radeon_uvd.c revision 9f464c52
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 */ 65struct 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 */ 103static 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 */ 109static 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 */ 116static 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 ?*/ 139static 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 */ 146static 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_TRANSFER_WRITE | RADEON_TRANSFER_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 */ 168static 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->msg = NULL; 182 dec->fb = NULL; 183 dec->it = NULL; 184 185 186 if (dec->sessionctx.res) 187 send_cmd(dec, RUVD_CMD_SESSION_CONTEXT_BUFFER, 188 dec->sessionctx.res->buf, 0, RADEON_USAGE_READWRITE, 189 RADEON_DOMAIN_VRAM); 190 191 /* and send it to the hardware */ 192 send_cmd(dec, RUVD_CMD_MSG_BUFFER, buf->res->buf, 0, 193 RADEON_USAGE_READ, RADEON_DOMAIN_GTT); 194} 195 196/* cycle to the next set of buffers */ 197static void next_buffer(struct ruvd_decoder *dec) 198{ 199 ++dec->cur_buffer; 200 dec->cur_buffer %= NUM_BUFFERS; 201} 202 203/* convert the profile into something UVD understands */ 204static uint32_t profile2stream_type(struct ruvd_decoder *dec, unsigned family) 205{ 206 switch (u_reduce_video_profile(dec->base.profile)) { 207 case PIPE_VIDEO_FORMAT_MPEG4_AVC: 208 return RUVD_CODEC_H264; 209 210 case PIPE_VIDEO_FORMAT_VC1: 211 return RUVD_CODEC_VC1; 212 213 case PIPE_VIDEO_FORMAT_MPEG12: 214 return RUVD_CODEC_MPEG2; 215 216 case PIPE_VIDEO_FORMAT_MPEG4: 217 return RUVD_CODEC_MPEG4; 218 219 case PIPE_VIDEO_FORMAT_HEVC: 220 return RUVD_CODEC_H265; 221 222 case PIPE_VIDEO_FORMAT_JPEG: 223 return RUVD_CODEC_MJPEG; 224 225 default: 226 assert(0); 227 return 0; 228 } 229} 230 231static unsigned calc_ctx_size_h265_main(struct ruvd_decoder *dec) 232{ 233 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH); 234 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT); 235 236 unsigned max_references = dec->base.max_references + 1; 237 238 if (dec->base.width * dec->base.height >= 4096*2000) 239 max_references = MAX2(max_references, 8); 240 else 241 max_references = MAX2(max_references, 17); 242 243 width = align (width, 16); 244 height = align (height, 16); 245 return ((width + 255) / 16) * ((height + 255) / 16) * 16 * max_references + 52 * 1024; 246} 247 248static unsigned calc_ctx_size_h265_main10(struct ruvd_decoder *dec, struct pipe_h265_picture_desc *pic) 249{ 250 unsigned log2_ctb_size, width_in_ctb, height_in_ctb, num_16x16_block_per_ctb; 251 unsigned context_buffer_size_per_ctb_row, cm_buffer_size, max_mb_address, db_left_tile_pxl_size; 252 unsigned db_left_tile_ctx_size = 4096 / 16 * (32 + 16 * 4); 253 254 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH); 255 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT); 256 unsigned coeff_10bit = (pic->pps->sps->bit_depth_luma_minus8 || pic->pps->sps->bit_depth_chroma_minus8) ? 2 : 1; 257 258 unsigned max_references = dec->base.max_references + 1; 259 260 if (dec->base.width * dec->base.height >= 4096*2000) 261 max_references = MAX2(max_references, 8); 262 else 263 max_references = MAX2(max_references, 17); 264 265 log2_ctb_size = pic->pps->sps->log2_min_luma_coding_block_size_minus3 + 3 + 266 pic->pps->sps->log2_diff_max_min_luma_coding_block_size; 267 268 width_in_ctb = (width + ((1 << log2_ctb_size) - 1)) >> log2_ctb_size; 269 height_in_ctb = (height + ((1 << log2_ctb_size) - 1)) >> log2_ctb_size; 270 271 num_16x16_block_per_ctb = ((1 << log2_ctb_size) >> 4) * ((1 << log2_ctb_size) >> 4); 272 context_buffer_size_per_ctb_row = align(width_in_ctb * num_16x16_block_per_ctb * 16, 256); 273 max_mb_address = (unsigned) ceil(height * 8 / 2048.0); 274 275 cm_buffer_size = max_references * context_buffer_size_per_ctb_row * height_in_ctb; 276 db_left_tile_pxl_size = coeff_10bit * (max_mb_address * 2 * 2048 + 1024); 277 278 return cm_buffer_size + db_left_tile_ctx_size + db_left_tile_pxl_size; 279} 280 281static unsigned get_db_pitch_alignment(struct ruvd_decoder *dec) 282{ 283 return 16; 284} 285 286/* calculate size of reference picture buffer */ 287static unsigned calc_dpb_size(struct ruvd_decoder *dec) 288{ 289 unsigned width_in_mb, height_in_mb, image_size, dpb_size; 290 291 // always align them to MB size for dpb calculation 292 unsigned width = align(dec->base.width, VL_MACROBLOCK_WIDTH); 293 unsigned height = align(dec->base.height, VL_MACROBLOCK_HEIGHT); 294 295 // always one more for currently decoded picture 296 unsigned max_references = dec->base.max_references + 1; 297 298 // aligned size of a single frame 299 image_size = align(width, get_db_pitch_alignment(dec)) * height; 300 image_size += image_size / 2; 301 image_size = align(image_size, 1024); 302 303 // picture width & height in 16 pixel units 304 width_in_mb = width / VL_MACROBLOCK_WIDTH; 305 height_in_mb = align(height / VL_MACROBLOCK_HEIGHT, 2); 306 307 switch (u_reduce_video_profile(dec->base.profile)) { 308 case PIPE_VIDEO_FORMAT_MPEG4_AVC: { 309 if (!dec->use_legacy) { 310 unsigned fs_in_mb = width_in_mb * height_in_mb; 311 unsigned alignment = 64, num_dpb_buffer; 312 313 if (dec->stream_type == RUVD_CODEC_H264_PERF) 314 alignment = 256; 315 switch(dec->base.level) { 316 case 30: 317 num_dpb_buffer = 8100 / fs_in_mb; 318 break; 319 case 31: 320 num_dpb_buffer = 18000 / fs_in_mb; 321 break; 322 case 32: 323 num_dpb_buffer = 20480 / fs_in_mb; 324 break; 325 case 41: 326 num_dpb_buffer = 32768 / fs_in_mb; 327 break; 328 case 42: 329 num_dpb_buffer = 34816 / fs_in_mb; 330 break; 331 case 50: 332 num_dpb_buffer = 110400 / fs_in_mb; 333 break; 334 case 51: 335 num_dpb_buffer = 184320 / fs_in_mb; 336 break; 337 default: 338 num_dpb_buffer = 184320 / fs_in_mb; 339 break; 340 } 341 num_dpb_buffer++; 342 max_references = MAX2(MIN2(NUM_H264_REFS, num_dpb_buffer), max_references); 343 dpb_size = image_size * max_references; 344 if ((dec->stream_type != RUVD_CODEC_H264_PERF)) { 345 dpb_size += max_references * align(width_in_mb * height_in_mb * 192, alignment); 346 dpb_size += align(width_in_mb * height_in_mb * 32, alignment); 347 } 348 } else { 349 // the firmware seems to allways assume a minimum of ref frames 350 max_references = MAX2(NUM_H264_REFS, max_references); 351 // reference picture buffer 352 dpb_size = image_size * max_references; 353 if ((dec->stream_type != RUVD_CODEC_H264_PERF)) { 354 // macroblock context buffer 355 dpb_size += width_in_mb * height_in_mb * max_references * 192; 356 // IT surface buffer 357 dpb_size += width_in_mb * height_in_mb * 32; 358 } 359 } 360 break; 361 } 362 363 case PIPE_VIDEO_FORMAT_HEVC: 364 if (dec->base.width * dec->base.height >= 4096*2000) 365 max_references = MAX2(max_references, 8); 366 else 367 max_references = MAX2(max_references, 17); 368 369 width = align (width, 16); 370 height = align (height, 16); 371 if (dec->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10) 372 dpb_size = align((align(width, get_db_pitch_alignment(dec)) * height * 9) / 4, 256) * max_references; 373 else 374 dpb_size = align((align(width, get_db_pitch_alignment(dec)) * height * 3) / 2, 256) * max_references; 375 break; 376 377 case PIPE_VIDEO_FORMAT_VC1: 378 // the firmware seems to allways assume a minimum of ref frames 379 max_references = MAX2(NUM_VC1_REFS, max_references); 380 381 // reference picture buffer 382 dpb_size = image_size * max_references; 383 384 // CONTEXT_BUFFER 385 dpb_size += width_in_mb * height_in_mb * 128; 386 387 // IT surface buffer 388 dpb_size += width_in_mb * 64; 389 390 // DB surface buffer 391 dpb_size += width_in_mb * 128; 392 393 // BP 394 dpb_size += align(MAX2(width_in_mb, height_in_mb) * 7 * 16, 64); 395 break; 396 397 case PIPE_VIDEO_FORMAT_MPEG12: 398 // reference picture buffer, must be big enough for all frames 399 dpb_size = image_size * NUM_MPEG2_REFS; 400 break; 401 402 case PIPE_VIDEO_FORMAT_MPEG4: 403 // reference picture buffer 404 dpb_size = image_size * max_references; 405 406 // CM 407 dpb_size += width_in_mb * height_in_mb * 64; 408 409 // IT surface buffer 410 dpb_size += align(width_in_mb * height_in_mb * 32, 64); 411 412 dpb_size = MAX2(dpb_size, 30 * 1024 * 1024); 413 break; 414 415 case PIPE_VIDEO_FORMAT_JPEG: 416 dpb_size = 0; 417 break; 418 419 default: 420 // something is missing here 421 assert(0); 422 423 // at least use a sane default value 424 dpb_size = 32 * 1024 * 1024; 425 break; 426 } 427 return dpb_size; 428} 429 430/* free associated data in the video buffer callback */ 431static void ruvd_destroy_associated_data(void *data) 432{ 433 /* NOOP, since we only use an intptr */ 434} 435 436/* get h264 specific message bits */ 437static struct ruvd_h264 get_h264_msg(struct ruvd_decoder *dec, struct pipe_h264_picture_desc *pic) 438{ 439 struct ruvd_h264 result; 440 441 memset(&result, 0, sizeof(result)); 442 switch (pic->base.profile) { 443 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE: 444 case PIPE_VIDEO_PROFILE_MPEG4_AVC_CONSTRAINED_BASELINE: 445 result.profile = RUVD_H264_PROFILE_BASELINE; 446 break; 447 448 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN: 449 result.profile = RUVD_H264_PROFILE_MAIN; 450 break; 451 452 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH: 453 result.profile = RUVD_H264_PROFILE_HIGH; 454 break; 455 456 default: 457 assert(0); 458 break; 459 } 460 461 result.level = dec->base.level; 462 463 result.sps_info_flags = 0; 464 result.sps_info_flags |= pic->pps->sps->direct_8x8_inference_flag << 0; 465 result.sps_info_flags |= pic->pps->sps->mb_adaptive_frame_field_flag << 1; 466 result.sps_info_flags |= pic->pps->sps->frame_mbs_only_flag << 2; 467 result.sps_info_flags |= pic->pps->sps->delta_pic_order_always_zero_flag << 3; 468 469 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8; 470 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8; 471 result.log2_max_frame_num_minus4 = pic->pps->sps->log2_max_frame_num_minus4; 472 result.pic_order_cnt_type = pic->pps->sps->pic_order_cnt_type; 473 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4; 474 475 switch (dec->base.chroma_format) { 476 case PIPE_VIDEO_CHROMA_FORMAT_NONE: 477 /* TODO: assert? */ 478 break; 479 case PIPE_VIDEO_CHROMA_FORMAT_400: 480 result.chroma_format = 0; 481 break; 482 case PIPE_VIDEO_CHROMA_FORMAT_420: 483 result.chroma_format = 1; 484 break; 485 case PIPE_VIDEO_CHROMA_FORMAT_422: 486 result.chroma_format = 2; 487 break; 488 case PIPE_VIDEO_CHROMA_FORMAT_444: 489 result.chroma_format = 3; 490 break; 491 } 492 493 result.pps_info_flags = 0; 494 result.pps_info_flags |= pic->pps->transform_8x8_mode_flag << 0; 495 result.pps_info_flags |= pic->pps->redundant_pic_cnt_present_flag << 1; 496 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 2; 497 result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag << 3; 498 result.pps_info_flags |= pic->pps->weighted_bipred_idc << 4; 499 result.pps_info_flags |= pic->pps->weighted_pred_flag << 6; 500 result.pps_info_flags |= pic->pps->bottom_field_pic_order_in_frame_present_flag << 7; 501 result.pps_info_flags |= pic->pps->entropy_coding_mode_flag << 8; 502 503 result.num_slice_groups_minus1 = pic->pps->num_slice_groups_minus1; 504 result.slice_group_map_type = pic->pps->slice_group_map_type; 505 result.slice_group_change_rate_minus1 = pic->pps->slice_group_change_rate_minus1; 506 result.pic_init_qp_minus26 = pic->pps->pic_init_qp_minus26; 507 result.chroma_qp_index_offset = pic->pps->chroma_qp_index_offset; 508 result.second_chroma_qp_index_offset = pic->pps->second_chroma_qp_index_offset; 509 510 memcpy(result.scaling_list_4x4, pic->pps->ScalingList4x4, 6*16); 511 memcpy(result.scaling_list_8x8, pic->pps->ScalingList8x8, 2*64); 512 513 if (dec->stream_type == RUVD_CODEC_H264_PERF) { 514 memcpy(dec->it, result.scaling_list_4x4, 6*16); 515 memcpy((dec->it + 96), result.scaling_list_8x8, 2*64); 516 } 517 518 result.num_ref_frames = pic->num_ref_frames; 519 520 result.num_ref_idx_l0_active_minus1 = pic->num_ref_idx_l0_active_minus1; 521 result.num_ref_idx_l1_active_minus1 = pic->num_ref_idx_l1_active_minus1; 522 523 result.frame_num = pic->frame_num; 524 memcpy(result.frame_num_list, pic->frame_num_list, 4*16); 525 result.curr_field_order_cnt_list[0] = pic->field_order_cnt[0]; 526 result.curr_field_order_cnt_list[1] = pic->field_order_cnt[1]; 527 memcpy(result.field_order_cnt_list, pic->field_order_cnt_list, 4*16*2); 528 529 result.decoded_pic_idx = pic->frame_num; 530 531 return result; 532} 533 534/* get h265 specific message bits */ 535static struct ruvd_h265 get_h265_msg(struct ruvd_decoder *dec, struct pipe_video_buffer *target, 536 struct pipe_h265_picture_desc *pic) 537{ 538 struct ruvd_h265 result; 539 unsigned i; 540 541 memset(&result, 0, sizeof(result)); 542 543 result.sps_info_flags = 0; 544 result.sps_info_flags |= pic->pps->sps->scaling_list_enabled_flag << 0; 545 result.sps_info_flags |= pic->pps->sps->amp_enabled_flag << 1; 546 result.sps_info_flags |= pic->pps->sps->sample_adaptive_offset_enabled_flag << 2; 547 result.sps_info_flags |= pic->pps->sps->pcm_enabled_flag << 3; 548 result.sps_info_flags |= pic->pps->sps->pcm_loop_filter_disabled_flag << 4; 549 result.sps_info_flags |= pic->pps->sps->long_term_ref_pics_present_flag << 5; 550 result.sps_info_flags |= pic->pps->sps->sps_temporal_mvp_enabled_flag << 6; 551 result.sps_info_flags |= pic->pps->sps->strong_intra_smoothing_enabled_flag << 7; 552 result.sps_info_flags |= pic->pps->sps->separate_colour_plane_flag << 8; 553 if (pic->UseRefPicList == true) 554 result.sps_info_flags |= 1 << 10; 555 556 result.chroma_format = pic->pps->sps->chroma_format_idc; 557 result.bit_depth_luma_minus8 = pic->pps->sps->bit_depth_luma_minus8; 558 result.bit_depth_chroma_minus8 = pic->pps->sps->bit_depth_chroma_minus8; 559 result.log2_max_pic_order_cnt_lsb_minus4 = pic->pps->sps->log2_max_pic_order_cnt_lsb_minus4; 560 result.sps_max_dec_pic_buffering_minus1 = pic->pps->sps->sps_max_dec_pic_buffering_minus1; 561 result.log2_min_luma_coding_block_size_minus3 = pic->pps->sps->log2_min_luma_coding_block_size_minus3; 562 result.log2_diff_max_min_luma_coding_block_size = pic->pps->sps->log2_diff_max_min_luma_coding_block_size; 563 result.log2_min_transform_block_size_minus2 = pic->pps->sps->log2_min_transform_block_size_minus2; 564 result.log2_diff_max_min_transform_block_size = pic->pps->sps->log2_diff_max_min_transform_block_size; 565 result.max_transform_hierarchy_depth_inter = pic->pps->sps->max_transform_hierarchy_depth_inter; 566 result.max_transform_hierarchy_depth_intra = pic->pps->sps->max_transform_hierarchy_depth_intra; 567 result.pcm_sample_bit_depth_luma_minus1 = pic->pps->sps->pcm_sample_bit_depth_luma_minus1; 568 result.pcm_sample_bit_depth_chroma_minus1 = pic->pps->sps->pcm_sample_bit_depth_chroma_minus1; 569 result.log2_min_pcm_luma_coding_block_size_minus3 = pic->pps->sps->log2_min_pcm_luma_coding_block_size_minus3; 570 result.log2_diff_max_min_pcm_luma_coding_block_size = pic->pps->sps->log2_diff_max_min_pcm_luma_coding_block_size; 571 result.num_short_term_ref_pic_sets = pic->pps->sps->num_short_term_ref_pic_sets; 572 573 result.pps_info_flags = 0; 574 result.pps_info_flags |= pic->pps->dependent_slice_segments_enabled_flag << 0; 575 result.pps_info_flags |= pic->pps->output_flag_present_flag << 1; 576 result.pps_info_flags |= pic->pps->sign_data_hiding_enabled_flag << 2; 577 result.pps_info_flags |= pic->pps->cabac_init_present_flag << 3; 578 result.pps_info_flags |= pic->pps->constrained_intra_pred_flag << 4; 579 result.pps_info_flags |= pic->pps->transform_skip_enabled_flag << 5; 580 result.pps_info_flags |= pic->pps->cu_qp_delta_enabled_flag << 6; 581 result.pps_info_flags |= pic->pps->pps_slice_chroma_qp_offsets_present_flag << 7; 582 result.pps_info_flags |= pic->pps->weighted_pred_flag << 8; 583 result.pps_info_flags |= pic->pps->weighted_bipred_flag << 9; 584 result.pps_info_flags |= pic->pps->transquant_bypass_enabled_flag << 10; 585 result.pps_info_flags |= pic->pps->tiles_enabled_flag << 11; 586 result.pps_info_flags |= pic->pps->entropy_coding_sync_enabled_flag << 12; 587 result.pps_info_flags |= pic->pps->uniform_spacing_flag << 13; 588 result.pps_info_flags |= pic->pps->loop_filter_across_tiles_enabled_flag << 14; 589 result.pps_info_flags |= pic->pps->pps_loop_filter_across_slices_enabled_flag << 15; 590 result.pps_info_flags |= pic->pps->deblocking_filter_override_enabled_flag << 16; 591 result.pps_info_flags |= pic->pps->pps_deblocking_filter_disabled_flag << 17; 592 result.pps_info_flags |= pic->pps->lists_modification_present_flag << 18; 593 result.pps_info_flags |= pic->pps->slice_segment_header_extension_present_flag << 19; 594 //result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag; ??? 595 596 result.num_extra_slice_header_bits = pic->pps->num_extra_slice_header_bits; 597 result.num_long_term_ref_pic_sps = pic->pps->sps->num_long_term_ref_pics_sps; 598 result.num_ref_idx_l0_default_active_minus1 = pic->pps->num_ref_idx_l0_default_active_minus1; 599 result.num_ref_idx_l1_default_active_minus1 = pic->pps->num_ref_idx_l1_default_active_minus1; 600 result.pps_cb_qp_offset = pic->pps->pps_cb_qp_offset; 601 result.pps_cr_qp_offset = pic->pps->pps_cr_qp_offset; 602 result.pps_beta_offset_div2 = pic->pps->pps_beta_offset_div2; 603 result.pps_tc_offset_div2 = pic->pps->pps_tc_offset_div2; 604 result.diff_cu_qp_delta_depth = pic->pps->diff_cu_qp_delta_depth; 605 result.num_tile_columns_minus1 = pic->pps->num_tile_columns_minus1; 606 result.num_tile_rows_minus1 = pic->pps->num_tile_rows_minus1; 607 result.log2_parallel_merge_level_minus2 = pic->pps->log2_parallel_merge_level_minus2; 608 result.init_qp_minus26 = pic->pps->init_qp_minus26; 609 610 for (i = 0; i < 19; ++i) 611 result.column_width_minus1[i] = pic->pps->column_width_minus1[i]; 612 613 for (i = 0; i < 21; ++i) 614 result.row_height_minus1[i] = pic->pps->row_height_minus1[i]; 615 616 result.num_delta_pocs_ref_rps_idx = pic->NumDeltaPocsOfRefRpsIdx; 617 result.curr_idx = pic->CurrPicOrderCntVal; 618 result.curr_poc = pic->CurrPicOrderCntVal; 619 620 vl_video_buffer_set_associated_data(target, &dec->base, 621 (void *)(uintptr_t)pic->CurrPicOrderCntVal, 622 &ruvd_destroy_associated_data); 623 624 for (i = 0; i < 16; ++i) { 625 struct pipe_video_buffer *ref = pic->ref[i]; 626 uintptr_t ref_pic = 0; 627 628 result.poc_list[i] = pic->PicOrderCntVal[i]; 629 630 if (ref) 631 ref_pic = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base); 632 else 633 ref_pic = 0x7F; 634 result.ref_pic_list[i] = ref_pic; 635 } 636 637 for (i = 0; i < 8; ++i) { 638 result.ref_pic_set_st_curr_before[i] = 0xFF; 639 result.ref_pic_set_st_curr_after[i] = 0xFF; 640 result.ref_pic_set_lt_curr[i] = 0xFF; 641 } 642 643 for (i = 0; i < pic->NumPocStCurrBefore; ++i) 644 result.ref_pic_set_st_curr_before[i] = pic->RefPicSetStCurrBefore[i]; 645 646 for (i = 0; i < pic->NumPocStCurrAfter; ++i) 647 result.ref_pic_set_st_curr_after[i] = pic->RefPicSetStCurrAfter[i]; 648 649 for (i = 0; i < pic->NumPocLtCurr; ++i) 650 result.ref_pic_set_lt_curr[i] = pic->RefPicSetLtCurr[i]; 651 652 for (i = 0; i < 6; ++i) 653 result.ucScalingListDCCoefSizeID2[i] = pic->pps->sps->ScalingListDCCoeff16x16[i]; 654 655 for (i = 0; i < 2; ++i) 656 result.ucScalingListDCCoefSizeID3[i] = pic->pps->sps->ScalingListDCCoeff32x32[i]; 657 658 memcpy(dec->it, pic->pps->sps->ScalingList4x4, 6 * 16); 659 memcpy(dec->it + 96, pic->pps->sps->ScalingList8x8, 6 * 64); 660 memcpy(dec->it + 480, pic->pps->sps->ScalingList16x16, 6 * 64); 661 memcpy(dec->it + 864, pic->pps->sps->ScalingList32x32, 2 * 64); 662 663 for (i = 0 ; i < 2 ; i++) { 664 for (int j = 0 ; j < 15 ; j++) 665 result.direct_reflist[i][j] = pic->RefPicList[i][j]; 666 } 667 668 if (pic->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10) { 669 if (target->buffer_format == PIPE_FORMAT_P016) { 670 result.p010_mode = 1; 671 result.msb_mode = 1; 672 } else { 673 result.luma_10to8 = 5; 674 result.chroma_10to8 = 5; 675 result.sclr_luma10to8 = 4; 676 result.sclr_chroma10to8 = 4; 677 } 678 } 679 680 /* TODO 681 result.highestTid; 682 result.isNonRef; 683 684 IDRPicFlag; 685 RAPPicFlag; 686 NumPocTotalCurr; 687 NumShortTermPictureSliceHeaderBits; 688 NumLongTermPictureSliceHeaderBits; 689 690 IsLongTerm[16]; 691 */ 692 693 return result; 694} 695 696/* get vc1 specific message bits */ 697static struct ruvd_vc1 get_vc1_msg(struct pipe_vc1_picture_desc *pic) 698{ 699 struct ruvd_vc1 result; 700 701 memset(&result, 0, sizeof(result)); 702 703 switch(pic->base.profile) { 704 case PIPE_VIDEO_PROFILE_VC1_SIMPLE: 705 result.profile = RUVD_VC1_PROFILE_SIMPLE; 706 result.level = 1; 707 break; 708 709 case PIPE_VIDEO_PROFILE_VC1_MAIN: 710 result.profile = RUVD_VC1_PROFILE_MAIN; 711 result.level = 2; 712 break; 713 714 case PIPE_VIDEO_PROFILE_VC1_ADVANCED: 715 result.profile = RUVD_VC1_PROFILE_ADVANCED; 716 result.level = 4; 717 break; 718 719 default: 720 assert(0); 721 } 722 723 /* fields common for all profiles */ 724 result.sps_info_flags |= pic->postprocflag << 7; 725 result.sps_info_flags |= pic->pulldown << 6; 726 result.sps_info_flags |= pic->interlace << 5; 727 result.sps_info_flags |= pic->tfcntrflag << 4; 728 result.sps_info_flags |= pic->finterpflag << 3; 729 result.sps_info_flags |= pic->psf << 1; 730 731 result.pps_info_flags |= pic->range_mapy_flag << 31; 732 result.pps_info_flags |= pic->range_mapy << 28; 733 result.pps_info_flags |= pic->range_mapuv_flag << 27; 734 result.pps_info_flags |= pic->range_mapuv << 24; 735 result.pps_info_flags |= pic->multires << 21; 736 result.pps_info_flags |= pic->maxbframes << 16; 737 result.pps_info_flags |= pic->overlap << 11; 738 result.pps_info_flags |= pic->quantizer << 9; 739 result.pps_info_flags |= pic->panscan_flag << 7; 740 result.pps_info_flags |= pic->refdist_flag << 6; 741 result.pps_info_flags |= pic->vstransform << 0; 742 743 /* some fields only apply to main/advanced profile */ 744 if (pic->base.profile != PIPE_VIDEO_PROFILE_VC1_SIMPLE) { 745 result.pps_info_flags |= pic->syncmarker << 20; 746 result.pps_info_flags |= pic->rangered << 19; 747 result.pps_info_flags |= pic->loopfilter << 5; 748 result.pps_info_flags |= pic->fastuvmc << 4; 749 result.pps_info_flags |= pic->extended_mv << 3; 750 result.pps_info_flags |= pic->extended_dmv << 8; 751 result.pps_info_flags |= pic->dquant << 1; 752 } 753 754 result.chroma_format = 1; 755 756#if 0 757//(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT) 758uint32_t slice_count 759uint8_t picture_type 760uint8_t frame_coding_mode 761uint8_t deblockEnable 762uint8_t pquant 763#endif 764 765 return result; 766} 767 768/* extract the frame number from a referenced video buffer */ 769static uint32_t get_ref_pic_idx(struct ruvd_decoder *dec, struct pipe_video_buffer *ref) 770{ 771 uint32_t min = MAX2(dec->frame_number, NUM_MPEG2_REFS) - NUM_MPEG2_REFS; 772 uint32_t max = MAX2(dec->frame_number, 1) - 1; 773 uintptr_t frame; 774 775 /* seems to be the most sane fallback */ 776 if (!ref) 777 return max; 778 779 /* get the frame number from the associated data */ 780 frame = (uintptr_t)vl_video_buffer_get_associated_data(ref, &dec->base); 781 782 /* limit the frame number to a valid range */ 783 return MAX2(MIN2(frame, max), min); 784} 785 786/* get mpeg2 specific msg bits */ 787static struct ruvd_mpeg2 get_mpeg2_msg(struct ruvd_decoder *dec, 788 struct pipe_mpeg12_picture_desc *pic) 789{ 790 const int *zscan = pic->alternate_scan ? vl_zscan_alternate : vl_zscan_normal; 791 struct ruvd_mpeg2 result; 792 unsigned i; 793 794 memset(&result, 0, sizeof(result)); 795 result.decoded_pic_idx = dec->frame_number; 796 for (i = 0; i < 2; ++i) 797 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]); 798 799 result.load_intra_quantiser_matrix = 1; 800 result.load_nonintra_quantiser_matrix = 1; 801 802 for (i = 0; i < 64; ++i) { 803 result.intra_quantiser_matrix[i] = pic->intra_matrix[zscan[i]]; 804 result.nonintra_quantiser_matrix[i] = pic->non_intra_matrix[zscan[i]]; 805 } 806 807 result.profile_and_level_indication = 0; 808 result.chroma_format = 0x1; 809 810 result.picture_coding_type = pic->picture_coding_type; 811 result.f_code[0][0] = pic->f_code[0][0] + 1; 812 result.f_code[0][1] = pic->f_code[0][1] + 1; 813 result.f_code[1][0] = pic->f_code[1][0] + 1; 814 result.f_code[1][1] = pic->f_code[1][1] + 1; 815 result.intra_dc_precision = pic->intra_dc_precision; 816 result.pic_structure = pic->picture_structure; 817 result.top_field_first = pic->top_field_first; 818 result.frame_pred_frame_dct = pic->frame_pred_frame_dct; 819 result.concealment_motion_vectors = pic->concealment_motion_vectors; 820 result.q_scale_type = pic->q_scale_type; 821 result.intra_vlc_format = pic->intra_vlc_format; 822 result.alternate_scan = pic->alternate_scan; 823 824 return result; 825} 826 827/* get mpeg4 specific msg bits */ 828static struct ruvd_mpeg4 get_mpeg4_msg(struct ruvd_decoder *dec, 829 struct pipe_mpeg4_picture_desc *pic) 830{ 831 struct ruvd_mpeg4 result; 832 unsigned i; 833 834 memset(&result, 0, sizeof(result)); 835 result.decoded_pic_idx = dec->frame_number; 836 for (i = 0; i < 2; ++i) 837 result.ref_pic_idx[i] = get_ref_pic_idx(dec, pic->ref[i]); 838 839 result.variant_type = 0; 840 result.profile_and_level_indication = 0xF0; // ASP Level0 841 842 result.video_object_layer_verid = 0x5; // advanced simple 843 result.video_object_layer_shape = 0x0; // rectangular 844 845 result.video_object_layer_width = dec->base.width; 846 result.video_object_layer_height = dec->base.height; 847 848 result.vop_time_increment_resolution = pic->vop_time_increment_resolution; 849 850 result.flags |= pic->short_video_header << 0; 851 //result.flags |= obmc_disable << 1; 852 result.flags |= pic->interlaced << 2; 853 result.flags |= 1 << 3; // load_intra_quant_mat 854 result.flags |= 1 << 4; // load_nonintra_quant_mat 855 result.flags |= pic->quarter_sample << 5; 856 result.flags |= 1 << 6; // complexity_estimation_disable 857 result.flags |= pic->resync_marker_disable << 7; 858 //result.flags |= data_partitioned << 8; 859 //result.flags |= reversible_vlc << 9; 860 result.flags |= 0 << 10; // newpred_enable 861 result.flags |= 0 << 11; // reduced_resolution_vop_enable 862 //result.flags |= scalability << 12; 863 //result.flags |= is_object_layer_identifier << 13; 864 //result.flags |= fixed_vop_rate << 14; 865 //result.flags |= newpred_segment_type << 15; 866 867 result.quant_type = pic->quant_type; 868 869 for (i = 0; i < 64; ++i) { 870 result.intra_quant_mat[i] = pic->intra_matrix[vl_zscan_normal[i]]; 871 result.nonintra_quant_mat[i] = pic->non_intra_matrix[vl_zscan_normal[i]]; 872 } 873 874 /* 875 int32_t trd [2] 876 int32_t trb [2] 877 uint8_t vop_coding_type 878 uint8_t vop_fcode_forward 879 uint8_t vop_fcode_backward 880 uint8_t rounding_control 881 uint8_t alternate_vertical_scan_flag 882 uint8_t top_field_first 883 */ 884 885 return result; 886} 887 888static void get_mjpeg_slice_header(struct ruvd_decoder *dec, struct pipe_mjpeg_picture_desc *pic) 889{ 890 int size = 0, saved_size, len_pos, i; 891 uint16_t *bs; 892 uint8_t *buf = dec->bs_ptr; 893 894 /* SOI */ 895 buf[size++] = 0xff; 896 buf[size++] = 0xd8; 897 898 /* DQT */ 899 buf[size++] = 0xff; 900 buf[size++] = 0xdb; 901 902 len_pos = size++; 903 size++; 904 905 for (i = 0; i < 4; ++i) { 906 if (pic->quantization_table.load_quantiser_table[i] == 0) 907 continue; 908 909 buf[size++] = i; 910 memcpy((buf + size), &pic->quantization_table.quantiser_table[i], 64); 911 size += 64; 912 } 913 914 bs = (uint16_t*)&buf[len_pos]; 915 *bs = util_bswap16(size - 4); 916 917 saved_size = size; 918 919 /* DHT */ 920 buf[size++] = 0xff; 921 buf[size++] = 0xc4; 922 923 len_pos = size++; 924 size++; 925 926 for (i = 0; i < 2; ++i) { 927 if (pic->huffman_table.load_huffman_table[i] == 0) 928 continue; 929 930 buf[size++] = 0x00 | i; 931 memcpy((buf + size), &pic->huffman_table.table[i].num_dc_codes, 16); 932 size += 16; 933 memcpy((buf + size), &pic->huffman_table.table[i].dc_values, 12); 934 size += 12; 935 } 936 937 for (i = 0; i < 2; ++i) { 938 if (pic->huffman_table.load_huffman_table[i] == 0) 939 continue; 940 941 buf[size++] = 0x10 | i; 942 memcpy((buf + size), &pic->huffman_table.table[i].num_ac_codes, 16); 943 size += 16; 944 memcpy((buf + size), &pic->huffman_table.table[i].ac_values, 162); 945 size += 162; 946 } 947 948 bs = (uint16_t*)&buf[len_pos]; 949 *bs = util_bswap16(size - saved_size - 2); 950 951 saved_size = size; 952 953 /* DRI */ 954 if (pic->slice_parameter.restart_interval) { 955 buf[size++] = 0xff; 956 buf[size++] = 0xdd; 957 buf[size++] = 0x00; 958 buf[size++] = 0x04; 959 bs = (uint16_t*)&buf[size++]; 960 *bs = util_bswap16(pic->slice_parameter.restart_interval); 961 saved_size = ++size; 962 } 963 964 /* SOF */ 965 buf[size++] = 0xff; 966 buf[size++] = 0xc0; 967 968 len_pos = size++; 969 size++; 970 971 buf[size++] = 0x08; 972 973 bs = (uint16_t*)&buf[size++]; 974 *bs = util_bswap16(pic->picture_parameter.picture_height); 975 size++; 976 977 bs = (uint16_t*)&buf[size++]; 978 *bs = util_bswap16(pic->picture_parameter.picture_width); 979 size++; 980 981 buf[size++] = pic->picture_parameter.num_components; 982 983 for (i = 0; i < pic->picture_parameter.num_components; ++i) { 984 buf[size++] = pic->picture_parameter.components[i].component_id; 985 buf[size++] = pic->picture_parameter.components[i].h_sampling_factor << 4 | 986 pic->picture_parameter.components[i].v_sampling_factor; 987 buf[size++] = pic->picture_parameter.components[i].quantiser_table_selector; 988 } 989 990 bs = (uint16_t*)&buf[len_pos]; 991 *bs = util_bswap16(size - saved_size - 2); 992 993 saved_size = size; 994 995 /* SOS */ 996 buf[size++] = 0xff; 997 buf[size++] = 0xda; 998 999 len_pos = size++; 1000 size++; 1001 1002 buf[size++] = pic->slice_parameter.num_components; 1003 1004 for (i = 0; i < pic->slice_parameter.num_components; ++i) { 1005 buf[size++] = pic->slice_parameter.components[i].component_selector; 1006 buf[size++] = pic->slice_parameter.components[i].dc_table_selector << 4 | 1007 pic->slice_parameter.components[i].ac_table_selector; 1008 } 1009 1010 buf[size++] = 0x00; 1011 buf[size++] = 0x3f; 1012 buf[size++] = 0x00; 1013 1014 bs = (uint16_t*)&buf[len_pos]; 1015 *bs = util_bswap16(size - saved_size - 2); 1016 1017 dec->bs_ptr += size; 1018 dec->bs_size += size; 1019} 1020 1021/** 1022 * destroy this video decoder 1023 */ 1024static void ruvd_destroy(struct pipe_video_codec *decoder) 1025{ 1026 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder; 1027 unsigned i; 1028 1029 assert(decoder); 1030 1031 map_msg_fb_it_buf(dec); 1032 dec->msg->size = sizeof(*dec->msg); 1033 dec->msg->msg_type = RUVD_MSG_DESTROY; 1034 dec->msg->stream_handle = dec->stream_handle; 1035 send_msg_buf(dec); 1036 1037 flush(dec, 0); 1038 1039 dec->ws->cs_destroy(dec->cs); 1040 1041 for (i = 0; i < NUM_BUFFERS; ++i) { 1042 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]); 1043 rvid_destroy_buffer(&dec->bs_buffers[i]); 1044 } 1045 1046 rvid_destroy_buffer(&dec->dpb); 1047 rvid_destroy_buffer(&dec->ctx); 1048 rvid_destroy_buffer(&dec->sessionctx); 1049 1050 FREE(dec); 1051} 1052 1053/** 1054 * start decoding of a new frame 1055 */ 1056static void ruvd_begin_frame(struct pipe_video_codec *decoder, 1057 struct pipe_video_buffer *target, 1058 struct pipe_picture_desc *picture) 1059{ 1060 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder; 1061 uintptr_t frame; 1062 1063 assert(decoder); 1064 1065 frame = ++dec->frame_number; 1066 vl_video_buffer_set_associated_data(target, decoder, (void *)frame, 1067 &ruvd_destroy_associated_data); 1068 1069 dec->bs_size = 0; 1070 dec->bs_ptr = dec->ws->buffer_map( 1071 dec->bs_buffers[dec->cur_buffer].res->buf, 1072 dec->cs, PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY); 1073} 1074 1075/** 1076 * decode a macroblock 1077 */ 1078static void ruvd_decode_macroblock(struct pipe_video_codec *decoder, 1079 struct pipe_video_buffer *target, 1080 struct pipe_picture_desc *picture, 1081 const struct pipe_macroblock *macroblocks, 1082 unsigned num_macroblocks) 1083{ 1084 /* not supported (yet) */ 1085 assert(0); 1086} 1087 1088/** 1089 * decode a bitstream 1090 */ 1091static void ruvd_decode_bitstream(struct pipe_video_codec *decoder, 1092 struct pipe_video_buffer *target, 1093 struct pipe_picture_desc *picture, 1094 unsigned num_buffers, 1095 const void * const *buffers, 1096 const unsigned *sizes) 1097{ 1098 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder; 1099 enum pipe_video_format format = u_reduce_video_profile(picture->profile); 1100 unsigned i; 1101 1102 assert(decoder); 1103 1104 if (!dec->bs_ptr) 1105 return; 1106 1107 if (format == PIPE_VIDEO_FORMAT_JPEG) 1108 get_mjpeg_slice_header(dec, (struct pipe_mjpeg_picture_desc*)picture); 1109 1110 for (i = 0; i < num_buffers; ++i) { 1111 struct rvid_buffer *buf = &dec->bs_buffers[dec->cur_buffer]; 1112 unsigned new_size = dec->bs_size + sizes[i]; 1113 1114 if (format == PIPE_VIDEO_FORMAT_JPEG) 1115 new_size += 2; /* save for EOI */ 1116 1117 if (new_size > buf->res->buf->size) { 1118 dec->ws->buffer_unmap(buf->res->buf); 1119 if (!rvid_resize_buffer(dec->screen, dec->cs, buf, new_size)) { 1120 RVID_ERR("Can't resize bitstream buffer!"); 1121 return; 1122 } 1123 1124 dec->bs_ptr = dec->ws->buffer_map(buf->res->buf, dec->cs, 1125 PIPE_TRANSFER_WRITE | 1126 RADEON_TRANSFER_TEMPORARY); 1127 if (!dec->bs_ptr) 1128 return; 1129 1130 dec->bs_ptr += dec->bs_size; 1131 } 1132 1133 memcpy(dec->bs_ptr, buffers[i], sizes[i]); 1134 dec->bs_size += sizes[i]; 1135 dec->bs_ptr += sizes[i]; 1136 } 1137 1138 if (format == PIPE_VIDEO_FORMAT_JPEG) { 1139 ((uint8_t *)dec->bs_ptr)[0] = 0xff; /* EOI */ 1140 ((uint8_t *)dec->bs_ptr)[1] = 0xd9; 1141 dec->bs_size += 2; 1142 dec->bs_ptr += 2; 1143 } 1144} 1145 1146/** 1147 * end decoding of the current frame 1148 */ 1149static void ruvd_end_frame(struct pipe_video_codec *decoder, 1150 struct pipe_video_buffer *target, 1151 struct pipe_picture_desc *picture) 1152{ 1153 struct ruvd_decoder *dec = (struct ruvd_decoder*)decoder; 1154 struct pb_buffer *dt; 1155 struct rvid_buffer *msg_fb_it_buf, *bs_buf; 1156 unsigned bs_size; 1157 1158 assert(decoder); 1159 1160 if (!dec->bs_ptr) 1161 return; 1162 1163 msg_fb_it_buf = &dec->msg_fb_it_buffers[dec->cur_buffer]; 1164 bs_buf = &dec->bs_buffers[dec->cur_buffer]; 1165 1166 bs_size = align(dec->bs_size, 128); 1167 memset(dec->bs_ptr, 0, bs_size - dec->bs_size); 1168 dec->ws->buffer_unmap(bs_buf->res->buf); 1169 1170 map_msg_fb_it_buf(dec); 1171 dec->msg->size = sizeof(*dec->msg); 1172 dec->msg->msg_type = RUVD_MSG_DECODE; 1173 dec->msg->stream_handle = dec->stream_handle; 1174 dec->msg->status_report_feedback_number = dec->frame_number; 1175 1176 dec->msg->body.decode.stream_type = dec->stream_type; 1177 dec->msg->body.decode.decode_flags = 0x1; 1178 dec->msg->body.decode.width_in_samples = dec->base.width; 1179 dec->msg->body.decode.height_in_samples = dec->base.height; 1180 1181 if ((picture->profile == PIPE_VIDEO_PROFILE_VC1_SIMPLE) || 1182 (picture->profile == PIPE_VIDEO_PROFILE_VC1_MAIN)) { 1183 dec->msg->body.decode.width_in_samples = align(dec->msg->body.decode.width_in_samples, 16) / 16; 1184 dec->msg->body.decode.height_in_samples = align(dec->msg->body.decode.height_in_samples, 16) / 16; 1185 } 1186 1187 if (dec->dpb.res) 1188 dec->msg->body.decode.dpb_size = dec->dpb.res->buf->size; 1189 dec->msg->body.decode.bsd_size = bs_size; 1190 dec->msg->body.decode.db_pitch = align(dec->base.width, get_db_pitch_alignment(dec)); 1191 1192 dt = dec->set_dtb(dec->msg, (struct vl_video_buffer *)target); 1193 1194 switch (u_reduce_video_profile(picture->profile)) { 1195 case PIPE_VIDEO_FORMAT_MPEG4_AVC: 1196 dec->msg->body.decode.codec.h264 = get_h264_msg(dec, (struct pipe_h264_picture_desc*)picture); 1197 break; 1198 1199 case PIPE_VIDEO_FORMAT_HEVC: 1200 dec->msg->body.decode.codec.h265 = get_h265_msg(dec, target, (struct pipe_h265_picture_desc*)picture); 1201 if (dec->ctx.res == NULL) { 1202 unsigned ctx_size; 1203 if (dec->base.profile == PIPE_VIDEO_PROFILE_HEVC_MAIN_10) 1204 ctx_size = calc_ctx_size_h265_main10(dec, (struct pipe_h265_picture_desc*)picture); 1205 else 1206 ctx_size = calc_ctx_size_h265_main(dec); 1207 if (!rvid_create_buffer(dec->screen, &dec->ctx, ctx_size, PIPE_USAGE_DEFAULT)) { 1208 RVID_ERR("Can't allocated context buffer.\n"); 1209 } 1210 rvid_clear_buffer(decoder->context, &dec->ctx); 1211 } 1212 1213 if (dec->ctx.res) 1214 dec->msg->body.decode.dpb_reserved = dec->ctx.res->buf->size; 1215 break; 1216 1217 case PIPE_VIDEO_FORMAT_VC1: 1218 dec->msg->body.decode.codec.vc1 = get_vc1_msg((struct pipe_vc1_picture_desc*)picture); 1219 break; 1220 1221 case PIPE_VIDEO_FORMAT_MPEG12: 1222 dec->msg->body.decode.codec.mpeg2 = get_mpeg2_msg(dec, (struct pipe_mpeg12_picture_desc*)picture); 1223 break; 1224 1225 case PIPE_VIDEO_FORMAT_MPEG4: 1226 dec->msg->body.decode.codec.mpeg4 = get_mpeg4_msg(dec, (struct pipe_mpeg4_picture_desc*)picture); 1227 break; 1228 1229 case PIPE_VIDEO_FORMAT_JPEG: 1230 break; 1231 1232 default: 1233 assert(0); 1234 return; 1235 } 1236 1237 dec->msg->body.decode.db_surf_tile_config = dec->msg->body.decode.dt_surf_tile_config; 1238 dec->msg->body.decode.extension_support = 0x1; 1239 1240 /* set at least the feedback buffer size */ 1241 dec->fb[0] = dec->fb_size; 1242 1243 send_msg_buf(dec); 1244 1245 if (dec->dpb.res) 1246 send_cmd(dec, RUVD_CMD_DPB_BUFFER, dec->dpb.res->buf, 0, 1247 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM); 1248 1249 if (dec->ctx.res) 1250 send_cmd(dec, RUVD_CMD_CONTEXT_BUFFER, dec->ctx.res->buf, 0, 1251 RADEON_USAGE_READWRITE, RADEON_DOMAIN_VRAM); 1252 send_cmd(dec, RUVD_CMD_BITSTREAM_BUFFER, bs_buf->res->buf, 1253 0, RADEON_USAGE_READ, RADEON_DOMAIN_GTT); 1254 send_cmd(dec, RUVD_CMD_DECODING_TARGET_BUFFER, dt, 0, 1255 RADEON_USAGE_WRITE, RADEON_DOMAIN_VRAM); 1256 send_cmd(dec, RUVD_CMD_FEEDBACK_BUFFER, msg_fb_it_buf->res->buf, 1257 FB_BUFFER_OFFSET, RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT); 1258 if (have_it(dec)) 1259 send_cmd(dec, RUVD_CMD_ITSCALING_TABLE_BUFFER, msg_fb_it_buf->res->buf, 1260 FB_BUFFER_OFFSET + dec->fb_size, RADEON_USAGE_READ, RADEON_DOMAIN_GTT); 1261 set_reg(dec, dec->reg.cntl, 1); 1262 1263 flush(dec, PIPE_FLUSH_ASYNC); 1264 next_buffer(dec); 1265} 1266 1267/** 1268 * flush any outstanding command buffers to the hardware 1269 */ 1270static void ruvd_flush(struct pipe_video_codec *decoder) 1271{ 1272} 1273 1274/** 1275 * create and UVD decoder 1276 */ 1277struct pipe_video_codec *ruvd_create_decoder(struct pipe_context *context, 1278 const struct pipe_video_codec *templ, 1279 ruvd_set_dtb set_dtb) 1280{ 1281 struct radeon_winsys* ws = ((struct r600_common_context *)context)->ws; 1282 struct r600_common_context *rctx = (struct r600_common_context*)context; 1283 unsigned dpb_size; 1284 unsigned width = templ->width, height = templ->height; 1285 unsigned bs_buf_size; 1286 struct radeon_info info; 1287 struct ruvd_decoder *dec; 1288 int r, i; 1289 1290 ws->query_info(ws, &info); 1291 1292 switch(u_reduce_video_profile(templ->profile)) { 1293 case PIPE_VIDEO_FORMAT_MPEG12: 1294 if (templ->entrypoint > PIPE_VIDEO_ENTRYPOINT_BITSTREAM || info.family < CHIP_PALM) 1295 return vl_create_mpeg12_decoder(context, templ); 1296 1297 /* fall through */ 1298 case PIPE_VIDEO_FORMAT_MPEG4: 1299 width = align(width, VL_MACROBLOCK_WIDTH); 1300 height = align(height, VL_MACROBLOCK_HEIGHT); 1301 break; 1302 case PIPE_VIDEO_FORMAT_MPEG4_AVC: 1303 width = align(width, VL_MACROBLOCK_WIDTH); 1304 height = align(height, VL_MACROBLOCK_HEIGHT); 1305 break; 1306 1307 default: 1308 break; 1309 } 1310 1311 1312 dec = CALLOC_STRUCT(ruvd_decoder); 1313 1314 if (!dec) 1315 return NULL; 1316 1317 if (info.drm_major < 3) 1318 dec->use_legacy = true; 1319 1320 dec->base = *templ; 1321 dec->base.context = context; 1322 dec->base.width = width; 1323 dec->base.height = height; 1324 1325 dec->base.destroy = ruvd_destroy; 1326 dec->base.begin_frame = ruvd_begin_frame; 1327 dec->base.decode_macroblock = ruvd_decode_macroblock; 1328 dec->base.decode_bitstream = ruvd_decode_bitstream; 1329 dec->base.end_frame = ruvd_end_frame; 1330 dec->base.flush = ruvd_flush; 1331 1332 dec->stream_type = profile2stream_type(dec, info.family); 1333 dec->set_dtb = set_dtb; 1334 dec->stream_handle = rvid_alloc_stream_handle(); 1335 dec->screen = context->screen; 1336 dec->ws = ws; 1337 dec->cs = ws->cs_create(rctx->ctx, RING_UVD, NULL, NULL, false); 1338 if (!dec->cs) { 1339 RVID_ERR("Can't get command submission context.\n"); 1340 goto error; 1341 } 1342 1343 dec->fb_size = FB_BUFFER_SIZE; 1344 bs_buf_size = width * height * (512 / (16 * 16)); 1345 for (i = 0; i < NUM_BUFFERS; ++i) { 1346 unsigned msg_fb_it_size = FB_BUFFER_OFFSET + dec->fb_size; 1347 STATIC_ASSERT(sizeof(struct ruvd_msg) <= FB_BUFFER_OFFSET); 1348 if (have_it(dec)) 1349 msg_fb_it_size += IT_SCALING_TABLE_SIZE; 1350 if (!rvid_create_buffer(dec->screen, &dec->msg_fb_it_buffers[i], 1351 msg_fb_it_size, PIPE_USAGE_STAGING)) { 1352 RVID_ERR("Can't allocated message buffers.\n"); 1353 goto error; 1354 } 1355 1356 if (!rvid_create_buffer(dec->screen, &dec->bs_buffers[i], 1357 bs_buf_size, PIPE_USAGE_STAGING)) { 1358 RVID_ERR("Can't allocated bitstream buffers.\n"); 1359 goto error; 1360 } 1361 1362 rvid_clear_buffer(context, &dec->msg_fb_it_buffers[i]); 1363 rvid_clear_buffer(context, &dec->bs_buffers[i]); 1364 } 1365 1366 dpb_size = calc_dpb_size(dec); 1367 if (dpb_size) { 1368 if (!rvid_create_buffer(dec->screen, &dec->dpb, dpb_size, PIPE_USAGE_DEFAULT)) { 1369 RVID_ERR("Can't allocated dpb.\n"); 1370 goto error; 1371 } 1372 rvid_clear_buffer(context, &dec->dpb); 1373 } 1374 1375 dec->reg.data0 = RUVD_GPCOM_VCPU_DATA0; 1376 dec->reg.data1 = RUVD_GPCOM_VCPU_DATA1; 1377 dec->reg.cmd = RUVD_GPCOM_VCPU_CMD; 1378 dec->reg.cntl = RUVD_ENGINE_CNTL; 1379 1380 map_msg_fb_it_buf(dec); 1381 dec->msg->size = sizeof(*dec->msg); 1382 dec->msg->msg_type = RUVD_MSG_CREATE; 1383 dec->msg->stream_handle = dec->stream_handle; 1384 dec->msg->body.create.stream_type = dec->stream_type; 1385 dec->msg->body.create.width_in_samples = dec->base.width; 1386 dec->msg->body.create.height_in_samples = dec->base.height; 1387 dec->msg->body.create.dpb_size = dpb_size; 1388 send_msg_buf(dec); 1389 r = flush(dec, 0); 1390 if (r) 1391 goto error; 1392 1393 next_buffer(dec); 1394 1395 return &dec->base; 1396 1397error: 1398 if (dec->cs) dec->ws->cs_destroy(dec->cs); 1399 1400 for (i = 0; i < NUM_BUFFERS; ++i) { 1401 rvid_destroy_buffer(&dec->msg_fb_it_buffers[i]); 1402 rvid_destroy_buffer(&dec->bs_buffers[i]); 1403 } 1404 1405 rvid_destroy_buffer(&dec->dpb); 1406 rvid_destroy_buffer(&dec->ctx); 1407 rvid_destroy_buffer(&dec->sessionctx); 1408 1409 FREE(dec); 1410 1411 return NULL; 1412} 1413 1414/* calculate top/bottom offset */ 1415static unsigned texture_offset(struct radeon_surf *surface, unsigned layer) 1416{ 1417 return surface->u.legacy.level[0].offset + 1418 layer * (uint64_t)surface->u.legacy.level[0].slice_size_dw * 4; 1419} 1420 1421/* hw encode the aspect of macro tiles */ 1422static unsigned macro_tile_aspect(unsigned macro_tile_aspect) 1423{ 1424 switch (macro_tile_aspect) { 1425 default: 1426 case 1: macro_tile_aspect = 0; break; 1427 case 2: macro_tile_aspect = 1; break; 1428 case 4: macro_tile_aspect = 2; break; 1429 case 8: macro_tile_aspect = 3; break; 1430 } 1431 return macro_tile_aspect; 1432} 1433 1434/* hw encode the bank width and height */ 1435static unsigned bank_wh(unsigned bankwh) 1436{ 1437 switch (bankwh) { 1438 default: 1439 case 1: bankwh = 0; break; 1440 case 2: bankwh = 1; break; 1441 case 4: bankwh = 2; break; 1442 case 8: bankwh = 3; break; 1443 } 1444 return bankwh; 1445} 1446 1447/** 1448 * fill decoding target field from the luma and chroma surfaces 1449 */ 1450void ruvd_set_dt_surfaces(struct ruvd_msg *msg, struct radeon_surf *luma, 1451 struct radeon_surf *chroma) 1452{ 1453 msg->body.decode.dt_pitch = luma->u.legacy.level[0].nblk_x * luma->blk_w; 1454 switch (luma->u.legacy.level[0].mode) { 1455 case RADEON_SURF_MODE_LINEAR_ALIGNED: 1456 msg->body.decode.dt_tiling_mode = RUVD_TILE_LINEAR; 1457 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_LINEAR; 1458 break; 1459 case RADEON_SURF_MODE_1D: 1460 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8; 1461 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_1D_THIN; 1462 break; 1463 case RADEON_SURF_MODE_2D: 1464 msg->body.decode.dt_tiling_mode = RUVD_TILE_8X8; 1465 msg->body.decode.dt_array_mode = RUVD_ARRAY_MODE_2D_THIN; 1466 break; 1467 default: 1468 assert(0); 1469 break; 1470 } 1471 1472 msg->body.decode.dt_luma_top_offset = texture_offset(luma, 0); 1473 if (chroma) 1474 msg->body.decode.dt_chroma_top_offset = texture_offset(chroma, 0); 1475 if (msg->body.decode.dt_field_mode) { 1476 msg->body.decode.dt_luma_bottom_offset = texture_offset(luma, 1); 1477 if (chroma) 1478 msg->body.decode.dt_chroma_bottom_offset = texture_offset(chroma, 1); 1479 } else { 1480 msg->body.decode.dt_luma_bottom_offset = msg->body.decode.dt_luma_top_offset; 1481 msg->body.decode.dt_chroma_bottom_offset = msg->body.decode.dt_chroma_top_offset; 1482 } 1483 1484 if (chroma) { 1485 assert(luma->u.legacy.bankw == chroma->u.legacy.bankw); 1486 assert(luma->u.legacy.bankh == chroma->u.legacy.bankh); 1487 assert(luma->u.legacy.mtilea == chroma->u.legacy.mtilea); 1488 } 1489 1490 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_WIDTH(bank_wh(luma->u.legacy.bankw)); 1491 msg->body.decode.dt_surf_tile_config |= RUVD_BANK_HEIGHT(bank_wh(luma->u.legacy.bankh)); 1492 msg->body.decode.dt_surf_tile_config |= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma->u.legacy.mtilea)); 1493} 1494