projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radeon/uvd: silence a warning
[mesa.git] / src / gallium / drivers / radeon / radeon_uvd.c
1 /**************************************************************************
2 *
3 * Copyright 2011 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 /*
29 * Authors:
30 * Christian König <christian.koenig@amd.com>
31 *
32 */
33
34 #include <sys/types.h>
35 #include <assert.h>
36 #include <errno.h>
37 #include <unistd.h>
38 #include <stdio.h>
39
40 #include "pipe/p_video_codec.h"
41
42 #include "util/u_memory.h"
43 #include "util/u_video.h"
44
45 #include "vl/vl_defines.h"
46 #include "vl/vl_mpeg12_decoder.h"
47
48 #include "r600_pipe_common.h"
49 #include "radeon_video.h"
50 #include "radeon_uvd.h"
51
52 #define NUM_BUFFERS 4
53
54 #define NUM_MPEG2_REFS 6
55 #define NUM_H264_REFS 17
56 #define NUM_VC1_REFS 5
57
58 #define FB_BUFFER_OFFSET 0x1000
59 #define FB_BUFFER_SIZE 2048
60 #define IT_SCALING_TABLE_SIZE 992
61
62 /* UVD decoder representation */
63 struct ruvd_decoder {
64 struct pipe_video_codec base;
65
66 ruvd_set_dtb set_dtb;
67
68 unsigned stream_handle;
69 unsigned stream_type;
70 unsigned frame_number;
71
72 struct pipe_screen *screen;
73 struct radeon_winsys* ws;
74 struct radeon_winsys_cs* cs;
75
76 unsigned cur_buffer;
77
78 struct rvid_buffer msg_fb_it_buffers[NUM_BUFFERS];
79 struct ruvd_msg *msg;
80 uint32_t *fb;
81 uint8_t *it;
82
83 struct rvid_buffer bs_buffers[NUM_BUFFERS];
84 void* bs_ptr;
85 unsigned bs_size;
86
87 struct rvid_buffer dpb;
88 bool use_legacy;
89 struct rvid_buffer ctx;
90 };
91
92 /* flush IB to the hardware */
93 static void flush(struct ruvd_decoder *dec)
94 {
95 dec->ws->cs_flush(dec->cs, RADEON_FLUSH_ASYNC, NULL, 0);
96 }
97
98 /* add a new set register command to the IB */
99 static void set_reg(struct ruvd_decoder *dec, unsigned reg, uint32_t val)
100 {
101 uint32_t *pm4 = dec->cs->buf;
102 pm4[dec->cs->cdw++] = RUVD_PKT0(reg >> 2, 0);
103 pm4[dec->cs->cdw++] = val;
104 }
105
106 /* send a command to the VCPU through the GPCOM registers */
107 static void send_cmd(struct ruvd_decoder *dec, unsigned cmd,
108 struct pb_buffer* buf, uint32_t off,
109 enum radeon_bo_usage usage, enum radeon_bo_domain domain)
110 {
111 int reloc_idx;
112
113 reloc_idx = dec->ws->cs_add_buffer(dec->cs, buf, usage, domain,
114 RADEON_PRIO_UVD);
115 if (!dec->use_legacy) {
116 uint64_t addr;
117 addr = dec->ws->buffer_get_virtual_address(buf);
118 addr = addr + off;
119 set_reg(dec, RUVD_GPCOM_VCPU_DATA0, addr);
120 set_reg(dec, RUVD_GPCOM_VCPU_DATA1, addr >> 32);
121 } else {
122 set_reg(dec, RUVD_GPCOM_VCPU_DATA0, off);
123 set_reg(dec, RUVD_GPCOM_VCPU_DATA1, reloc_idx * 4);
124 }
125 set_reg(dec, RUVD_GPCOM_VCPU_CMD, cmd << 1);
126 }
127
128 /* do the codec needs an IT buffer ?*/
129 static bool have_it(struct ruvd_decoder *dec)
130 {
131 return dec->stream_type == RUVD_CODEC_H264_PERF ||
132 dec->stream_type == RUVD_CODEC_H265;
133 }
134
135 /* map the next available message/feedback/itscaling buffer */
136 static void map_msg_fb_it_buf(struct ruvd_decoder *dec)
137 {
138 struct rvid_buffer* buf;
139 uint8_t *ptr;
140
141 /* grab the current message/feedback buffer */
142 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
143
144 /* and map it for CPU access */
145 ptr = dec->ws->buffer_map(buf->res->buf, dec->cs, PIPE_TRANSFER_WRITE);
146
147 /* calc buffer offsets */
148 dec->msg = (struct ruvd_msg *)ptr;
149 dec->fb = (uint32_t *)(ptr + FB_BUFFER_OFFSET);
150 if (have_it(dec))
151 dec->it = (uint8_t *)(ptr + FB_BUFFER_OFFSET + FB_BUFFER_SIZE);
152 }
153
154 /* unmap and send a message command to the VCPU */
155 static void send_msg_buf(struct ruvd_decoder *dec)
156 {
157 struct rvid_buffer* buf;
158
159 /* ignore the request if message/feedback buffer isn't mapped */
160 if (!dec->msg || !dec->fb)
161 return;
162
163 /* grab the current message buffer */
164 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
165
166 /* unmap the buffer */
167 dec->ws->buffer_unmap(buf->res->buf);
168 dec->msg = NULL;
169 dec->fb = NULL;
170 dec->it = NULL;
171
172 /* and send it to the hardware */
173 send_cmd(dec, RUVD_CMD_MSG_BUFFER, buf->res->buf, 0,
174 RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
175 }
176
177 /* cycle to the next set of buffers */
178 static void next_buffer(struct ruvd_decoder *dec)
179 {
180 ++dec->cur_buffer;
181 dec->cur_buffer %= NUM_BUFFERS;
182 }
183
184 /* convert the profile into something UVD understands */
185 static uint32_t profile2stream_type(struct ruvd_decoder *dec, unsigned family)
186 {
187 switch (u_reduce_video_profile(dec->base.profile)) {
188 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
189 return (family >= CHIP_TONGA) ?
190 RUVD_CODEC_H264_PERF : RUVD_CODEC_H264;
191
192 case PIPE_VIDEO_FORMAT_VC1:
193 return RUVD_CODEC_VC1;
194
195 case PIPE_VIDEO_FORMAT_MPEG12:
196 return RUVD_CODEC_MPEG2;
197
198 case PIPE_VIDEO_FORMAT_MPEG4:
199 return RUVD_CODEC_MPEG4;
200
201 case PIPE_VIDEO_FORMAT_HEVC:
202 return RUVD_CODEC_H265;
203
204 default:
205 assert(0);
206 return 0;
207 }
208 }
209
210 static unsigned calc_ctx_size(struct ruvd_decoder *dec)
211 {
212 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
213 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
214
215 unsigned max_references = dec->base.max_references + 1;
216
217 if (dec->base.width * dec->base.height >= 4096*2000)
218 max_references = MAX2(max_references, 8);
219 else
220 max_references = MAX2(max_references, 17);
221
222 width = align (width, 16);
223 height = align (height, 16);
224 return ((width + 255) / 16) * ((height + 255) / 16) * 16 * max_references + 52 * 1024;
225 }
226
227 /* calculate size of reference picture buffer */
228 static unsigned calc_dpb_size(struct ruvd_decoder *dec)
229 {
230 unsigned width_in_mb, height_in_mb, image_size, dpb_size;
231
232 // always align them to MB size for dpb calculation
233 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
234 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
235
236 // always one more for currently decoded picture
237 unsigned max_references = dec->base.max_references + 1;
238
239 // aligned size of a single frame
240 image_size = width * height;
241 image_size += image_size / 2;
242 image_size = align(image_size, 1024);
243
244 // picture width & height in 16 pixel units
245 width_in_mb = width / VL_MACROBLOCK_WIDTH;
246 height_in_mb = align(height / VL_MACROBLOCK_HEIGHT, 2);
247
248 switch (u_reduce_video_profile(dec->base.profile)) {
249 case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
250 if (!dec->use_legacy) {
251 unsigned fs_in_mb = width_in_mb * height_in_mb;
252 unsigned alignment = 64, num_dpb_buffer;
253
254 if (dec->stream_type == RUVD_CODEC_H264_PERF)
255 alignment = 256;
256 switch(dec->base.level) {
257 case 30:
258 num_dpb_buffer = 8100 / fs_in_mb;
259 break;
260 case 31:
261 num_dpb_buffer = 18000 / fs_in_mb;
262 break;
263 case 32:
264 num_dpb_buffer = 20480 / fs_in_mb;
265 break;
266 case 41:
267 num_dpb_buffer = 32768 / fs_in_mb;
268 break;
269 case 42:
270 num_dpb_buffer = 34816 / fs_in_mb;
271 break;
272 case 50:
273 num_dpb_buffer = 110400 / fs_in_mb;
274 break;
275 case 51:
276 num_dpb_buffer = 184320 / fs_in_mb;
277 break;
278 default:
279 num_dpb_buffer = 184320 / fs_in_mb;
280 break;
281 }
282 num_dpb_buffer++;
283 max_references = MAX2(MIN2(NUM_H264_REFS, num_dpb_buffer), max_references);
284 dpb_size = image_size * max_references;
285 dpb_size += max_references * align(width_in_mb * height_in_mb * 192, alignment);
286 dpb_size += align(width_in_mb * height_in_mb * 32, alignment);
287 } else {
288 // the firmware seems to allways assume a minimum of ref frames
289 max_references = MAX2(NUM_H264_REFS, max_references);
290 // reference picture buffer
291 dpb_size = image_size * max_references;
292 // macroblock context buffer
293 dpb_size += width_in_mb * height_in_mb * max_references * 192;
294 // IT surface buffer
295 dpb_size += width_in_mb * height_in_mb * 32;
296 }
297 break;
298 }
299
300 case PIPE_VIDEO_FORMAT_HEVC:
301 if (dec->base.width * dec->base.height >= 4096*2000)
302 max_references = MAX2(max_references, 8);
303 else
304 max_references = MAX2(max_references, 17);
305
306 width = align (width, 16);
307 height = align (height, 16);
308 dpb_size = align((width * height * 3) / 2, 256) * max_references;
309 break;
310
311 case PIPE_VIDEO_FORMAT_VC1:
312 // the firmware seems to allways assume a minimum of ref frames
313 max_references = MAX2(NUM_VC1_REFS, max_references);
314
315 // reference picture buffer
316 dpb_size = image_size * max_references;
317
318 // CONTEXT_BUFFER
319 dpb_size += width_in_mb * height_in_mb * 128;
320
321 // IT surface buffer
322 dpb_size += width_in_mb * 64;
323
324 // DB surface buffer
325 dpb_size += width_in_mb * 128;
326
327 // BP
328 dpb_size += align(MAX2(width_in_mb, height_in_mb) * 7 * 16, 64);
329 break;
330
331 case PIPE_VIDEO_FORMAT_MPEG12:
332 // reference picture buffer, must be big enough for all frames
333 dpb_size = image_size * NUM_MPEG2_REFS;
334 break;
335
336 case PIPE_VIDEO_FORMAT_MPEG4:
337 // reference picture buffer
338 dpb_size = image_size * max_references;
339
340 // CM
341 dpb_size += width_in_mb * height_in_mb * 64;
342
343 // IT surface buffer
344 dpb_size += align(width_in_mb * height_in_mb * 32, 64);
345
346 dpb_size = MAX2(dpb_size, 30 * 1024 * 1024);
347 break;
348
349 default:
350 // something is missing here
351 assert(0);
352
353 // at least use a sane default value
354 dpb_size = 32 * 1024 * 1024;
355 break;
356 }
357 return dpb_size;
358 }
359
360 /* free associated data in the video buffer callback */
361 static void ruvd_destroy_associated_data(void *data)
362 {
363 /* NOOP, since we only use an intptr */
364 }
365
366 /* get h264 specific message bits */
367 static struct ruvd_h264 get_h264_msg(struct ruvd_decoder *dec, struct pipe_h264_picture_desc *pic)
368 {
369 struct ruvd_h264 result;
370
371 memset(&result, 0, sizeof(result));
372 switch (pic->base.profile) {
373 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE:
374 result.profile = RUVD_H264_PROFILE_BASELINE;
375 break;
376
377 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN:
378 result.profile = RUVD_H264_PROFILE_MAIN;
379 break;
380
381 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH:
382 result.profile = RUVD_H264_PROFILE_HIGH;
383 break;
384
385 default:
386 assert(0);
387 break;
388 }
389
390 result.level = dec->base.level;
391
392 result.sps_info_flags = 0;
393 result.sps_info_flags |= pic->pps->sps->direct_8x8_inference_flag << 0;
394 result.sps_info_flags |= pic->pps->sps->mb_adaptive_frame_field_flag << 1;
395 result.sps_info_flags |= pic->pps->sps->frame_mbs_only_flag << 2;
396 result.sps_info_flags |= pic->pps->sps->delta_pic_order_always_zero_flag << 3;
397
398 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8;
399 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8;
400 result.log2_max_frame_num_minus4 = pic->pps->sps->log2_max_frame_num_minus4;
401 result.pic_order_cnt_type = pic->pps->sps->pic_order_cnt_type;
402 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4;
403
404 switch (dec->base.chroma_format) {
405 case PIPE_VIDEO_CHROMA_FORMAT_NONE:
406 /* TODO: assert? */
407 break;
408 case PIPE_VIDEO_CHROMA_FORMAT_400:
409 result.chroma_format = 0;
410 break;
411 case PIPE_VIDEO_CHROMA_FORMAT_420:
412 result.chroma_format = 1;
413 break;
414 case PIPE_VIDEO_CHROMA_FORMAT_422:
415 result.chroma_format = 2;
416 break;
417 case PIPE_VIDEO_CHROMA_FORMAT_444:
418 result.chroma_format = 3;
419 break;
420 }
421
422 result.pps_info_flags = 0;
423 result.pps_info_flags |= pic->pps->transform_8x8_mode_flag << 0;
424 result.pps_info_flags |= pic->pps->redundant_pic_cnt_present_flag << 1;
425 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 2;
426 result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag << 3;
427 result.pps_info_flags |= pic->pps->weighted_bipred_idc << 4;
428 result.pps_info_flags |= pic->pps->weighted_pred_flag << 6;
429 result.pps_info_flags |= pic->pps->bottom_field_pic_order_in_frame_present_flag << 7;
430 result.pps_info_flags |= pic->pps->entropy_coding_mode_flag << 8;
431
432 result.num_slice_groups_minus1 = pic->pps->num_slice_groups_minus1;
433 result.slice_group_map_type = pic->pps->slice_group_map_type;
434 result.slice_group_change_rate_minus1 = pic->pps->slice_group_change_rate_minus1;
435 result.pic_init_qp_minus26 = pic->pps->pic_init_qp_minus26;
436 result.chroma_qp_index_offset = pic->pps->chroma_qp_index_offset;
437 result.second_chroma_qp_index_offset = pic->pps->second_chroma_qp_index_offset;
438
439 memcpy(result.scaling_list_4x4, pic->pps->ScalingList4x4, 6*16);
440 memcpy(result.scaling_list_8x8, pic->pps->ScalingList8x8, 2*64);
441
442 if (dec->stream_type == RUVD_CODEC_H264_PERF) {
443 memcpy(dec->it, result.scaling_list_4x4, 6*16);
444 memcpy((dec->it + 96), result.scaling_list_8x8, 2*64);
445 }
446
447 result.num_ref_frames = pic->num_ref_frames;
448
449 result.num_ref_idx_l0_active_minus1 = pic->num_ref_idx_l0_active_minus1;
450 result.num_ref_idx_l1_active_minus1 = pic->num_ref_idx_l1_active_minus1;
451
452 result.frame_num = pic->frame_num;
453 memcpy(result.frame_num_list, pic->frame_num_list, 4*16);
454 result.curr_field_order_cnt_list[0] = pic->field_order_cnt[0];
455 result.curr_field_order_cnt_list[1] = pic->field_order_cnt[1];
456 memcpy(result.field_order_cnt_list, pic->field_order_cnt_list, 4*16*2);
457
458 result.decoded_pic_idx = pic->frame_num;
459
460 return result;
461 }
462
463 /* get h265 specific message bits */
464 static struct ruvd_h265 get_h265_msg(struct ruvd_decoder *dec, struct pipe_video_buffer *target,
465 struct pipe_h265_picture_desc *pic)
466 {
467 struct ruvd_h265 result;
468 unsigned i;
469
470 memset(&result, 0, sizeof(result));
471
472 result.sps_info_flags = 0;
473 result.sps_info_flags |= pic->pps->sps->scaling_list_enabled_flag << 0;
474 result.sps_info_flags |= pic->pps->sps->amp_enabled_flag << 1;
475 result.sps_info_flags |= pic->pps->sps->sample_adaptive_offset_enabled_flag << 2;
476 result.sps_info_flags |= pic->pps->sps->pcm_enabled_flag << 3;
477 result.sps_info_flags |= pic->pps->sps->pcm_loop_filter_disabled_flag << 4;
478 result.sps_info_flags |= pic->pps->sps->long_term_ref_pics_present_flag << 5;
479 result.sps_info_flags |= pic->pps->sps->sps_temporal_mvp_enabled_flag << 6;
480 result.sps_info_flags |= pic->pps->sps->strong_intra_smoothing_enabled_flag << 7;
481 result.sps_info_flags |= pic->pps->sps->separate_colour_plane_flag << 8;
482 if (((struct r600_common_screen*)dec->screen)->family == CHIP_CARRIZO)
483 result.sps_info_flags |= 1 << 9;
484 if (pic->UseRefPicList == true)
485 result.sps_info_flags |= 1 << 10;
486
487 result.chroma_format = pic->pps->sps->chroma_format_idc;
488 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8;
489 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8;
490 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4;
491 result.sps_max_dec_pic_buffering_minus1 = pic->pps->sps->sps_max_dec_pic_buffering_minus1;
492 result.log2_min_luma_coding_block_size_minus3 = pic->pps->sps->log2_min_luma_coding_block_size_minus3;
493 result.log2_diff_max_min_luma_coding_block_size = pic->pps->sps->log2_diff_max_min_luma_coding_block_size;
494 result.log2_min_transform_block_size_minus2 = pic->pps->sps->log2_min_transform_block_size_minus2;
495 result.log2_diff_max_min_transform_block_size = pic->pps->sps->log2_diff_max_min_transform_block_size;
496 result.max_transform_hierarchy_depth_inter = pic->pps->sps->max_transform_hierarchy_depth_inter;
497 result.max_transform_hierarchy_depth_intra = pic->pps->sps->max_transform_hierarchy_depth_intra;
498 result.pcm_sample_bit_depth_luma_minus1 = pic->pps->sps->pcm_sample_bit_depth_luma_minus1;
499 result.pcm_sample_bit_depth_chroma_minus1 = pic->pps->sps->pcm_sample_bit_depth_chroma_minus1;
500 result.log2_min_pcm_luma_coding_block_size_minus3 = pic->pps->sps->log2_min_pcm_luma_coding_block_size_minus3;
501 result.log2_diff_max_min_pcm_luma_coding_block_size = pic->pps->sps->log2_diff_max_min_pcm_luma_coding_block_size;
502 result.num_short_term_ref_pic_sets = pic->pps->sps->num_short_term_ref_pic_sets;
503
504 result.pps_info_flags = 0;
505 result.pps_info_flags |= pic->pps->dependent_slice_segments_enabled_flag << 0;
506 result.pps_info_flags |= pic->pps->output_flag_present_flag << 1;
507 result.pps_info_flags |= pic->pps->sign_data_hiding_enabled_flag << 2;
508 result.pps_info_flags |= pic->pps->cabac_init_present_flag << 3;
509 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 4;
510 result.pps_info_flags |= pic->pps->transform_skip_enabled_flag << 5;
511 result.pps_info_flags |= pic->pps->cu_qp_delta_enabled_flag << 6;
512 result.pps_info_flags |= pic->pps->pps_slice_chroma_qp_offsets_present_flag << 7;
513 result.pps_info_flags |= pic->pps->weighted_pred_flag << 8;
514 result.pps_info_flags |= pic->pps->weighted_bipred_flag << 9;
515 result.pps_info_flags |= pic->pps->transquant_bypass_enabled_flag << 10;
516 result.pps_info_flags |= pic->pps->tiles_enabled_flag << 11;
517 result.pps_info_flags |= pic->pps->entropy_coding_sync_enabled_flag << 12;
518 result.pps_info_flags |= pic->pps->uniform_spacing_flag << 13;
519 result.pps_info_flags |= pic->pps->loop_filter_across_tiles_enabled_flag << 14;
520 result.pps_info_flags |= pic->pps->pps_loop_filter_across_slices_enabled_flag << 15;
521 result.pps_info_flags |= pic->pps->deblocking_filter_override_enabled_flag << 16;
522 result.pps_info_flags |= pic->pps->pps_deblocking_filter_disabled_flag << 17;
523 result.pps_info_flags |= pic->pps->lists_modification_present_flag << 18;
524 result.pps_info_flags |= pic->pps->slice_segment_header_extension_present_flag << 19;
525 //result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag; ???
526
527 result.num_extra_slice_header_bits = pic->pps->num_extra_slice_header_bits;
528 result.num_long_term_ref_pic_sps = pic->pps->sps->num_long_term_ref_pics_sps;
529 result.num_ref_idx_l0_default_active_minus1 = pic->pps->num_ref_idx_l0_default_active_minus1;
530 result.num_ref_idx_l1_default_active_minus1 = pic->pps->num_ref_idx_l1_default_active_minus1;
531 result.pps_cb_qp_offset = pic->pps->pps_cb_qp_offset;
532 result.pps_cr_qp_offset = pic->pps->pps_cr_qp_offset;
533 result.pps_beta_offset_div2 = pic->pps->pps_beta_offset_div2;
534 result.pps_tc_offset_div2 = pic->pps->pps_tc_offset_div2;
535 result.diff_cu_qp_delta_depth = pic->pps->diff_cu_qp_delta_depth;
536 result.num_tile_columns_minus1 = pic->pps->num_tile_columns_minus1;
537 result.num_tile_rows_minus1 = pic->pps->num_tile_rows_minus1;
538 result.log2_parallel_merge_level_minus2 = pic->pps->log2_parallel_merge_level_minus2;
539 result.init_qp_minus26 = pic->pps->init_qp_minus26;
540
541 for (i = 0; i < 19; ++i)
542 result.column_width_minus1[i] = pic->pps->column_width_minus1[i];
543
544 for (i = 0; i < 21; ++i)
545 result.row_height_minus1[i] = pic->pps->row_height_minus1[i];
546
547 result.num_delta_pocs_ref_rps_idx = pic->NumDeltaPocsOfRefRpsIdx;
548 result.curr_idx = pic->CurrPicOrderCntVal;
549 result.curr_poc = pic->CurrPicOrderCntVal;
550
551 vl_video_buffer_set_associated_data(target, &dec->base,
552 (void *)(uintptr_t)pic->CurrPicOrderCntVal,
553 &ruvd_destroy_associated_data);
554
555 for (i = 0; i < 16; ++i) {
556 struct pipe_video_buffer *ref = pic->ref[i];
557 uintptr_t ref_pic = 0;
558
559 result.poc_list[i] = pic->PicOrderCntVal[i];
560
561 if (ref)
562 ref_pic = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base);
563 else
564 ref_pic = 0x7F;
565 result.ref_pic_list[i] = ref_pic;
566 }
567
568 for (i = 0; i < 8; ++i) {
569 result.ref_pic_set_st_curr_before[i] = 0xFF;
570 result.ref_pic_set_st_curr_after[i] = 0xFF;
571 result.ref_pic_set_lt_curr[i] = 0xFF;
572 }
573
574 for (i = 0; i < pic->NumPocStCurrBefore; ++i)
575 result.ref_pic_set_st_curr_before[i] = pic->RefPicSetStCurrBefore[i];
576
577 for (i = 0; i < pic->NumPocStCurrAfter; ++i)
578 result.ref_pic_set_st_curr_after[i] = pic->RefPicSetStCurrAfter[i];
579
580 for (i = 0; i < pic->NumPocLtCurr; ++i)
581 result.ref_pic_set_lt_curr[i] = pic->RefPicSetLtCurr[i];
582
583 for (i = 0; i < 6; ++i)
584 result.ucScalingListDCCoefSizeID2[i] = pic->pps->sps->ScalingListDCCoeff16x16[i];
585
586 for (i = 0; i < 2; ++i)
587 result.ucScalingListDCCoefSizeID3[i] = pic->pps->sps->ScalingListDCCoeff32x32[i];
588
589 memcpy(dec->it, pic->pps->sps->ScalingList4x4, 6 * 16);
590 memcpy(dec->it + 96, pic->pps->sps->ScalingList8x8, 6 * 64);
591 memcpy(dec->it + 480, pic->pps->sps->ScalingList16x16, 6 * 64);
592 memcpy(dec->it + 864, pic->pps->sps->ScalingList32x32, 2 * 64);
593
594 for (i = 0 ; i < 2 ; i++) {
595 for (int j = 0 ; j < 15 ; j++)
596 result.direct_reflist[i][j] = pic->RefPicList[i][j];
597 }
598
599 /* TODO
600 result.highestTid;
601 result.isNonRef;
602
603 IDRPicFlag;
604 RAPPicFlag;
605 NumPocTotalCurr;
606 NumShortTermPictureSliceHeaderBits;
607 NumLongTermPictureSliceHeaderBits;
608
609 IsLongTerm[16];
610 */
611
612 return result;
613 }
614
615 /* get vc1 specific message bits */
616 static struct ruvd_vc1 get_vc1_msg(struct pipe_vc1_picture_desc *pic)
617 {
618 struct ruvd_vc1 result;
619
620 memset(&result, 0, sizeof(result));
621
622 switch(pic->base.profile) {
623 case PIPE_VIDEO_PROFILE_VC1_SIMPLE:
624 result.profile = RUVD_VC1_PROFILE_SIMPLE;
625 result.level = 1;
626 break;
627
628 case PIPE_VIDEO_PROFILE_VC1_MAIN:
629 result.profile = RUVD_VC1_PROFILE_MAIN;
630 result.level = 2;
631 break;
632
633 case PIPE_VIDEO_PROFILE_VC1_ADVANCED:
634 result.profile = RUVD_VC1_PROFILE_ADVANCED;
635 result.level = 4;
636 break;
637
638 default:
639 assert(0);
640 }
641
642 /* fields common for all profiles */
643 result.sps_info_flags |= pic->postprocflag << 7;
644 result.sps_info_flags |= pic->pulldown << 6;
645 result.sps_info_flags |= pic->interlace << 5;
646 result.sps_info_flags |= pic->tfcntrflag << 4;
647 result.sps_info_flags |= pic->finterpflag << 3;
648 result.sps_info_flags |= pic->psf << 1;
649
650 result.pps_info_flags |= pic->range_mapy_flag << 31;
651 result.pps_info_flags |= pic->range_mapy << 28;
652 result.pps_info_flags |= pic->range_mapuv_flag << 27;
653 result.pps_info_flags |= pic->range_mapuv << 24;
654 result.pps_info_flags |= pic->multires << 21;
655 result.pps_info_flags |= pic->maxbframes << 16;
656 result.pps_info_flags |= pic->overlap << 11;
657 result.pps_info_flags |= pic->quantizer << 9;
658 result.pps_info_flags |= pic->panscan_flag << 7;
659 result.pps_info_flags |= pic->refdist_flag << 6;
660 result.pps_info_flags |= pic->vstransform << 0;
661
662 /* some fields only apply to main/advanced profile */
663 if (pic->base.profile != PIPE_VIDEO_PROFILE_VC1_SIMPLE) {
664 result.pps_info_flags |= pic->syncmarker << 20;
665 result.pps_info_flags |= pic->rangered << 19;
666 result.pps_info_flags |= pic->loopfilter << 5;
667 result.pps_info_flags |= pic->fastuvmc << 4;
668 result.pps_info_flags |= pic->extended_mv << 3;
669 result.pps_info_flags |= pic->extended_dmv << 8;
670 result.pps_info_flags |= pic->dquant << 1;
671 }
672
673 result.chroma_format = 1;
674
675 #if 0
676 //(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT)
677 uint32_t slice_count
678 uint8_t picture_type
679 uint8_t frame_coding_mode
680 uint8_t deblockEnable
681 uint8_t pquant
682 #endif
683
684 return result;
685 }
686
687 /* extract the frame number from a referenced video buffer */
688 static uint32_t get_ref_pic_idx(struct ruvd_decoder *dec, struct pipe_video_buffer *ref)
689 {
690 uint32_t min = MAX2(dec->frame_number, NUM_MPEG2_REFS) - NUM_MPEG2_REFS;
691 uint32_t max = MAX2(dec->frame_number, 1) - 1;
692 uintptr_t frame;
693
694 /* seems to be the most sane fallback */
695 if (!ref)
696 return max;
697
698 /* get the frame number from the associated data */
699 frame = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base);
700
701 /* limit the frame number to a valid range */
702 return MAX2(MIN2(frame, max), min);
703 }
704
705 /* get mpeg2 specific msg bits */
706 static struct ruvd_mpeg2 get_mpeg2_msg(struct ruvd_decoder *dec,
707 struct pipe_mpeg12_picture_desc *pic)
708 {
709 const int *zscan = pic->alternate_scan ? vl_zscan_alternate : vl_zscan_normal;
710 struct ruvd_mpeg2 result;
711 unsigned i;
712
713 memset(&result, 0, sizeof(result));
714 result.decoded_pic_idx = dec->frame_number;
715 for (i = 0; i < 2; ++i)
716 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]);
717
718 result.load_intra_quantiser_matrix = 1;
719 result.load_nonintra_quantiser_matrix = 1;
720
721 for (i = 0; i < 64; ++i) {
722 result.intra_quantiser_matrix[i] = pic->intra_matrix[zscan[i]];
723 result.nonintra_quantiser_matrix[i] = pic->non_intra_matrix[zscan[i]];
724 }
725
726 result.profile_and_level_indication = 0;
727 result.chroma_format = 0x1;
728
729 result.picture_coding_type = pic->picture_coding_type;
730 result.f_code[0][0] = pic->f_code[0][0] + 1;
731 result.f_code[0][1] = pic->f_code[0][1] + 1;
732 result.f_code[1][0] = pic->f_code[1][0] + 1;
733 result.f_code[1][1] = pic->f_code[1][1] + 1;
734 result.intra_dc_precision = pic->intra_dc_precision;
735 result.pic_structure = pic->picture_structure;
736 result.top_field_first = pic->top_field_first;
737 result.frame_pred_frame_dct = pic->frame_pred_frame_dct;
738 result.concealment_motion_vectors = pic->concealment_motion_vectors;
739 result.q_scale_type = pic->q_scale_type;
740 result.intra_vlc_format = pic->intra_vlc_format;
741 result.alternate_scan = pic->alternate_scan;
742
743 return result;
744 }
745
746 /* get mpeg4 specific msg bits */
747 static struct ruvd_mpeg4 get_mpeg4_msg(struct ruvd_decoder *dec,
748 struct pipe_mpeg4_picture_desc *pic)
749 {
750 struct ruvd_mpeg4 result;
751 unsigned i;
752
753 memset(&result, 0, sizeof(result));
754 result.decoded_pic_idx = dec->frame_number;
755 for (i = 0; i < 2; ++i)
756 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]);
757
758 result.variant_type = 0;
759 result.profile_and_level_indication = 0xF0; // ASP Level0
760
761 result.video_object_layer_verid = 0x5; // advanced simple
762 result.video_object_layer_shape = 0x0; // rectangular
763
764 result.video_object_layer_width = dec->base.width;
765 result.video_object_layer_height = dec->base.height;
766
767 result.vop_time_increment_resolution = pic->vop_time_increment_resolution;
768
769 result.flags |= pic->short_video_header << 0;
770 //result.flags |= obmc_disable << 1;
771 result.flags |= pic->interlaced << 2;
772 result.flags |= 1 << 3; // load_intra_quant_mat
773 result.flags |= 1 << 4; // load_nonintra_quant_mat
774 result.flags |= pic->quarter_sample << 5;
775 result.flags |= 1 << 6; // complexity_estimation_disable
776 result.flags |= pic->resync_marker_disable << 7;
777 //result.flags |= data_partitioned << 8;
778 //result.flags |= reversible_vlc << 9;
779 result.flags |= 0 << 10; // newpred_enable
780 result.flags |= 0 << 11; // reduced_resolution_vop_enable
781 //result.flags |= scalability << 12;
782 //result.flags |= is_object_layer_identifier << 13;
783 //result.flags |= fixed_vop_rate << 14;
784 //result.flags |= newpred_segment_type << 15;
785
786 result.quant_type = pic->quant_type;
787
788 for (i = 0; i < 64; ++i) {
789 result.intra_quant_mat[i] = pic->intra_matrix[vl_zscan_normal[i]];
790 result.nonintra_quant_mat[i] = pic->non_intra_matrix[vl_zscan_normal[i]];
791 }
792
793 /*
794 int32_t trd [2]
795 int32_t trb [2]
796 uint8_t vop_coding_type
797 uint8_t vop_fcode_forward
798 uint8_t vop_fcode_backward
799 uint8_t rounding_control
800 uint8_t alternate_vertical_scan_flag
801 uint8_t top_field_first
802 */
803
804 return result;
805 }
806
807 /**
808 * destroy this video decoder
809 */
810 static void ruvd_destroy(struct pipe_video_codec *decoder)
811 {
812 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
813 unsigned i;
814
815 assert(decoder);
816
817 map_msg_fb_it_buf(dec);
818 memset(dec->msg, 0, sizeof(*dec->msg));
819 dec->msg->size = sizeof(*dec->msg);
820 dec->msg->msg_type = RUVD_MSG_DESTROY;
821 dec->msg->stream_handle = dec->stream_handle;
822 send_msg_buf(dec);
823
824 flush(dec);
825
826 dec->ws->cs_destroy(dec->cs);
827
828 for (i = 0; i < NUM_BUFFERS; ++i) {
829 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
830 rvid_destroy_buffer(&dec->bs_buffers[i]);
831 }
832
833 rvid_destroy_buffer(&dec->dpb);
834 if (u_reduce_video_profile(dec->base.profile) == PIPE_VIDEO_FORMAT_HEVC)
835 rvid_destroy_buffer(&dec->ctx);
836
837 FREE(dec);
838 }
839
840 /**
841 * start decoding of a new frame
842 */
843 static void ruvd_begin_frame(struct pipe_video_codec *decoder,
844 struct pipe_video_buffer *target,
845 struct pipe_picture_desc *picture)
846 {
847 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
848 uintptr_t frame;
849
850 assert(decoder);
851
852 frame = ++dec->frame_number;
853 vl_video_buffer_set_associated_data(target, decoder, (void *)frame,
854 &ruvd_destroy_associated_data);
855
856 dec->bs_size = 0;
857 dec->bs_ptr = dec->ws->buffer_map(
858 dec->bs_buffers[dec->cur_buffer].res->buf,
859 dec->cs, PIPE_TRANSFER_WRITE);
860 }
861
862 /**
863 * decode a macroblock
864 */
865 static void ruvd_decode_macroblock(struct pipe_video_codec *decoder,
866 struct pipe_video_buffer *target,
867 struct pipe_picture_desc *picture,
868 const struct pipe_macroblock *macroblocks,
869 unsigned num_macroblocks)
870 {
871 /* not supported (yet) */
872 assert(0);
873 }
874
875 /**
876 * decode a bitstream
877 */
878 static void ruvd_decode_bitstream(struct pipe_video_codec *decoder,
879 struct pipe_video_buffer *target,
880 struct pipe_picture_desc *picture,
881 unsigned num_buffers,
882 const void * const *buffers,
883 const unsigned *sizes)
884 {
885 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
886 unsigned i;
887
888 assert(decoder);
889
890 if (!dec->bs_ptr)
891 return;
892
893 for (i = 0; i < num_buffers; ++i) {
894 struct rvid_buffer *buf = &dec->bs_buffers[dec->cur_buffer];
895 unsigned new_size = dec->bs_size + sizes[i];
896
897 if (new_size > buf->res->buf->size) {
898 dec->ws->buffer_unmap(buf->res->buf);
899 if (!rvid_resize_buffer(dec->screen, dec->cs, buf, new_size)) {
900 RVID_ERR("Can't resize bitstream buffer!");
901 return;
902 }
903
904 dec->bs_ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
905 PIPE_TRANSFER_WRITE);
906 if (!dec->bs_ptr)
907 return;
908
909 dec->bs_ptr += dec->bs_size;
910 }
911
912 memcpy(dec->bs_ptr, buffers[i], sizes[i]);
913 dec->bs_size += sizes[i];
914 dec->bs_ptr += sizes[i];
915 }
916 }
917
918 /**
919 * end decoding of the current frame
920 */
921 static void ruvd_end_frame(struct pipe_video_codec *decoder,
922 struct pipe_video_buffer *target,
923 struct pipe_picture_desc *picture)
924 {
925 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
926 struct pb_buffer *dt;
927 struct rvid_buffer *msg_fb_it_buf, *bs_buf;
928 unsigned bs_size;
929
930 assert(decoder);
931
932 if (!dec->bs_ptr)
933 return;
934
935 msg_fb_it_buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
936 bs_buf = &dec->bs_buffers[dec->cur_buffer];
937
938 bs_size = align(dec->bs_size, 128);
939 memset(dec->bs_ptr, 0, bs_size - dec->bs_size);
940 dec->ws->buffer_unmap(bs_buf->res->buf);
941
942 map_msg_fb_it_buf(dec);
943 dec->msg->size = sizeof(*dec->msg);
944 dec->msg->msg_type = RUVD_MSG_DECODE;
945 dec->msg->stream_handle = dec->stream_handle;
946 dec->msg->status_report_feedback_number = dec->frame_number;
947
948 dec->msg->body.decode.stream_type = dec->stream_type;
949 dec->msg->body.decode.decode_flags = 0x1;
950 dec->msg->body.decode.width_in_samples = dec->base.width;
951 dec->msg->body.decode.height_in_samples = dec->base.height;
952
953 if ((picture->profile == PIPE_VIDEO_PROFILE_VC1_SIMPLE) ||
954 (picture->profile == PIPE_VIDEO_PROFILE_VC1_MAIN)) {
955 dec->msg->body.decode.width_in_samples = align(dec->msg->body.decode.width_in_samples, 16) / 16;
956 dec->msg->body.decode.height_in_samples = align(dec->msg->body.decode.height_in_samples, 16) / 16;
957 }
958
959 dec->msg->body.decode.dpb_size = dec->dpb.res->buf->size;
960 dec->msg->body.decode.bsd_size = bs_size;
961 dec->msg->body.decode.db_pitch = dec->base.width;
962
963 dt = dec->set_dtb(dec->msg, (struct vl_video_buffer *)target);
964 if (((struct r600_common_screen*)dec->screen)->family >= CHIP_STONEY)
965 dec->msg->body.decode.dt_wa_chroma_top_offset = dec->msg->body.decode.dt_pitch / 2;
966
967 switch (u_reduce_video_profile(picture->profile)) {
968 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
969 dec->msg->body.decode.codec.h264 = get_h264_msg(dec, (struct pipe_h264_picture_desc*)picture);
970 break;
971
972 case PIPE_VIDEO_FORMAT_HEVC:
973 dec->msg->body.decode.codec.h265 = get_h265_msg(dec, target, (struct pipe_h265_picture_desc*)picture);
974 break;
975
976 case PIPE_VIDEO_FORMAT_VC1:
977 dec->msg->body.decode.codec.vc1 = get_vc1_msg((struct pipe_vc1_picture_desc*)picture);
978 break;
979
980 case PIPE_VIDEO_FORMAT_MPEG12:
981 dec->msg->body.decode.codec.mpeg2 = get_mpeg2_msg(dec, (struct pipe_mpeg12_picture_desc*)picture);
982 break;
983
984 case PIPE_VIDEO_FORMAT_MPEG4:
985 dec->msg->body.decode.codec.mpeg4 = get_mpeg4_msg(dec, (struct pipe_mpeg4_picture_desc*)picture);
986 break;
987
988 default:
989 assert(0);
990 return;
991 }
992
993 dec->msg->body.decode.db_surf_tile_config = dec->msg->body.decode.dt_surf_tile_config;
994 dec->msg->body.decode.extension_support = 0x1;
995
996 /* set at least the feedback buffer size */
997 dec->fb[0] = FB_BUFFER_SIZE;
998
999 send_msg_buf(dec);
1000
1001 send_cmd(dec, RUVD_CMD_DPB_BUFFER, dec->dpb.res->buf, 0,
1002 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1003 if (u_reduce_video_profile(picture->profile) == PIPE_VIDEO_FORMAT_HEVC) {
1004 send_cmd(dec, RUVD_CMD_CONTEXT_BUFFER, dec->ctx.res->buf, 0,
1005 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1006 }
1007 send_cmd(dec, RUVD_CMD_BITSTREAM_BUFFER, bs_buf->res->buf,
1008 0, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1009 send_cmd(dec, RUVD_CMD_DECODING_TARGET_BUFFER, dt, 0,
1010 RADEON_USAGE_WRITE, RADEON_DOMAIN_VRAM);
1011 send_cmd(dec, RUVD_CMD_FEEDBACK_BUFFER, msg_fb_it_buf->res->buf,
1012 FB_BUFFER_OFFSET, RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT);
1013 if (have_it(dec))
1014 send_cmd(dec, RUVD_CMD_ITSCALING_TABLE_BUFFER, msg_fb_it_buf->res->buf,
1015 FB_BUFFER_OFFSET + FB_BUFFER_SIZE, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1016 set_reg(dec, RUVD_ENGINE_CNTL, 1);
1017
1018 flush(dec);
1019 next_buffer(dec);
1020 }
1021
1022 /**
1023 * flush any outstanding command buffers to the hardware
1024 */
1025 static void ruvd_flush(struct pipe_video_codec *decoder)
1026 {
1027 }
1028
1029 /**
1030 * create and UVD decoder
1031 */
1032 struct pipe_video_codec *ruvd_create_decoder(struct pipe_context *context,
1033 const struct pipe_video_codec *templ,
1034 ruvd_set_dtb set_dtb)
1035 {
1036 struct radeon_winsys* ws = ((struct r600_common_context *)context)->ws;
1037 struct r600_common_context *rctx = (struct r600_common_context*)context;
1038 unsigned dpb_size;
1039 unsigned width = templ->width, height = templ->height;
1040 unsigned bs_buf_size;
1041 struct radeon_info info;
1042 struct ruvd_decoder *dec;
1043 int i;
1044
1045 ws->query_info(ws, &info);
1046
1047 switch(u_reduce_video_profile(templ->profile)) {
1048 case PIPE_VIDEO_FORMAT_MPEG12:
1049 if (templ->entrypoint > PIPE_VIDEO_ENTRYPOINT_BITSTREAM || info.family < CHIP_PALM)
1050 return vl_create_mpeg12_decoder(context, templ);
1051
1052 /* fall through */
1053 case PIPE_VIDEO_FORMAT_MPEG4:
1054 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
1055 width = align(width, VL_MACROBLOCK_WIDTH);
1056 height = align(height, VL_MACROBLOCK_HEIGHT);
1057 break;
1058
1059 default:
1060 break;
1061 }
1062
1063
1064 dec = CALLOC_STRUCT(ruvd_decoder);
1065
1066 if (!dec)
1067 return NULL;
1068
1069 if (info.drm_major < 3)
1070 dec->use_legacy = TRUE;
1071
1072 dec->base = *templ;
1073 dec->base.context = context;
1074 dec->base.width = width;
1075 dec->base.height = height;
1076
1077 dec->base.destroy = ruvd_destroy;
1078 dec->base.begin_frame = ruvd_begin_frame;
1079 dec->base.decode_macroblock = ruvd_decode_macroblock;
1080 dec->base.decode_bitstream = ruvd_decode_bitstream;
1081 dec->base.end_frame = ruvd_end_frame;
1082 dec->base.flush = ruvd_flush;
1083
1084 dec->stream_type = profile2stream_type(dec, info.family);
1085 dec->set_dtb = set_dtb;
1086 dec->stream_handle = rvid_alloc_stream_handle();
1087 dec->screen = context->screen;
1088 dec->ws = ws;
1089 dec->cs = ws->cs_create(rctx->ctx, RING_UVD, NULL, NULL, NULL);
1090 if (!dec->cs) {
1091 RVID_ERR("Can't get command submission context.\n");
1092 goto error;
1093 }
1094
1095 bs_buf_size = width * height * 512 / (16 * 16);
1096 for (i = 0; i < NUM_BUFFERS; ++i) {
1097 unsigned msg_fb_it_size = FB_BUFFER_OFFSET + FB_BUFFER_SIZE;
1098 STATIC_ASSERT(sizeof(struct ruvd_msg) <= FB_BUFFER_OFFSET);
1099 if (have_it(dec))
1100 msg_fb_it_size += IT_SCALING_TABLE_SIZE;
1101 if (!rvid_create_buffer(dec->screen, &dec->msg_fb_it_buffers[i],
1102 msg_fb_it_size, PIPE_USAGE_STAGING)) {
1103 RVID_ERR("Can't allocated message buffers.\n");
1104 goto error;
1105 }
1106
1107 if (!rvid_create_buffer(dec->screen, &dec->bs_buffers[i],
1108 bs_buf_size, PIPE_USAGE_STAGING)) {
1109 RVID_ERR("Can't allocated bitstream buffers.\n");
1110 goto error;
1111 }
1112
1113 rvid_clear_buffer(context, &dec->msg_fb_it_buffers[i]);
1114 rvid_clear_buffer(context, &dec->bs_buffers[i]);
1115 }
1116
1117 dpb_size = calc_dpb_size(dec);
1118
1119 if (!rvid_create_buffer(dec->screen, &dec->dpb, dpb_size, PIPE_USAGE_DEFAULT)) {
1120 RVID_ERR("Can't allocated dpb.\n");
1121 goto error;
1122 }
1123
1124 rvid_clear_buffer(context, &dec->dpb);
1125
1126 if (u_reduce_video_profile(dec->base.profile) == PIPE_VIDEO_FORMAT_HEVC) {
1127 unsigned ctx_size = calc_ctx_size(dec);
1128 if (!rvid_create_buffer(dec->screen, &dec->ctx, ctx_size, PIPE_USAGE_DEFAULT)) {
1129 RVID_ERR("Can't allocated context buffer.\n");
1130 goto error;
1131 }
1132 rvid_clear_buffer(context, &dec->ctx);
1133 }
1134
1135 map_msg_fb_it_buf(dec);
1136 dec->msg->size = sizeof(*dec->msg);
1137 dec->msg->msg_type = RUVD_MSG_CREATE;
1138 dec->msg->stream_handle = dec->stream_handle;
1139 dec->msg->body.create.stream_type = dec->stream_type;
1140 dec->msg->body.create.width_in_samples = dec->base.width;
1141 dec->msg->body.create.height_in_samples = dec->base.height;
1142 dec->msg->body.create.dpb_size = dpb_size;
1143 send_msg_buf(dec);
1144 flush(dec);
1145 next_buffer(dec);
1146
1147 return &dec->base;
1148
1149 error:
1150 if (dec->cs) dec->ws->cs_destroy(dec->cs);
1151
1152 for (i = 0; i < NUM_BUFFERS; ++i) {
1153 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
1154 rvid_destroy_buffer(&dec->bs_buffers[i]);
1155 }
1156
1157 rvid_destroy_buffer(&dec->dpb);
1158 if (u_reduce_video_profile(dec->base.profile) == PIPE_VIDEO_FORMAT_HEVC)
1159 rvid_destroy_buffer(&dec->ctx);
1160
1161 FREE(dec);
1162
1163 return NULL;
1164 }
1165
1166 /* calculate top/bottom offset */
1167 static unsigned texture_offset(struct radeon_surf *surface, unsigned layer)
1168 {
1169 return surface->level[0].offset +
1170 layer * surface->level[0].slice_size;
1171 }
1172
1173 /* hw encode the aspect of macro tiles */
1174 static unsigned macro_tile_aspect(unsigned macro_tile_aspect)
1175 {
1176 switch (macro_tile_aspect) {
1177 default:
1178 case 1: macro_tile_aspect = 0; break;
1179 case 2: macro_tile_aspect = 1; break;
1180 case 4: macro_tile_aspect = 2; break;
1181 case 8: macro_tile_aspect = 3; break;
1182 }
1183 return macro_tile_aspect;
1184 }
1185
1186 /* hw encode the bank width and height */
1187 static unsigned bank_wh(unsigned bankwh)
1188 {
1189 switch (bankwh) {
1190 default:
1191 case 1: bankwh = 0; break;
1192 case 2: bankwh = 1; break;
1193 case 4: bankwh = 2; break;
1194 case 8: bankwh = 3; break;
1195 }
1196 return bankwh;
1197 }
1198
1199 /**
1200 * fill decoding target field from the luma and chroma surfaces
1201 */
1202 void ruvd_set_dt_surfaces(struct ruvd_msg *msg, struct radeon_surf *luma,
1203 struct radeon_surf *chroma)
1204 {
1205 msg->body.decode.dt_pitch = luma->level[0].pitch_bytes;
1206 switch (luma->level[0].mode) {
1207 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1208 msg->body.decode.dt_tiling_mode = RUVD_TILE_LINEAR;
1209 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_LINEAR;
1210 break;
1211 case RADEON_SURF_MODE_1D:
1212 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8;
1213 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_1D_THIN;
1214 break;
1215 case RADEON_SURF_MODE_2D:
1216 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8;
1217 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_2D_THIN;
1218 break;
1219 default:
1220 assert(0);
1221 break;
1222 }
1223
1224 msg->body.decode.dt_luma_top_offset = texture_offset(luma, 0);
1225 msg->body.decode.dt_chroma_top_offset = texture_offset(chroma, 0);
1226 if (msg->body.decode.dt_field_mode) {
1227 msg->body.decode.dt_luma_bottom_offset = texture_offset(luma, 1);
1228 msg->body.decode.dt_chroma_bottom_offset = texture_offset(chroma, 1);
1229 } else {
1230 msg->body.decode.dt_luma_bottom_offset = msg->body.decode.dt_luma_top_offset;
1231 msg->body.decode.dt_chroma_bottom_offset = msg->body.decode.dt_chroma_top_offset;
1232 }
1233
1234 assert(luma->bankw == chroma->bankw);
1235 assert(luma->bankh == chroma->bankh);
1236 assert(luma->mtilea == chroma->mtilea);
1237
1238 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_WIDTH(bank_wh(luma->bankw));
1239 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_HEIGHT(bank_wh(luma->bankh));
1240 msg->body.decode.dt_surf_tile_config |= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma->mtilea));
1241 }