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 FB_BUFFER_SIZE_TONGA (2048 * 64)
61 #define IT_SCALING_TABLE_SIZE 992
62 #define UVD_SESSION_CONTEXT_SIZE (128 * 1024)
63
64 /* UVD decoder representation */
65 struct ruvd_decoder {
66 struct pipe_video_codec base;
67
68 ruvd_set_dtb set_dtb;
69
70 unsigned stream_handle;
71 unsigned stream_type;
72 unsigned frame_number;
73
74 struct pipe_screen *screen;
75 struct radeon_winsys* ws;
76 struct radeon_cmdbuf* cs;
77
78 unsigned cur_buffer;
79
80 struct rvid_buffer msg_fb_it_buffers[NUM_BUFFERS];
81 struct ruvd_msg *msg;
82 uint32_t *fb;
83 unsigned fb_size;
84 uint8_t *it;
85
86 struct rvid_buffer bs_buffers[NUM_BUFFERS];
87 void* bs_ptr;
88 unsigned bs_size;
89
90 struct rvid_buffer dpb;
91 bool use_legacy;
92 struct rvid_buffer ctx;
93 struct rvid_buffer sessionctx;
94 struct {
95 unsigned data0;
96 unsigned data1;
97 unsigned cmd;
98 unsigned cntl;
99 } reg;
100 };
101
102 /* flush IB to the hardware */
flush(struct ruvd_decoder * dec,unsigned flags)103 static int flush(struct ruvd_decoder *dec, unsigned flags)
104 {
105 return dec->ws->cs_flush(dec->cs, flags, NULL);
106 }
107
108 /* add a new set register command to the IB */
set_reg(struct ruvd_decoder * dec,unsigned reg,uint32_t val)109 static void set_reg(struct ruvd_decoder *dec, unsigned reg, uint32_t val)
110 {
111 radeon_emit(dec->cs, RUVD_PKT0(reg >> 2, 0));
112 radeon_emit(dec->cs, val);
113 }
114
115 /* send a command to the VCPU through the GPCOM registers */
send_cmd(struct ruvd_decoder * dec,unsigned cmd,struct pb_buffer * buf,uint32_t off,enum radeon_bo_usage usage,enum radeon_bo_domain domain)116 static void send_cmd(struct ruvd_decoder *dec, unsigned cmd,
117 struct pb_buffer* buf, uint32_t off,
118 enum radeon_bo_usage usage, enum radeon_bo_domain domain)
119 {
120 int reloc_idx;
121
122 reloc_idx = dec->ws->cs_add_buffer(dec->cs, buf, usage | RADEON_USAGE_SYNCHRONIZED,
123 domain, 0);
124 if (!dec->use_legacy) {
125 uint64_t addr;
126 addr = dec->ws->buffer_get_virtual_address(buf);
127 addr = addr + off;
128 set_reg(dec, dec->reg.data0, addr);
129 set_reg(dec, dec->reg.data1, addr >> 32);
130 } else {
131 off += dec->ws->buffer_get_reloc_offset(buf);
132 set_reg(dec, RUVD_GPCOM_VCPU_DATA0, off);
133 set_reg(dec, RUVD_GPCOM_VCPU_DATA1, reloc_idx * 4);
134 }
135 set_reg(dec, dec->reg.cmd, cmd << 1);
136 }
137
138 /* do the codec needs an IT buffer ?*/
have_it(struct ruvd_decoder * dec)139 static bool have_it(struct ruvd_decoder *dec)
140 {
141 return dec->stream_type == RUVD_CODEC_H264_PERF ||
142 dec->stream_type == RUVD_CODEC_H265;
143 }
144
145 /* map the next available message/feedback/itscaling buffer */
map_msg_fb_it_buf(struct ruvd_decoder * dec)146 static void map_msg_fb_it_buf(struct ruvd_decoder *dec)
147 {
148 struct rvid_buffer* buf;
149 uint8_t *ptr;
150
151 /* grab the current message/feedback buffer */
152 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
153
154 /* and map it for CPU access */
155 ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
156 PIPE_MAP_WRITE | RADEON_MAP_TEMPORARY);
157
158 /* calc buffer offsets */
159 dec->msg = (struct ruvd_msg *)ptr;
160 memset(dec->msg, 0, sizeof(*dec->msg));
161
162 dec->fb = (uint32_t *)(ptr + FB_BUFFER_OFFSET);
163 if (have_it(dec))
164 dec->it = (uint8_t *)(ptr + FB_BUFFER_OFFSET + dec->fb_size);
165 }
166
167 /* unmap and send a message command to the VCPU */
send_msg_buf(struct ruvd_decoder * dec)168 static void send_msg_buf(struct ruvd_decoder *dec)
169 {
170 struct rvid_buffer* buf;
171
172 /* ignore the request if message/feedback buffer isn't mapped */
173 if (!dec->msg || !dec->fb)
174 return;
175
176 /* grab the current message buffer */
177 buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
178
179 /* unmap the buffer */
180 dec->ws->buffer_unmap(buf->res->buf);
181 dec->bs_ptr = NULL;
182 dec->msg = NULL;
183 dec->fb = NULL;
184 dec->it = NULL;
185
186
187 if (dec->sessionctx.res)
188 send_cmd(dec, RUVD_CMD_SESSION_CONTEXT_BUFFER,
189 dec->sessionctx.res->buf, 0, RADEON_USAGE_READWRITE,
190 RADEON_DOMAIN_VRAM);
191
192 /* and send it to the hardware */
193 send_cmd(dec, RUVD_CMD_MSG_BUFFER, buf->res->buf, 0,
194 RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
195 }
196
197 /* cycle to the next set of buffers */
next_buffer(struct ruvd_decoder * dec)198 static void next_buffer(struct ruvd_decoder *dec)
199 {
200 ++dec->cur_buffer;
201 dec->cur_buffer %= NUM_BUFFERS;
202 }
203
204 /* convert the profile into something UVD understands */
profile2stream_type(struct ruvd_decoder * dec,unsigned family)205 static uint32_t profile2stream_type(struct ruvd_decoder *dec, unsigned family)
206 {
207 switch (u_reduce_video_profile(dec->base.profile)) {
208 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
209 return RUVD_CODEC_H264;
210
211 case PIPE_VIDEO_FORMAT_VC1:
212 return RUVD_CODEC_VC1;
213
214 case PIPE_VIDEO_FORMAT_MPEG12:
215 return RUVD_CODEC_MPEG2;
216
217 case PIPE_VIDEO_FORMAT_MPEG4:
218 return RUVD_CODEC_MPEG4;
219
220 case PIPE_VIDEO_FORMAT_JPEG:
221 return RUVD_CODEC_MJPEG;
222
223 default:
224 assert(0);
225 return 0;
226 }
227 }
228
229
get_db_pitch_alignment(struct ruvd_decoder * dec)230 static unsigned get_db_pitch_alignment(struct ruvd_decoder *dec)
231 {
232 return 16;
233 }
234
235 /* calculate size of reference picture buffer */
calc_dpb_size(struct ruvd_decoder * dec)236 static unsigned calc_dpb_size(struct ruvd_decoder *dec)
237 {
238 unsigned width_in_mb, height_in_mb, image_size, dpb_size;
239
240 // always align them to MB size for dpb calculation
241 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH);
242 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT);
243
244 // always one more for currently decoded picture
245 unsigned max_references = dec->base.max_references + 1;
246
247 // aligned size of a single frame
248 image_size = align(width, get_db_pitch_alignment(dec)) * height;
249 image_size += image_size / 2;
250 image_size = align(image_size, 1024);
251
252 // picture width & height in 16 pixel units
253 width_in_mb = width / VL_MACROBLOCK_WIDTH;
254 height_in_mb = align(height / VL_MACROBLOCK_HEIGHT, 2);
255
256 switch (u_reduce_video_profile(dec->base.profile)) {
257 case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
258 if (!dec->use_legacy) {
259 unsigned fs_in_mb = width_in_mb * height_in_mb;
260 unsigned alignment = 64, num_dpb_buffer;
261
262 if (dec->stream_type == RUVD_CODEC_H264_PERF)
263 alignment = 256;
264 switch(dec->base.level) {
265 case 30:
266 num_dpb_buffer = 8100 / fs_in_mb;
267 break;
268 case 31:
269 num_dpb_buffer = 18000 / fs_in_mb;
270 break;
271 case 32:
272 num_dpb_buffer = 20480 / fs_in_mb;
273 break;
274 case 41:
275 num_dpb_buffer = 32768 / fs_in_mb;
276 break;
277 case 42:
278 num_dpb_buffer = 34816 / fs_in_mb;
279 break;
280 case 50:
281 num_dpb_buffer = 110400 / fs_in_mb;
282 break;
283 case 51:
284 num_dpb_buffer = 184320 / fs_in_mb;
285 break;
286 default:
287 num_dpb_buffer = 184320 / fs_in_mb;
288 break;
289 }
290 num_dpb_buffer++;
291 max_references = MAX2(MIN2(NUM_H264_REFS, num_dpb_buffer), max_references);
292 dpb_size = image_size * max_references;
293 if ((dec->stream_type != RUVD_CODEC_H264_PERF)) {
294 dpb_size += max_references * align(width_in_mb * height_in_mb * 192, alignment);
295 dpb_size += align(width_in_mb * height_in_mb * 32, alignment);
296 }
297 } else {
298 // the firmware seems to allways assume a minimum of ref frames
299 max_references = MAX2(NUM_H264_REFS, max_references);
300 // reference picture buffer
301 dpb_size = image_size * max_references;
302 if ((dec->stream_type != RUVD_CODEC_H264_PERF)) {
303 // macroblock context buffer
304 dpb_size += width_in_mb * height_in_mb * max_references * 192;
305 // IT surface buffer
306 dpb_size += width_in_mb * height_in_mb * 32;
307 }
308 }
309 break;
310 }
311
312 case PIPE_VIDEO_FORMAT_VC1:
313 // the firmware seems to allways assume a minimum of ref frames
314 max_references = MAX2(NUM_VC1_REFS, max_references);
315
316 // reference picture buffer
317 dpb_size = image_size * max_references;
318
319 // CONTEXT_BUFFER
320 dpb_size += width_in_mb * height_in_mb * 128;
321
322 // IT surface buffer
323 dpb_size += width_in_mb * 64;
324
325 // DB surface buffer
326 dpb_size += width_in_mb * 128;
327
328 // BP
329 dpb_size += align(MAX2(width_in_mb, height_in_mb) * 7 * 16, 64);
330 break;
331
332 case PIPE_VIDEO_FORMAT_MPEG12:
333 // reference picture buffer, must be big enough for all frames
334 dpb_size = image_size * NUM_MPEG2_REFS;
335 break;
336
337 case PIPE_VIDEO_FORMAT_MPEG4:
338 // reference picture buffer
339 dpb_size = image_size * max_references;
340
341 // CM
342 dpb_size += width_in_mb * height_in_mb * 64;
343
344 // IT surface buffer
345 dpb_size += align(width_in_mb * height_in_mb * 32, 64);
346
347 dpb_size = MAX2(dpb_size, 30 * 1024 * 1024);
348 break;
349
350 case PIPE_VIDEO_FORMAT_JPEG:
351 dpb_size = 0;
352 break;
353
354 default:
355 // something is missing here
356 assert(0);
357
358 // at least use a sane default value
359 dpb_size = 32 * 1024 * 1024;
360 break;
361 }
362 return dpb_size;
363 }
364
365 /* free associated data in the video buffer callback */
ruvd_destroy_associated_data(void * data)366 static void ruvd_destroy_associated_data(void *data)
367 {
368 /* NOOP, since we only use an intptr */
369 }
370
371 /* get h264 specific message bits */
get_h264_msg(struct ruvd_decoder * dec,struct pipe_h264_picture_desc * pic)372 static struct ruvd_h264 get_h264_msg(struct ruvd_decoder *dec, struct pipe_h264_picture_desc *pic)
373 {
374 struct ruvd_h264 result;
375
376 memset(&result, 0, sizeof(result));
377 switch (pic->base.profile) {
378 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE:
379 case PIPE_VIDEO_PROFILE_MPEG4_AVC_CONSTRAINED_BASELINE:
380 result.profile = RUVD_H264_PROFILE_BASELINE;
381 break;
382
383 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN:
384 result.profile = RUVD_H264_PROFILE_MAIN;
385 break;
386
387 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH:
388 result.profile = RUVD_H264_PROFILE_HIGH;
389 break;
390
391 default:
392 assert(0);
393 break;
394 }
395
396 result.level = dec->base.level;
397
398 result.sps_info_flags = 0;
399 result.sps_info_flags |= pic->pps->sps->direct_8x8_inference_flag << 0;
400 result.sps_info_flags |= pic->pps->sps->mb_adaptive_frame_field_flag << 1;
401 result.sps_info_flags |= pic->pps->sps->frame_mbs_only_flag << 2;
402 result.sps_info_flags |= pic->pps->sps->delta_pic_order_always_zero_flag << 3;
403
404 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8;
405 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8;
406 result.log2_max_frame_num_minus4 = pic->pps->sps->log2_max_frame_num_minus4;
407 result.pic_order_cnt_type = pic->pps->sps->pic_order_cnt_type;
408 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4;
409
410 switch (dec->base.chroma_format) {
411 case PIPE_VIDEO_CHROMA_FORMAT_NONE:
412 /* TODO: assert? */
413 break;
414 case PIPE_VIDEO_CHROMA_FORMAT_400:
415 result.chroma_format = 0;
416 break;
417 case PIPE_VIDEO_CHROMA_FORMAT_420:
418 result.chroma_format = 1;
419 break;
420 case PIPE_VIDEO_CHROMA_FORMAT_422:
421 result.chroma_format = 2;
422 break;
423 case PIPE_VIDEO_CHROMA_FORMAT_444:
424 result.chroma_format = 3;
425 break;
426 }
427
428 result.pps_info_flags = 0;
429 result.pps_info_flags |= pic->pps->transform_8x8_mode_flag << 0;
430 result.pps_info_flags |= pic->pps->redundant_pic_cnt_present_flag << 1;
431 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 2;
432 result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag << 3;
433 result.pps_info_flags |= pic->pps->weighted_bipred_idc << 4;
434 result.pps_info_flags |= pic->pps->weighted_pred_flag << 6;
435 result.pps_info_flags |= pic->pps->bottom_field_pic_order_in_frame_present_flag << 7;
436 result.pps_info_flags |= pic->pps->entropy_coding_mode_flag << 8;
437
438 result.num_slice_groups_minus1 = pic->pps->num_slice_groups_minus1;
439 result.slice_group_map_type = pic->pps->slice_group_map_type;
440 result.slice_group_change_rate_minus1 = pic->pps->slice_group_change_rate_minus1;
441 result.pic_init_qp_minus26 = pic->pps->pic_init_qp_minus26;
442 result.chroma_qp_index_offset = pic->pps->chroma_qp_index_offset;
443 result.second_chroma_qp_index_offset = pic->pps->second_chroma_qp_index_offset;
444
445 memcpy(result.scaling_list_4x4, pic->pps->ScalingList4x4, 6*16);
446 memcpy(result.scaling_list_8x8, pic->pps->ScalingList8x8, 2*64);
447
448 if (dec->stream_type == RUVD_CODEC_H264_PERF) {
449 memcpy(dec->it, result.scaling_list_4x4, 6*16);
450 memcpy((dec->it + 96), result.scaling_list_8x8, 2*64);
451 }
452
453 result.num_ref_frames = pic->num_ref_frames;
454
455 result.num_ref_idx_l0_active_minus1 = pic->num_ref_idx_l0_active_minus1;
456 result.num_ref_idx_l1_active_minus1 = pic->num_ref_idx_l1_active_minus1;
457
458 result.frame_num = pic->frame_num;
459 memcpy(result.frame_num_list, pic->frame_num_list, 4*16);
460 result.curr_field_order_cnt_list[0] = pic->field_order_cnt[0];
461 result.curr_field_order_cnt_list[1] = pic->field_order_cnt[1];
462 memcpy(result.field_order_cnt_list, pic->field_order_cnt_list, 4*16*2);
463
464 result.decoded_pic_idx = pic->frame_num;
465
466 return result;
467 }
468
469 /* get vc1 specific message bits */
get_vc1_msg(struct pipe_vc1_picture_desc * pic)470 static struct ruvd_vc1 get_vc1_msg(struct pipe_vc1_picture_desc *pic)
471 {
472 struct ruvd_vc1 result;
473
474 memset(&result, 0, sizeof(result));
475
476 switch(pic->base.profile) {
477 case PIPE_VIDEO_PROFILE_VC1_SIMPLE:
478 result.profile = RUVD_VC1_PROFILE_SIMPLE;
479 result.level = 1;
480 break;
481
482 case PIPE_VIDEO_PROFILE_VC1_MAIN:
483 result.profile = RUVD_VC1_PROFILE_MAIN;
484 result.level = 2;
485 break;
486
487 case PIPE_VIDEO_PROFILE_VC1_ADVANCED:
488 result.profile = RUVD_VC1_PROFILE_ADVANCED;
489 result.level = 4;
490 break;
491
492 default:
493 assert(0);
494 }
495
496 /* fields common for all profiles */
497 result.sps_info_flags |= pic->postprocflag << 7;
498 result.sps_info_flags |= pic->pulldown << 6;
499 result.sps_info_flags |= pic->interlace << 5;
500 result.sps_info_flags |= pic->tfcntrflag << 4;
501 result.sps_info_flags |= pic->finterpflag << 3;
502 result.sps_info_flags |= pic->psf << 1;
503
504 result.pps_info_flags |= pic->range_mapy_flag << 31;
505 result.pps_info_flags |= pic->range_mapy << 28;
506 result.pps_info_flags |= pic->range_mapuv_flag << 27;
507 result.pps_info_flags |= pic->range_mapuv << 24;
508 result.pps_info_flags |= pic->multires << 21;
509 result.pps_info_flags |= pic->maxbframes << 16;
510 result.pps_info_flags |= pic->overlap << 11;
511 result.pps_info_flags |= pic->quantizer << 9;
512 result.pps_info_flags |= pic->panscan_flag << 7;
513 result.pps_info_flags |= pic->refdist_flag << 6;
514 result.pps_info_flags |= pic->vstransform << 0;
515
516 /* some fields only apply to main/advanced profile */
517 if (pic->base.profile != PIPE_VIDEO_PROFILE_VC1_SIMPLE) {
518 result.pps_info_flags |= pic->syncmarker << 20;
519 result.pps_info_flags |= pic->rangered << 19;
520 result.pps_info_flags |= pic->loopfilter << 5;
521 result.pps_info_flags |= pic->fastuvmc << 4;
522 result.pps_info_flags |= pic->extended_mv << 3;
523 result.pps_info_flags |= pic->extended_dmv << 8;
524 result.pps_info_flags |= pic->dquant << 1;
525 }
526
527 result.chroma_format = 1;
528
529 #if 0
530 //(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT)
531 uint32_t slice_count
532 uint8_t picture_type
533 uint8_t frame_coding_mode
534 uint8_t deblockEnable
535 uint8_t pquant
536 #endif
537
538 return result;
539 }
540
541 /* extract the frame number from a referenced video buffer */
get_ref_pic_idx(struct ruvd_decoder * dec,struct pipe_video_buffer * ref)542 static uint32_t get_ref_pic_idx(struct ruvd_decoder *dec, struct pipe_video_buffer *ref)
543 {
544 uint32_t min = MAX2(dec->frame_number, NUM_MPEG2_REFS) - NUM_MPEG2_REFS;
545 uint32_t max = MAX2(dec->frame_number, 1) - 1;
546 uintptr_t frame;
547
548 /* seems to be the most sane fallback */
549 if (!ref)
550 return max;
551
552 /* get the frame number from the associated data */
553 frame = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base);
554
555 /* limit the frame number to a valid range */
556 return MAX2(MIN2(frame, max), min);
557 }
558
559 /* get mpeg2 specific msg bits */
get_mpeg2_msg(struct ruvd_decoder * dec,struct pipe_mpeg12_picture_desc * pic)560 static struct ruvd_mpeg2 get_mpeg2_msg(struct ruvd_decoder *dec,
561 struct pipe_mpeg12_picture_desc *pic)
562 {
563 const int *zscan = pic->alternate_scan ? vl_zscan_alternate : vl_zscan_normal;
564 struct ruvd_mpeg2 result;
565 unsigned i;
566
567 memset(&result, 0, sizeof(result));
568 result.decoded_pic_idx = dec->frame_number;
569 for (i = 0; i < 2; ++i)
570 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]);
571
572 result.load_intra_quantiser_matrix = 1;
573 result.load_nonintra_quantiser_matrix = 1;
574
575 for (i = 0; i < 64; ++i) {
576 result.intra_quantiser_matrix[i] = pic->intra_matrix[zscan[i]];
577 result.nonintra_quantiser_matrix[i] = pic->non_intra_matrix[zscan[i]];
578 }
579
580 result.profile_and_level_indication = 0;
581 result.chroma_format = 0x1;
582
583 result.picture_coding_type = pic->picture_coding_type;
584 result.f_code[0][0] = pic->f_code[0][0] + 1;
585 result.f_code[0][1] = pic->f_code[0][1] + 1;
586 result.f_code[1][0] = pic->f_code[1][0] + 1;
587 result.f_code[1][1] = pic->f_code[1][1] + 1;
588 result.intra_dc_precision = pic->intra_dc_precision;
589 result.pic_structure = pic->picture_structure;
590 result.top_field_first = pic->top_field_first;
591 result.frame_pred_frame_dct = pic->frame_pred_frame_dct;
592 result.concealment_motion_vectors = pic->concealment_motion_vectors;
593 result.q_scale_type = pic->q_scale_type;
594 result.intra_vlc_format = pic->intra_vlc_format;
595 result.alternate_scan = pic->alternate_scan;
596
597 return result;
598 }
599
600 /* get mpeg4 specific msg bits */
get_mpeg4_msg(struct ruvd_decoder * dec,struct pipe_mpeg4_picture_desc * pic)601 static struct ruvd_mpeg4 get_mpeg4_msg(struct ruvd_decoder *dec,
602 struct pipe_mpeg4_picture_desc *pic)
603 {
604 struct ruvd_mpeg4 result;
605 unsigned i;
606
607 memset(&result, 0, sizeof(result));
608 result.decoded_pic_idx = dec->frame_number;
609 for (i = 0; i < 2; ++i)
610 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]);
611
612 result.variant_type = 0;
613 result.profile_and_level_indication = 0xF0; // ASP Level0
614
615 result.video_object_layer_verid = 0x5; // advanced simple
616 result.video_object_layer_shape = 0x0; // rectangular
617
618 result.video_object_layer_width = dec->base.width;
619 result.video_object_layer_height = dec->base.height;
620
621 result.vop_time_increment_resolution = pic->vop_time_increment_resolution;
622
623 result.flags |= pic->short_video_header << 0;
624 //result.flags |= obmc_disable << 1;
625 result.flags |= pic->interlaced << 2;
626 result.flags |= 1 << 3; // load_intra_quant_mat
627 result.flags |= 1 << 4; // load_nonintra_quant_mat
628 result.flags |= pic->quarter_sample << 5;
629 result.flags |= 1 << 6; // complexity_estimation_disable
630 result.flags |= pic->resync_marker_disable << 7;
631 //result.flags |= data_partitioned << 8;
632 //result.flags |= reversible_vlc << 9;
633 result.flags |= 0 << 10; // newpred_enable
634 result.flags |= 0 << 11; // reduced_resolution_vop_enable
635 //result.flags |= scalability << 12;
636 //result.flags |= is_object_layer_identifier << 13;
637 //result.flags |= fixed_vop_rate << 14;
638 //result.flags |= newpred_segment_type << 15;
639
640 result.quant_type = pic->quant_type;
641
642 for (i = 0; i < 64; ++i) {
643 result.intra_quant_mat[i] = pic->intra_matrix[vl_zscan_normal[i]];
644 result.nonintra_quant_mat[i] = pic->non_intra_matrix[vl_zscan_normal[i]];
645 }
646
647 /*
648 int32_t trd [2]
649 int32_t trb [2]
650 uint8_t vop_coding_type
651 uint8_t vop_fcode_forward
652 uint8_t vop_fcode_backward
653 uint8_t rounding_control
654 uint8_t alternate_vertical_scan_flag
655 uint8_t top_field_first
656 */
657
658 return result;
659 }
660
get_mjpeg_slice_header(struct ruvd_decoder * dec,struct pipe_mjpeg_picture_desc * pic)661 static void get_mjpeg_slice_header(struct ruvd_decoder *dec, struct pipe_mjpeg_picture_desc *pic)
662 {
663 int size = 0, saved_size, len_pos, i;
664 uint16_t *bs;
665 uint8_t *buf = dec->bs_ptr;
666
667 /* SOI */
668 buf[size++] = 0xff;
669 buf[size++] = 0xd8;
670
671 /* DQT */
672 buf[size++] = 0xff;
673 buf[size++] = 0xdb;
674
675 len_pos = size++;
676 size++;
677
678 for (i = 0; i < 4; ++i) {
679 if (pic->quantization_table.load_quantiser_table[i] == 0)
680 continue;
681
682 buf[size++] = i;
683 memcpy((buf + size), &pic->quantization_table.quantiser_table[i], 64);
684 size += 64;
685 }
686
687 bs = (uint16_t*)&buf[len_pos];
688 *bs = util_bswap16(size - 4);
689
690 saved_size = size;
691
692 /* DHT */
693 buf[size++] = 0xff;
694 buf[size++] = 0xc4;
695
696 len_pos = size++;
697 size++;
698
699 for (i = 0; i < 2; ++i) {
700 if (pic->huffman_table.load_huffman_table[i] == 0)
701 continue;
702
703 buf[size++] = 0x00 | i;
704 memcpy((buf + size), &pic->huffman_table.table[i].num_dc_codes, 16);
705 size += 16;
706 memcpy((buf + size), &pic->huffman_table.table[i].dc_values, 12);
707 size += 12;
708 }
709
710 for (i = 0; i < 2; ++i) {
711 if (pic->huffman_table.load_huffman_table[i] == 0)
712 continue;
713
714 buf[size++] = 0x10 | i;
715 memcpy((buf + size), &pic->huffman_table.table[i].num_ac_codes, 16);
716 size += 16;
717 memcpy((buf + size), &pic->huffman_table.table[i].ac_values, 162);
718 size += 162;
719 }
720
721 bs = (uint16_t*)&buf[len_pos];
722 *bs = util_bswap16(size - saved_size - 2);
723
724 saved_size = size;
725
726 /* DRI */
727 if (pic->slice_parameter.restart_interval) {
728 buf[size++] = 0xff;
729 buf[size++] = 0xdd;
730 buf[size++] = 0x00;
731 buf[size++] = 0x04;
732 bs = (uint16_t*)&buf[size++];
733 *bs = util_bswap16(pic->slice_parameter.restart_interval);
734 saved_size = ++size;
735 }
736
737 /* SOF */
738 buf[size++] = 0xff;
739 buf[size++] = 0xc0;
740
741 len_pos = size++;
742 size++;
743
744 buf[size++] = 0x08;
745
746 bs = (uint16_t*)&buf[size++];
747 *bs = util_bswap16(pic->picture_parameter.picture_height);
748 size++;
749
750 bs = (uint16_t*)&buf[size++];
751 *bs = util_bswap16(pic->picture_parameter.picture_width);
752 size++;
753
754 buf[size++] = pic->picture_parameter.num_components;
755
756 for (i = 0; i < pic->picture_parameter.num_components; ++i) {
757 buf[size++] = pic->picture_parameter.components[i].component_id;
758 buf[size++] = pic->picture_parameter.components[i].h_sampling_factor << 4 |
759 pic->picture_parameter.components[i].v_sampling_factor;
760 buf[size++] = pic->picture_parameter.components[i].quantiser_table_selector;
761 }
762
763 bs = (uint16_t*)&buf[len_pos];
764 *bs = util_bswap16(size - saved_size - 2);
765
766 saved_size = size;
767
768 /* SOS */
769 buf[size++] = 0xff;
770 buf[size++] = 0xda;
771
772 len_pos = size++;
773 size++;
774
775 buf[size++] = pic->slice_parameter.num_components;
776
777 for (i = 0; i < pic->slice_parameter.num_components; ++i) {
778 buf[size++] = pic->slice_parameter.components[i].component_selector;
779 buf[size++] = pic->slice_parameter.components[i].dc_table_selector << 4 |
780 pic->slice_parameter.components[i].ac_table_selector;
781 }
782
783 buf[size++] = 0x00;
784 buf[size++] = 0x3f;
785 buf[size++] = 0x00;
786
787 bs = (uint16_t*)&buf[len_pos];
788 *bs = util_bswap16(size - saved_size - 2);
789
790 dec->bs_ptr += size;
791 dec->bs_size += size;
792 }
793
794 /**
795 * destroy this video decoder
796 */
ruvd_destroy(struct pipe_video_codec * decoder)797 static void ruvd_destroy(struct pipe_video_codec *decoder)
798 {
799 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
800 unsigned i;
801
802 assert(decoder);
803
804 map_msg_fb_it_buf(dec);
805 dec->msg->size = sizeof(*dec->msg);
806 dec->msg->msg_type = RUVD_MSG_DESTROY;
807 dec->msg->stream_handle = dec->stream_handle;
808 send_msg_buf(dec);
809
810 flush(dec, 0);
811
812 dec->ws->cs_destroy(dec->cs);
813
814 for (i = 0; i < NUM_BUFFERS; ++i) {
815 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
816 rvid_destroy_buffer(&dec->bs_buffers[i]);
817 }
818
819 rvid_destroy_buffer(&dec->dpb);
820 rvid_destroy_buffer(&dec->ctx);
821 rvid_destroy_buffer(&dec->sessionctx);
822
823 FREE(dec);
824 }
825
826 /**
827 * start decoding of a new frame
828 */
ruvd_begin_frame(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture)829 static void ruvd_begin_frame(struct pipe_video_codec *decoder,
830 struct pipe_video_buffer *target,
831 struct pipe_picture_desc *picture)
832 {
833 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
834 uintptr_t frame;
835
836 assert(decoder);
837
838 frame = ++dec->frame_number;
839 vl_video_buffer_set_associated_data(target, decoder, (void *)frame,
840 &ruvd_destroy_associated_data);
841
842 dec->bs_size = 0;
843 dec->bs_ptr = dec->ws->buffer_map(
844 dec->bs_buffers[dec->cur_buffer].res->buf,
845 dec->cs, PIPE_MAP_WRITE | RADEON_MAP_TEMPORARY);
846 }
847
848 /**
849 * decode a macroblock
850 */
ruvd_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)851 static void ruvd_decode_macroblock(struct pipe_video_codec *decoder,
852 struct pipe_video_buffer *target,
853 struct pipe_picture_desc *picture,
854 const struct pipe_macroblock *macroblocks,
855 unsigned num_macroblocks)
856 {
857 /* not supported (yet) */
858 assert(0);
859 }
860
861 /**
862 * decode a bitstream
863 */
ruvd_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)864 static void ruvd_decode_bitstream(struct pipe_video_codec *decoder,
865 struct pipe_video_buffer *target,
866 struct pipe_picture_desc *picture,
867 unsigned num_buffers,
868 const void * const *buffers,
869 const unsigned *sizes)
870 {
871 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
872 enum pipe_video_format format = u_reduce_video_profile(picture->profile);
873 unsigned i;
874
875 assert(decoder);
876
877 if (!dec->bs_ptr)
878 return;
879
880 if (format == PIPE_VIDEO_FORMAT_JPEG)
881 get_mjpeg_slice_header(dec, (struct pipe_mjpeg_picture_desc*)picture);
882
883 for (i = 0; i < num_buffers; ++i) {
884 struct rvid_buffer *buf = &dec->bs_buffers[dec->cur_buffer];
885 unsigned new_size = dec->bs_size + sizes[i];
886
887 if (format == PIPE_VIDEO_FORMAT_JPEG)
888 new_size += 2; /* save for EOI */
889
890 if (new_size > buf->res->buf->size) {
891 dec->ws->buffer_unmap(buf->res->buf);
892 dec->bs_ptr = NULL;
893 if (!rvid_resize_buffer(dec->screen, dec->cs, buf, new_size)) {
894 RVID_ERR("Can't resize bitstream buffer!");
895 return;
896 }
897
898 dec->bs_ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
899 PIPE_MAP_WRITE |
900 RADEON_MAP_TEMPORARY);
901 if (!dec->bs_ptr)
902 return;
903
904 dec->bs_ptr += dec->bs_size;
905 }
906
907 memcpy(dec->bs_ptr, buffers[i], sizes[i]);
908 dec->bs_size += sizes[i];
909 dec->bs_ptr += sizes[i];
910 }
911
912 if (format == PIPE_VIDEO_FORMAT_JPEG) {
913 ((uint8_t *)dec->bs_ptr)[0] = 0xff; /* EOI */
914 ((uint8_t *)dec->bs_ptr)[1] = 0xd9;
915 dec->bs_size += 2;
916 dec->bs_ptr += 2;
917 }
918 }
919
920 /**
921 * end decoding of the current frame
922 */
ruvd_end_frame(struct pipe_video_codec * decoder,struct pipe_video_buffer * target,struct pipe_picture_desc * picture)923 static void ruvd_end_frame(struct pipe_video_codec *decoder,
924 struct pipe_video_buffer *target,
925 struct pipe_picture_desc *picture)
926 {
927 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder;
928 struct pb_buffer *dt;
929 struct rvid_buffer *msg_fb_it_buf, *bs_buf;
930 unsigned bs_size;
931
932 assert(decoder);
933
934 if (!dec->bs_ptr)
935 return;
936
937 msg_fb_it_buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
938 bs_buf = &dec->bs_buffers[dec->cur_buffer];
939
940 bs_size = align(dec->bs_size, 128);
941 memset(dec->bs_ptr, 0, bs_size - dec->bs_size);
942 dec->ws->buffer_unmap(bs_buf->res->buf);
943 dec->bs_ptr = NULL;
944
945 map_msg_fb_it_buf(dec);
946 dec->msg->size = sizeof(*dec->msg);
947 dec->msg->msg_type = RUVD_MSG_DECODE;
948 dec->msg->stream_handle = dec->stream_handle;
949 dec->msg->status_report_feedback_number = dec->frame_number;
950
951 dec->msg->body.decode.stream_type = dec->stream_type;
952 dec->msg->body.decode.decode_flags = 0x1;
953 dec->msg->body.decode.width_in_samples = dec->base.width;
954 dec->msg->body.decode.height_in_samples = dec->base.height;
955
956 if ((picture->profile == PIPE_VIDEO_PROFILE_VC1_SIMPLE) ||
957 (picture->profile == PIPE_VIDEO_PROFILE_VC1_MAIN)) {
958 dec->msg->body.decode.width_in_samples = align(dec->msg->body.decode.width_in_samples, 16) / 16;
959 dec->msg->body.decode.height_in_samples = align(dec->msg->body.decode.height_in_samples, 16) / 16;
960 }
961
962 if (dec->dpb.res)
963 dec->msg->body.decode.dpb_size = dec->dpb.res->buf->size;
964 dec->msg->body.decode.bsd_size = bs_size;
965 dec->msg->body.decode.db_pitch = align(dec->base.width, get_db_pitch_alignment(dec));
966
967 dt = dec->set_dtb(dec->msg, (struct vl_video_buffer *)target);
968
969 switch (u_reduce_video_profile(picture->profile)) {
970 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
971 dec->msg->body.decode.codec.h264 = get_h264_msg(dec, (struct pipe_h264_picture_desc*)picture);
972 break;
973
974 case PIPE_VIDEO_FORMAT_VC1:
975 dec->msg->body.decode.codec.vc1 = get_vc1_msg((struct pipe_vc1_picture_desc*)picture);
976 break;
977
978 case PIPE_VIDEO_FORMAT_MPEG12:
979 dec->msg->body.decode.codec.mpeg2 = get_mpeg2_msg(dec, (struct pipe_mpeg12_picture_desc*)picture);
980 break;
981
982 case PIPE_VIDEO_FORMAT_MPEG4:
983 dec->msg->body.decode.codec.mpeg4 = get_mpeg4_msg(dec, (struct pipe_mpeg4_picture_desc*)picture);
984 break;
985
986 case PIPE_VIDEO_FORMAT_JPEG:
987 break;
988
989 default:
990 assert(0);
991 return;
992 }
993
994 dec->msg->body.decode.db_surf_tile_config = dec->msg->body.decode.dt_surf_tile_config;
995 dec->msg->body.decode.extension_support = 0x1;
996
997 /* set at least the feedback buffer size */
998 dec->fb[0] = dec->fb_size;
999
1000 send_msg_buf(dec);
1001
1002 if (dec->dpb.res)
1003 send_cmd(dec, RUVD_CMD_DPB_BUFFER, dec->dpb.res->buf, 0,
1004 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1005
1006 if (dec->ctx.res)
1007 send_cmd(dec, RUVD_CMD_CONTEXT_BUFFER, dec->ctx.res->buf, 0,
1008 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM);
1009 send_cmd(dec, RUVD_CMD_BITSTREAM_BUFFER, bs_buf->res->buf,
1010 0, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1011 send_cmd(dec, RUVD_CMD_DECODING_TARGET_BUFFER, dt, 0,
1012 RADEON_USAGE_WRITE, RADEON_DOMAIN_VRAM);
1013 send_cmd(dec, RUVD_CMD_FEEDBACK_BUFFER, msg_fb_it_buf->res->buf,
1014 FB_BUFFER_OFFSET, RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT);
1015 if (have_it(dec))
1016 send_cmd(dec, RUVD_CMD_ITSCALING_TABLE_BUFFER, msg_fb_it_buf->res->buf,
1017 FB_BUFFER_OFFSET + dec->fb_size, RADEON_USAGE_READ, RADEON_DOMAIN_GTT);
1018 set_reg(dec, dec->reg.cntl, 1);
1019
1020 flush(dec, PIPE_FLUSH_ASYNC);
1021 next_buffer(dec);
1022 }
1023
1024 /**
1025 * flush any outstanding command buffers to the hardware
1026 */
ruvd_flush(struct pipe_video_codec * decoder)1027 static void ruvd_flush(struct pipe_video_codec *decoder)
1028 {
1029 }
1030
1031 /**
1032 * create and UVD decoder
1033 */
ruvd_create_decoder(struct pipe_context * context,const struct pipe_video_codec * templ,ruvd_set_dtb set_dtb)1034 struct pipe_video_codec *ruvd_create_decoder(struct pipe_context *context,
1035 const struct pipe_video_codec *templ,
1036 ruvd_set_dtb set_dtb)
1037 {
1038 struct radeon_winsys* ws = ((struct r600_common_context *)context)->ws;
1039 struct r600_common_context *rctx = (struct r600_common_context*)context;
1040 unsigned dpb_size;
1041 unsigned width = templ->width, height = templ->height;
1042 unsigned bs_buf_size;
1043 struct radeon_info info;
1044 struct ruvd_decoder *dec;
1045 int r, i;
1046
1047 ws->query_info(ws, &info);
1048
1049 switch(u_reduce_video_profile(templ->profile)) {
1050 case PIPE_VIDEO_FORMAT_MPEG12:
1051 if (templ->entrypoint > PIPE_VIDEO_ENTRYPOINT_BITSTREAM || info.family < CHIP_PALM)
1052 return vl_create_mpeg12_decoder(context, templ);
1053
1054 /* fall through */
1055 case PIPE_VIDEO_FORMAT_MPEG4:
1056 width = align(width, VL_MACROBLOCK_WIDTH);
1057 height = align(height, VL_MACROBLOCK_HEIGHT);
1058 break;
1059 case PIPE_VIDEO_FORMAT_MPEG4_AVC:
1060 width = align(width, VL_MACROBLOCK_WIDTH);
1061 height = align(height, VL_MACROBLOCK_HEIGHT);
1062 break;
1063
1064 default:
1065 break;
1066 }
1067
1068
1069 dec = CALLOC_STRUCT(ruvd_decoder);
1070
1071 if (!dec)
1072 return NULL;
1073
1074 dec->use_legacy = true;
1075
1076 dec->base = *templ;
1077 dec->base.context = context;
1078 dec->base.width = width;
1079 dec->base.height = height;
1080
1081 dec->base.destroy = ruvd_destroy;
1082 dec->base.begin_frame = ruvd_begin_frame;
1083 dec->base.decode_macroblock = ruvd_decode_macroblock;
1084 dec->base.decode_bitstream = ruvd_decode_bitstream;
1085 dec->base.end_frame = ruvd_end_frame;
1086 dec->base.flush = ruvd_flush;
1087
1088 dec->stream_type = profile2stream_type(dec, info.family);
1089 dec->set_dtb = set_dtb;
1090 dec->stream_handle = rvid_alloc_stream_handle();
1091 dec->screen = context->screen;
1092 dec->ws = ws;
1093 dec->cs = ws->cs_create(rctx->ctx, RING_UVD, NULL, NULL, false);
1094 if (!dec->cs) {
1095 RVID_ERR("Can't get command submission context.\n");
1096 goto error;
1097 }
1098
1099 dec->fb_size = FB_BUFFER_SIZE;
1100 bs_buf_size = width * height * (512 / (16 * 16));
1101 for (i = 0; i < NUM_BUFFERS; ++i) {
1102 unsigned msg_fb_it_size = FB_BUFFER_OFFSET + dec->fb_size;
1103 STATIC_ASSERT(sizeof(struct ruvd_msg) <= FB_BUFFER_OFFSET);
1104 if (have_it(dec))
1105 msg_fb_it_size += IT_SCALING_TABLE_SIZE;
1106 if (!rvid_create_buffer(dec->screen, &dec->msg_fb_it_buffers[i],
1107 msg_fb_it_size, PIPE_USAGE_STAGING)) {
1108 RVID_ERR("Can't allocated message buffers.\n");
1109 goto error;
1110 }
1111
1112 if (!rvid_create_buffer(dec->screen, &dec->bs_buffers[i],
1113 bs_buf_size, PIPE_USAGE_STAGING)) {
1114 RVID_ERR("Can't allocated bitstream buffers.\n");
1115 goto error;
1116 }
1117
1118 rvid_clear_buffer(context, &dec->msg_fb_it_buffers[i]);
1119 rvid_clear_buffer(context, &dec->bs_buffers[i]);
1120 }
1121
1122 dpb_size = calc_dpb_size(dec);
1123 if (dpb_size) {
1124 if (!rvid_create_buffer(dec->screen, &dec->dpb, dpb_size, PIPE_USAGE_DEFAULT)) {
1125 RVID_ERR("Can't allocated dpb.\n");
1126 goto error;
1127 }
1128 rvid_clear_buffer(context, &dec->dpb);
1129 }
1130
1131 dec->reg.data0 = RUVD_GPCOM_VCPU_DATA0;
1132 dec->reg.data1 = RUVD_GPCOM_VCPU_DATA1;
1133 dec->reg.cmd = RUVD_GPCOM_VCPU_CMD;
1134 dec->reg.cntl = RUVD_ENGINE_CNTL;
1135
1136 map_msg_fb_it_buf(dec);
1137 dec->msg->size = sizeof(*dec->msg);
1138 dec->msg->msg_type = RUVD_MSG_CREATE;
1139 dec->msg->stream_handle = dec->stream_handle;
1140 dec->msg->body.create.stream_type = dec->stream_type;
1141 dec->msg->body.create.width_in_samples = dec->base.width;
1142 dec->msg->body.create.height_in_samples = dec->base.height;
1143 dec->msg->body.create.dpb_size = dpb_size;
1144 send_msg_buf(dec);
1145 r = flush(dec, 0);
1146 if (r)
1147 goto error;
1148
1149 next_buffer(dec);
1150
1151 return &dec->base;
1152
1153 error:
1154 if (dec->cs) dec->ws->cs_destroy(dec->cs);
1155
1156 for (i = 0; i < NUM_BUFFERS; ++i) {
1157 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]);
1158 rvid_destroy_buffer(&dec->bs_buffers[i]);
1159 }
1160
1161 rvid_destroy_buffer(&dec->dpb);
1162 rvid_destroy_buffer(&dec->ctx);
1163 rvid_destroy_buffer(&dec->sessionctx);
1164
1165 FREE(dec);
1166
1167 return NULL;
1168 }
1169
1170 /* calculate top/bottom offset */
texture_offset(struct radeon_surf * surface,unsigned layer)1171 static unsigned texture_offset(struct radeon_surf *surface, unsigned layer)
1172 {
1173 return surface->u.legacy.level[0].offset +
1174 layer * (uint64_t)surface->u.legacy.level[0].slice_size_dw * 4;
1175 }
1176
1177 /* hw encode the aspect of macro tiles */
macro_tile_aspect(unsigned macro_tile_aspect)1178 static unsigned macro_tile_aspect(unsigned macro_tile_aspect)
1179 {
1180 switch (macro_tile_aspect) {
1181 default:
1182 case 1: macro_tile_aspect = 0; break;
1183 case 2: macro_tile_aspect = 1; break;
1184 case 4: macro_tile_aspect = 2; break;
1185 case 8: macro_tile_aspect = 3; break;
1186 }
1187 return macro_tile_aspect;
1188 }
1189
1190 /* hw encode the bank width and height */
bank_wh(unsigned bankwh)1191 static unsigned bank_wh(unsigned bankwh)
1192 {
1193 switch (bankwh) {
1194 default:
1195 case 1: bankwh = 0; break;
1196 case 2: bankwh = 1; break;
1197 case 4: bankwh = 2; break;
1198 case 8: bankwh = 3; break;
1199 }
1200 return bankwh;
1201 }
1202
1203 /**
1204 * fill decoding target field from the luma and chroma surfaces
1205 */
ruvd_set_dt_surfaces(struct ruvd_msg * msg,struct radeon_surf * luma,struct radeon_surf * chroma)1206 void ruvd_set_dt_surfaces(struct ruvd_msg *msg, struct radeon_surf *luma,
1207 struct radeon_surf *chroma)
1208 {
1209 msg->body.decode.dt_pitch = luma->u.legacy.level[0].nblk_x * luma->blk_w;
1210 switch (luma->u.legacy.level[0].mode) {
1211 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1212 msg->body.decode.dt_tiling_mode = RUVD_TILE_LINEAR;
1213 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_LINEAR;
1214 break;
1215 case RADEON_SURF_MODE_1D:
1216 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8;
1217 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_1D_THIN;
1218 break;
1219 case RADEON_SURF_MODE_2D:
1220 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8;
1221 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_2D_THIN;
1222 break;
1223 default:
1224 assert(0);
1225 break;
1226 }
1227
1228 msg->body.decode.dt_luma_top_offset = texture_offset(luma, 0);
1229 if (chroma)
1230 msg->body.decode.dt_chroma_top_offset = texture_offset(chroma, 0);
1231 if (msg->body.decode.dt_field_mode) {
1232 msg->body.decode.dt_luma_bottom_offset = texture_offset(luma, 1);
1233 if (chroma)
1234 msg->body.decode.dt_chroma_bottom_offset = texture_offset(chroma, 1);
1235 } else {
1236 msg->body.decode.dt_luma_bottom_offset = msg->body.decode.dt_luma_top_offset;
1237 msg->body.decode.dt_chroma_bottom_offset = msg->body.decode.dt_chroma_top_offset;
1238 }
1239
1240 if (chroma) {
1241 assert(luma->u.legacy.bankw == chroma->u.legacy.bankw);
1242 assert(luma->u.legacy.bankh == chroma->u.legacy.bankh);
1243 assert(luma->u.legacy.mtilea == chroma->u.legacy.mtilea);
1244 }
1245
1246 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_WIDTH(bank_wh(luma->u.legacy.bankw));
1247 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_HEIGHT(bank_wh(luma->u.legacy.bankh));
1248 msg->body.decode.dt_surf_tile_config |= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma->u.legacy.mtilea));
1249 }
1250