset(dec->msg, 0, sizes);
584 header->header_size = sizeof(rvcn_dec_message_header_t);
585 header->total_size = sizes;
586 header->num_buffers = 1;
587 header->msg_type = RDECODE_MSG_CREATE;
588 header->stream_handle = dec->stream_handle;
589 header->status_report_feedback_number = 0;
590
591 header->index[0].message_id = RDECODE_MESSAGE_CREATE;
592 header->index[0].offset = sizeof(rvcn_dec_message_header_t);
593 header->index[0].size = sizeof(rvcn_dec_message_create_t);
594 header->index[0].filled = 0;
595
596 create->stream_type = dec->stream_type;
597 create->session_flags = 0;
598 create->width_in_samples = dec->base.width;
599 create->height_in_samples = dec->base.height;
600 }
601
rvcn_dec_message_decode(struct radeon_decoder * dec,struct pipe_video_buffer * target,struct pipe_picture_desc * picture)602 static struct pb_buffer *rvcn_dec_message_decode(struct radeon_decoder *dec,
603 struct pipe_video_buffer *target,
604 struct pipe_picture_desc *picture)
605 {
606 struct r600_texture *luma = (struct r600_texture *)
607 ((struct vl_video_buffer *)target)->resources[0];
608 struct r600_texture *chroma = (struct r600_texture *)
609 ((struct vl_video_buffer *)target)->resources[1];
610 rvcn_dec_message_header_t *header;
611 rvcn_dec_message_index_t *index;
612 rvcn_dec_message_decode_t *decode;
613 unsigned sizes = 0, offset_decode, offset_codec;
614 void *codec;
615
616 header = dec->msg;
617 sizes += sizeof(rvcn_dec_message_header_t);
618 index = (void*)header + sizeof(rvcn_dec_message_header_t);
619 sizes += sizeof(rvcn_dec_message_index_t);
620 offset_decode = sizes;
621 decode = (void*)index + sizeof(rvcn_dec_message_index_t);
622 sizes += sizeof(rvcn_dec_message_decode_t);
623 offset_codec = sizes;
624 codec = (void*)decode + sizeof(rvcn_dec_message_decode_t);
625
626 memset(dec->msg, 0, sizes);
627 header->header_size = sizeof(rvcn_dec_message_header_t);
628 header->total_size = sizes;
629 header->num_buffers = 2;
630 header->msg_type = RDECODE_MSG_DECODE;
631 header->stream_handle = dec->stream_handle;
632 header->status_report_feedback_number = dec->frame_number;
633
634 header->index[0].message_id = RDECODE_MESSAGE_DECODE;
635 header->index[0].offset = offset_decode;
636 header->index[0].size = sizeof(rvcn_dec_message_decode_t);
637 header->index[0].filled = 0;
638
639 index->offset = offset_codec;
640 index->size = sizeof(rvcn_dec_message_avc_t);
641 index->filled = 0;
642
643 decode->stream_type = dec->stream_type;
644 decode->decode_flags = 0x1;
645 decode->width_in_samples = dec->base.width;
646 decode->height_in_samples = dec->base.height;
647
648 decode->bsd_size = align(dec->bs_size, 128);
649 decode->dpb_size = dec->dpb.res->buf->size;
650 decode->dt_size =
651 ((struct r600_resource *)((struct vl_video_buffer *)target)->resources[0])->buf->size +
652 ((struct r600_resource *)((struct vl_video_buffer *)target)->resources[1])->buf->size;
653
654 decode->sct_size = 0;
655 decode->sc_coeff_size = 0;
656
657 decode->sw_ctxt_size = RDECODE_SESSION_CONTEXT_SIZE;
658 decode->db_pitch = align(dec->base.width, 32);
659 decode->db_surf_tile_config = 0;
660
661 decode->dt_pitch = luma->surface.u.gfx9.surf_pitch * luma->surface.blk_w;
662 decode->dt_uv_pitch = decode->dt_pitch / 2;
663
664 decode->dt_tiling_mode = 0;
665 decode->dt_swizzle_mode = RDECODE_SW_MODE_LINEAR;
666 decode->dt_array_mode = RDECODE_ARRAY_MODE_LINEAR;
667 decode->dt_field_mode = ((struct vl_video_buffer *)target)->base.interlaced;
668 decode->dt_surf_tile_config = 0;
669 decode->dt_uv_surf_tile_config = 0;
670
671 decode->dt_luma_top_offset = luma->surface.u.gfx9.surf_offset;
672 decode->dt_chroma_top_offset = chroma->surface.u.gfx9.surf_offset;
673 if (decode->dt_field_mode) {
674 decode->dt_luma_bottom_offset = luma->surface.u.gfx9.surf_offset +
675 luma->surface.u.gfx9.surf_slice_size;
676 decode->dt_chroma_bottom_offset = chroma->surface.u.gfx9.surf_offset +
677 chroma->surface.u.gfx9.surf_slice_size;
678 } else {
679 decode->dt_luma_bottom_offset = decode->dt_luma_top_offset;
680 decode->dt_chroma_bottom_offset = decode->dt_chroma_top_offset;
681 }
682
683 switch (u_reduce_video_profile(picture->profile)) {
684 case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
685 rvcn_dec_message_avc_t avc =
686 get_h264_msg(dec, (struct pipe_h264_picture_desc*)picture);
687 memcpy(codec, (void*)&avc, sizeof(rvcn_dec_message_avc_t));
688 index->message_id = RDECODE_MESSAGE_AVC;
689 break;
690 }
691 case PIPE_VIDEO_FORMAT_HEVC: {
692 rvcn_dec_message_hevc_t hevc =
693 get_h265_msg(dec, target, (struct pipe_h265_picture_desc*)picture);
694
695 memcpy(codec, (void*)&hevc, sizeof(rvcn_dec_message_hevc_t));
696 index->message_id = RDECODE_MESSAGE_HEVC;
697 if (dec->ctx.res == NULL) {
698 unsigned ctx_size;
699 if (dec->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10)
700 ctx_size = calc_ctx_size_h265_main10(dec,
701 (struct pipe_h265_picture_desc*)picture);
702 else
703 ctx_size = calc_ctx_size_h265_main(dec);
704 if (!si_vid_create_buffer(dec->screen, &dec->ctx, ctx_size, PIPE_USAGE_DEFAULT))
705 RVID_ERR("Can't allocated context buffer.\n");
706 si_vid_clear_buffer(dec->base.context, &dec->ctx);
707 }
708 break;
709 }
710 case PIPE_VIDEO_FORMAT_VC1: {
711 rvcn_dec_message_vc1_t vc1 = get_vc1_msg((struct pipe_vc1_picture_desc*)picture);
712
713 memcpy(codec, (void*)&vc1, sizeof(rvcn_dec_message_vc1_t));
714 if ((picture->profile == PIPE_VIDEO_PROFILE_VC1_SIMPLE) ||
715 (picture->profile == PIPE_VIDEO_PROFILE_VC1_MAIN)) {
716 decode->width_in_samples = align(decode->width_in_samples, 16) / 16;
717 decode->height_in_samples = align(decode->height_in_samples, 16) / 16;
718 }
719 index->message_id = RDECODE_MESSAGE_VC1;
720 break;
721
722 }
723 case PIPE_VIDEO_FORMAT_MPEG12: {
724 rvcn_dec_message_mpeg2_vld_t mpeg2 =
725 get_mpeg2_msg(dec, (struct pipe_mpeg12_picture_desc*)picture);
726
727 memcpy(codec, (void*)&mpeg2, sizeof(rvcn_dec_message_mpeg2_vld_t));
728 index->message_id = RDECODE_MESSAGE_MPEG2_VLD;
729 break;
730 }
731 case PIPE_VIDEO_FORMAT_MPEG4: {
732 rvcn_dec_message_mpeg4_asp_vld_t mpeg4 =
733 get_mpeg4_msg(dec, (struct pipe_mpeg4_picture_desc*)picture);
734
735 memcpy(codec, (void*)&mpeg4, sizeof(rvcn_dec_message_mpeg4_asp_vld_t));
736 index->message_id = RDECODE_MESSAGE_MPEG4_ASP_VLD;
737 break;
738 }
739 default:
740 assert(0);
741 return NULL;
742 }
743
744 if (dec->ctx.res)
745 decode->hw_ctxt_size = dec->ctx.res->buf->size;
746
747 return luma->resource.buf;
748 }
749
rvcn_dec_message_destroy(struct radeon_decoder * dec)750 static void rvcn_dec_message_destroy(struct radeon_decoder *dec)
751 {
752 rvcn_dec_message_header_t *header = dec->msg;
753
754 memset(dec->msg, 0, sizeof(rvcn_dec_message_header_t));
755 header->header_size = sizeof(rvcn_dec_message_header_t);
756 header->total_size = sizeof(rvcn_dec_message_header_t) -
757 sizeof(rvcn_dec_message_index_t);
758 header->num_buffers = 0;
759 header->msg_type = RDECODE_MSG_DESTROY;
760 header->stream_handle = dec->stream_handle;
761 header->status_report_feedback_number = 0;
762 }
763
rvcn_dec_message_feedback(struct radeon_decoder * dec)764 static void rvcn_dec_message_feedback(struct radeon_decoder *dec)
765 {
766 rvcn_dec_feedback_header_t *header = (void*)dec->fb;
767
768 header->header_size = sizeof(rvcn_dec_feedback_header_t);
769 header->total_size = sizeof(rvcn_dec_feedback_header_t);
770 header->num_buffers = 0;
771 }
772
773 /* flush IB to the hardware */
flush(struct radeon_decoder * dec,unsigned flags)774 static int flush(struct radeon_decoder *dec, unsigned flags)
775 {
776 return dec->ws->cs_flush(dec->cs, flags, NULL);
777 }
778
779 /* add a new set register command to the IB */
set_reg(struct radeon_decoder * dec,unsigned reg,uint32_t val)780 static void set_reg(struct radeon_decoder *dec, unsigned reg, uint32_t val)
781 {
782 radeon_emit(dec->cs, RDECODE_PKT0(reg >> 2, 0));
783 radeon_emit(dec->cs, val);
784 }
785
786 /* send a command to the VCPU through the GPCOM registers */
send_cmd(struct radeon_decoder * dec,unsigned cmd,struct pb_buffer * buf,uint32_t off,enum radeon_bo_usage usage,enum radeon_bo_domain domain)787 static void send_cmd(struct radeon_decoder *dec, unsigned cmd,
788 struct pb_buffer* buf, uint32_t off,
789 enum radeon_bo_usage usage, enum radeon_bo_domain domain)
790 {
791 uint64_t addr;
792
793 dec->ws->cs_add_buffer(dec->cs, buf, usage | RADEON_USAGE_SYNCHRONIZED,
794 domain, RADEON_PRIO_UVD);
795 addr = dec->ws->buffer_get_virtual_address(buf);
796 addr = addr + off;
797
798 set_reg(dec, RDECODE_GPCOM_VCPU_DATA0, addr);
799 set_reg(dec, RDECODE_GPCOM_VCPU_DATA1, addr >> 32);
800 set_reg(dec, RDECODE_GPCOM_VCPU_CMD, cmd << 1);
801 }
802
803 /* do the codec needs an IT buffer ?*/
have_it(struct radeon_decoder * dec)804 static bool have_it(struct radeon_decoder *dec)
805 {
806 return dec->stream_type == RDECODE_CODEC_H264_PERF ||
807 dec->stream_type == RDECODE_CODEC_H265;
808 }
809
810 /* map the next available message/feedback/itscaling buffer */
map_msg_fb_it_buf(struct radeon_decoder * dec)811 static void map_msg_fb_it_buf(struct radeon_decoder *dec)
812 {
813 struct rvid_buffer* buf;
814 uint8_t *ptr;
815
816 /* grab the current message/feedback buffer */
817 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
818
819 /* and map it for CPU access */
820 ptr = dec->ws->buffer_map(buf->res->buf, dec->cs, PIPE_TRANSFER_WRITE);
821
822 /* calc buffer offsets */
823 dec->msg = ptr;
824
825 dec->fb = (uint32_t *)(ptr + FB_BUFFER_OFFSET);
826 if (have_it(dec))
827 dec->it = (uint8_t *)(ptr + FB_BUFFER_OFFSET + FB_BUFFER_SIZE);
828 }
829
830 /* unmap and send a message command to the VCPU */
send_msg_buf(struct radeon_decoder * dec)831 static void send_msg_buf(struct radeon_decoder *dec)
832 {
833 struct rvid_buffer* buf;
834
835 /* ignore the request if message/feedback buffer isn't mapped */
836 if (!dec->msg || !dec->fb)
837 return;
838
839 /* grab the current message buffer */
840 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
841
842 /* unmap the buffer */
843 dec->ws->buffer_unmap(buf->res->buf);
844 dec->msg = NULL;
845 dec->fb = NULL;
846 dec->it = NULL;
847
848 if (dec->sessionctx.res)
849 send_cmd(dec, RDECODE_CMD_SESSION_CONTEXT_BUFFER,
850 dec->sessionctx.res->buf, 0, RADEON_USAGE_READWRITE,
851 RADEON_DOMAIN_VRAM);
852
853 /* and send it to the hardware */
854 send_cmd(dec, RDECODE_CMD_MSG_BUFFER, buf->res->buf, 0,
855 RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
856 }
857
858 /* cycle to the next set of buffers */
next_buffer(struct radeon_decoder * dec)859 static void next_buffer(struct radeon_decoder *dec)
860 {
861 ++dec->cur_buffer;
862 dec->cur_buffer %= NUM_BUFFERS;
863 }
864
calc_ctx_size_h264_perf(struct radeon_decoder * dec)865 static unsigned calc_ctx_size_h264_perf(struct radeon_decoder *dec)
866 {
867 unsigned width_in_mb, height_in_mb, ctx_size;
868 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
869 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
870
871 unsigned max_references = dec->base.max_references + 1;
872
873 // picture width & height in 16 pixel units
874 width_in_mb = width / VL_MACROBLOCK_WIDTH;
875 height_in_mb = align(height / VL_MACROBLOCK_HEIGHT, 2);
876
877 unsigned fs_in_mb = width_in_mb * height_in_mb;
878 unsigned num_dpb_buffer;
879 switch(dec->base.level) {
880 case 30:
881 num_dpb_buffer = 8100 / fs_in_mb;
882 break;
883 case 31:
884 num_dpb_buffer = 18000 / fs_in_mb;
885 break;
886 case 32:
887 num_dpb_buffer = 20480 / fs_in_mb;
888 break;
889 case 41:
890 num_dpb_buffer = 32768 / fs_in_mb;
891 break;
892 case 42:
893 num_dpb_buffer = 34816 / fs_in_mb;
894 break;
895 case 50:
896 num_dpb_buffer = 110400 / fs_in_mb;
897 break;
898 case 51:
899 num_dpb_buffer = 184320 / fs_in_mb;
900 break;
901 default:
902 num_dpb_buffer = 184320 / fs_in_mb;
903 break;
904 }
905 num_dpb_buffer++;
906 max_references = MAX2(MIN2(NUM_H264_REFS, num_dpb_buffer), max_references);
907 ctx_size = max_references * align(width_in_mb * height_in_mb * 192, 256);
908
909 return ctx_size;
910 }
911
912 /* calculate size of reference picture buffer */
calc_dpb_size(struct radeon_decoder * dec)913 static unsigned calc_dpb_size(struct radeon_decoder *dec)
914 {
915 unsigned width_in_mb, height_in_mb, image_size, dpb_size;
916
917 // always align them to MB size for dpb calculation
918 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
919 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
920
921 // always one more for currently decoded picture
922 unsigned max_references = dec->base.max_references + 1;
923
924 // aligned size of a single frame
925 image_size = align(width, 32) * height;
926 image_size += image_size / 2;
927 image_size = align(image_size, 1024);
928
929 // picture width & height in 16 pixel units
930 width_in_mb = width / VL_MACROBLOCK_WIDTH;
931 height_in_mb = align(height / VL_MACROBLOCK_HEIGHT, 2);
932
933 switch (u_reduce_video_profile(dec->base.profile)) {
934 case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
935 unsigned fs_in_mb = width_in_mb * height_in_mb;
936 unsigned num_dpb_buffer;
937
938 switch(dec->base.level) {
939 case 30:
940 num_dpb_buffer = 8100 / fs_in_mb;
941 break;
942 case 31:
943 num_dpb_buffer = 18000 / fs_in_mb;
944 break;
945 case 32:
946 num_dpb_buffer = 20480 / fs_in_mb;
947 break;
948 case 41:
949 num_dpb_buffer = 32768 / fs_in_mb;
950 break;
951 case 42:
952 num_dpb_buffer = 34816 / fs_in_mb;
953 break;
954 case 50:
955 num_dpb_buffer = 110400 / fs_in_mb;
956 break;
957 case 51:
958 num_dpb_buffer = 184320 / fs_in_mb;
959 break;
960 default:
961 num_dpb_buffer = 184320 / fs_in_mb;
962 break;
963 }
964 num_dpb_buffer++;
965 max_references = MAX2(MIN2(NUM_H264_REFS, num_dpb_buffer), max_references);
966 dpb_size = image_size * max_references;
967 break;
968 }
969
970 case PIPE_VIDEO_FORMAT_HEVC:
971 if (dec->base.width * dec->base.height >= 4096*2000)
972 max_references = MAX2(max_references, 8);
973 else
974 max_references = MAX2(max_references, 17);
975
976 width = align (width, 16);
977 height = align (height, 16);
978 if (dec->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10)
979 dpb_size = align((align(width, 32) * height * 9) / 4, 256) * max_references;
980 else
981 dpb_size = align((align(width, 32) * height * 3) / 2, 256) * max_references;
982 break;
983
984 case PIPE_VIDEO_FORMAT_VC1:
985 // the firmware seems to allways assume a minimum of ref frames
986 max_references = MAX2(NUM_VC1_REFS, max_references);
987
988 // reference picture buffer
989 dpb_size = image_size * max_references;
990
991 // CONTEXT_BUFFER
992 dpb_size += width_in_mb * height_in_mb * 128;
993
994 // IT surface buffer
995 dpb_size += width_in_mb * 64;
996
997 // DB surface buffer
998 dpb_size += width_in_mb * 128;
999
1000 // BP
1001 dpb_size += align(MAX2(width_in_mb, height_in_mb) * 7 * 16, 64);
1002 break;
1003
1004 case PIPE_VIDEO_FORMAT_MPEG12:
1005 // reference picture buffer, must be big enough for all frames
1006 dpb_size = image_size * NUM_MPEG2_REFS;
1007 break;
1008
1009 case PIPE_VIDEO_FORMAT_MPEG4:
1010 // reference picture buffer
1011 dpb_size = image_size * max_references;
1012
1013 // CM
1014 dpb_size += width_in_mb * height_in_mb * 64;
1015
1016 // IT surface buffer
1017 dpb_size += align(width_in_mb * height_in_mb * 32, 64);
1018
1019 dpb_size = MAX2(dpb_size, 30 * 1024 * 1024);
1020 break;
1021
1022 default:
1023 // something is missing here
1024 assert(0);
1025
1026 // at least use a sane default value
1027 dpb_size = 32 * 1024 * 1024;
1028 break;
1029 }
1030 return dpb_size;
1031 }
1032
1033 /**
1034 * destroy this video decoder
1035 */
radeon_dec_destroy(struct pipe_video_codec * decoder)1036 static void radeon_dec_destroy(struct pipe_video_codec *decoder)
1037 {
1038 struct radeon_decoder *dec = (struct radeon_decoder*)decoder;
1039 unsigned i;
1040
1041 assert(decoder);
1042
1043 map_msg_fb_it_buf(dec);
1044 rvcn_dec_message_destroy(dec);
1045 send_msg_buf(dec);
1046
1047 flush(dec, 0);
1048
1049 dec->ws->cs_destroy(dec->cs);
1050
1051 for (i = 0; i < NUM_BUFFERS; ++i) {
1052 si_vid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
1053 si_vid_destroy_buffer(&dec->bs_buffers[i]);
1054 }
1055
1056 si_vid_destroy_buffer(&dec->dpb);
1057 si_vid_destroy_buffer(&dec->ctx);
1058 si_vid_destroy_buffer(&dec->sessionctx);
1059
1060 FREE(dec);
1061 }
1062
1063 /**
1064 * start decoding of a new frame
1065 */
radeon_dec_begin_frame(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture)1066 static void radeon_dec_begin_frame(struct pipe_video_codec *decoder,
1067 struct pipe_video_buffer *target,
1068 struct pipe_picture_desc *picture)
1069 {
1070 struct radeon_decoder *dec = (struct radeon_decoder*)decoder;
1071 uintptr_t frame;
1072
1073 assert(decoder);
1074
1075 frame = ++dec->frame_number;
1076 vl_video_buffer_set_associated_data(target, decoder, (void *)frame,
1077 &radeon_dec_destroy_associated_data);
1078
1079 dec->bs_size = 0;
1080 dec->bs_ptr = dec->ws->buffer_map(
1081 dec->bs_buffers[dec->cur_buffer].res->buf,
1082 dec->cs, PIPE_TRANSFER_WRITE);
1083 }
1084
1085 /**
1086 * decode a macroblock
1087 */
radeon_dec_decode_macroblock(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture,const struct pipe_macroblock * macroblocks,unsigned num_macroblocks)1088 static void radeon_dec_decode_macroblock(struct pipe_video_codec *decoder,
1089 struct pipe_video_buffer *target,
1090 struct pipe_picture_desc *picture,
1091 const struct pipe_macroblock *macroblocks,
1092 unsigned num_macroblocks)
1093 {
1094 /* not supported (yet) */
1095 assert(0);
1096 }
1097
1098 /**
1099 * decode a bitstream
1100 */
radeon_dec_decode_bitstream(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture,unsigned num_buffers,const void * const * buffers,const unsigned * sizes)1101 static void radeon_dec_decode_bitstream(struct pipe_video_codec *decoder,
1102 struct pipe_video_buffer *target,
1103 struct pipe_picture_desc *picture,
1104 unsigned num_buffers,
1105 const void * const *buffers,
1106 const unsigned *sizes)
1107 {
1108 struct radeon_decoder *dec = (struct radeon_decoder*)decoder;
1109 unsigned i;
1110
1111 assert(decoder);
1112
1113 if (!dec->bs_ptr)
1114 return;
1115
1116 for (i = 0; i < num_buffers; ++i) {
1117 struct rvid_buffer *buf = &dec->bs_buffers[dec->cur_buffer];
1118 unsigned new_size = dec->bs_size + sizes[i];
1119
1120 if (new_size > buf->res->buf->size) {
1121 dec->ws->buffer_unmap(buf->res->buf);
1122 if (!si_vid_resize_buffer(dec->screen, dec->cs, buf, new_size)) {
1123 RVID_ERR("Can't resize bitstream buffer!");
1124 return;
1125 }
1126
1127 dec->bs_ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
1128 PIPE_TRANSFER_WRITE);
1129 if (!dec->bs_ptr)
1130 return;
1131
1132 dec->bs_ptr += dec->bs_size;
1133 }
1134
1135 memcpy(dec->bs_ptr, buffers[i], sizes[i]);
1136 dec->bs_size += sizes[i];
1137 dec->bs_ptr += sizes[i];
1138 }
1139 }
1140
1141 /**
1142 * end decoding of the current frame
1143 */
radeon_dec_end_frame(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture)1144 static void radeon_dec_end_frame(struct pipe_video_codec *decoder,
1145 struct pipe_video_buffer *target,
1146 struct pipe_picture_desc *picture)
1147 {
1148 struct radeon_decoder *dec = (struct radeon_decoder*)decoder;
1149 struct pb_buffer *dt;
1150 struct rvid_buffer *msg_fb_it_buf, *bs_buf;
1151
1152 assert(decoder);
1153
1154 if (!dec->bs_ptr)
1155 return;
1156
1157 msg_fb_it_buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
1158 bs_buf = &dec->bs_buffers[dec->cur_buffer];
1159
1160 memset(dec->bs_ptr, 0, align(dec->bs_size, 128) - dec->bs_size);
1161 dec->ws->buffer_unmap(bs_buf->res->buf);
1162
1163 map_msg_fb_it_buf(dec);
1164 dt = rvcn_dec_message_decode(dec, target, picture);
1165 rvcn_dec_message_feedback(dec);
1166 send_msg_buf(dec);
1167
1168 send_cmd(dec, RDECODE_CMD_DPB_BUFFER, dec->dpb.res->buf, 0,
1169 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1170 if (dec->ctx.res)
1171 send_cmd(dec, RDECODE_CMD_CONTEXT_BUFFER, dec->ctx.res->buf, 0,
1172 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1173 send_cmd(dec, RDECODE_CMD_BITSTREAM_BUFFER, bs_buf->res->buf,
1174 0, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1175 send_cmd(dec, RDECODE_CMD_DECODING_TARGET_BUFFER, dt, 0,
1176 RADEON_USAGE_WRITE, RADEON_DOMAIN_VRAM);
1177 send_cmd(dec, RDECODE_CMD_FEEDBACK_BUFFER, msg_fb_it_buf->res->buf,
1178 FB_BUFFER_OFFSET, RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT);
1179 if (have_it(dec))
1180 send_cmd(dec, RDECODE_CMD_IT_SCALING_TABLE_BUFFER, msg_fb_it_buf->res->buf,
1181 FB_BUFFER_OFFSET + FB_BUFFER_SIZE, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1182 set_reg(dec, RDECODE_ENGINE_CNTL, 1);
1183
1184 flush(dec, PIPE_FLUSH_ASYNC);
1185 next_buffer(dec);
1186 }
1187
1188 /**
1189 * flush any outstanding command buffers to the hardware
1190 */
radeon_dec_flush(struct pipe_video_codec * decoder)1191 static void radeon_dec_flush(struct pipe_video_codec *decoder)
1192 {
1193 }
1194
1195 /**
1196 * create and HW decoder
1197 */
radeon_create_decoder(struct pipe_context * context,const struct pipe_video_codec * templ)1198 struct pipe_video_codec *radeon_create_decoder(struct pipe_context *context,
1199 const struct pipe_video_codec *templ)
1200 {
1201 struct radeon_winsys* ws = ((struct r600_common_context *)context)->ws;
1202 struct r600_common_context *rctx = (struct r600_common_context*)context;
1203 unsigned width = templ->width, height = templ->height;
1204 unsigned dpb_size, bs_buf_size, stream_type = 0;
1205 struct radeon_decoder *dec;
1206 int r, i;
1207
1208 switch(u_reduce_video_profile(templ->profile)) {
1209 case PIPE_VIDEO_FORMAT_MPEG12:
1210 if (templ->entrypoint > PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
1211 return vl_create_mpeg12_decoder(context, templ);
1212 stream_type = RDECODE_CODEC_MPEG2_VLD;
1213 break;
1214 case PIPE_VIDEO_FORMAT_MPEG4:
1215 width = align(width, VL_MACROBLOCK_WIDTH);
1216 height = align(height, VL_MACROBLOCK_HEIGHT);
1217 stream_type = RDECODE_CODEC_MPEG4;
1218 break;
1219 case PIPE_VIDEO_FORMAT_VC1:
1220 stream_type = RDECODE_CODEC_VC1;
1221 break;
1222 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
1223 width = align(width, VL_MACROBLOCK_WIDTH);
1224 height = align(height, VL_MACROBLOCK_HEIGHT);
1225 stream_type = RDECODE_CODEC_H264_PERF;
1226 break;
1227 case PIPE_VIDEO_FORMAT_HEVC:
1228 stream_type = RDECODE_CODEC_H265;
1229 break;
1230 default:
1231 assert(0);
1232 break;
1233 }
1234
1235 dec = CALLOC_STRUCT(radeon_decoder);
1236
1237 if (!dec)
1238 return NULL;
1239
1240 dec->base = *templ;
1241 dec->base.context = context;
1242 dec->base.width = width;
1243 dec->base.height = height;
1244
1245 dec->base.destroy = radeon_dec_destroy;
1246 dec->base.begin_frame = radeon_dec_begin_frame;
1247 dec->base.decode_macroblock = radeon_dec_decode_macroblock;
1248 dec->base.decode_bitstream = radeon_dec_decode_bitstream;
1249 dec->base.end_frame = radeon_dec_end_frame;
1250 dec->base.flush = radeon_dec_flush;
1251
1252 dec->stream_type = stream_type;
1253 dec->stream_handle = si_vid_alloc_stream_handle();
1254 dec->screen = context->screen;
1255 dec->ws = ws;
1256 dec->cs = ws->cs_create(rctx->ctx, RING_VCN_DEC, NULL, NULL);
1257 if (!dec->cs) {
1258 RVID_ERR("Can't get command submission context.\n");
1259 goto error;
1260 }
1261
1262 for (i = 0; i < 16; i++)
1263 dec->render_pic_list[i] = NULL;
1264 bs_buf_size = width * height * (512 / (16 * 16));
1265 for (i = 0; i < NUM_BUFFERS; ++i) {
1266 unsigned msg_fb_it_size = FB_BUFFER_OFFSET + FB_BUFFER_SIZE;
1267 if (have_it(dec))
1268 msg_fb_it_size += IT_SCALING_TABLE_SIZE;
1269 /* use vram to improve performance, workaround an unknown bug */
1270 if (!si_vid_create_buffer(dec->screen, &dec->msg_fb_it_buffers[i],
1271 msg_fb_it_size, PIPE_USAGE_DEFAULT)) {
1272 RVID_ERR("Can't allocated message buffers.\n");
1273 goto error;
1274 }
1275
1276 if (!si_vid_create_buffer(dec->screen, &dec->bs_buffers[i],
1277 bs_buf_size, PIPE_USAGE_STAGING)) {
1278 RVID_ERR("Can't allocated bitstream buffers.\n");
1279 goto error;
1280 }
1281
1282 si_vid_clear_buffer(context, &dec->msg_fb_it_buffers[i]);
1283 si_vid_clear_buffer(context, &dec->bs_buffers[i]);
1284 }
1285
1286 dpb_size = calc_dpb_size(dec);
1287
1288 if (!si_vid_create_buffer(dec->screen, &dec->dpb, dpb_size, PIPE_USAGE_DEFAULT)) {
1289 RVID_ERR("Can't allocated dpb.\n");
1290 goto error;
1291 }
1292
1293 si_vid_clear_buffer(context, &dec->dpb);
1294
1295 if (dec->stream_type == RDECODE_CODEC_H264_PERF) {
1296 unsigned ctx_size = calc_ctx_size_h264_perf(dec);
1297 if (!si_vid_create_buffer(dec->screen, &dec->ctx, ctx_size, PIPE_USAGE_DEFAULT)) {
1298 RVID_ERR("Can't allocated context buffer.\n");
1299 goto error;
1300 }
1301 si_vid_clear_buffer(context, &dec->ctx);
1302 }
1303
1304 if (!si_vid_create_buffer(dec->screen, &dec->sessionctx,
1305 RDECODE_SESSION_CONTEXT_SIZE,
1306 PIPE_USAGE_DEFAULT)) {
1307 RVID_ERR("Can't allocated session ctx.\n");
1308 goto error;
1309 }
1310 si_vid_clear_buffer(context, &dec->sessionctx);
1311
1312 map_msg_fb_it_buf(dec);
1313 rvcn_dec_message_create(dec);
1314 send_msg_buf(dec);
1315 r = flush(dec, 0);
1316 if (r)
1317 goto error;
1318
1319 next_buffer(dec);
1320
1321 return &dec->base;
1322
1323 error:
1324 if (dec->cs) dec->ws->cs_destroy(dec->cs);
1325
1326 for (i = 0; i < NUM_BUFFERS; ++i) {
1327 si_vid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
1328 si_vid_destroy_buffer(&dec->bs_buffers[i]);
1329 }
1330
1331 si_vid_destroy_buffer(&dec->dpb);
1332 si_vid_destroy_buffer(&dec->ctx);
1333 si_vid_destroy_buffer(&dec->sessionctx);
1334
1335 FREE(dec);
1336
1337 return NULL;
1338 }
1339