src/gallium/drivers/radeon/radeon_vce.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2013 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 <stdio.h>
  35
  36 #include "pipe/p_video_codec.h"
  37
  38 #include "util/u_video.h"
  39 #include "util/u_memory.h"
  40
  41 #include "vl/vl_video_buffer.h"
  42
  43 #include "r600_pipe_common.h"
  44 #include "radeon_video.h"
  45 #include "radeon_vce.h"
  46
  47 #define FW_40_2_2 ((40 << 24) | (2 << 16) | (2 << 8))
  48 #define FW_50_0_1 ((50 << 24) | (0 << 16) | (1 << 8))
  49 #define FW_50_1_2 ((50 << 24) | (1 << 16) | (2 << 8))
  50 #define FW_50_10_2 ((50 << 24) | (10 << 16) | (2 << 8))
  51 #define FW_50_17_3 ((50 << 24) | (17 << 16) | (3 << 8))
  52 #define FW_52_0_3 ((52 << 24) | (0 << 16) | (3 << 8))
  53 #define FW_52_4_3 ((52 << 24) | (4 << 16) | (3 << 8))
  54
  55 /**
  56  * flush commands to the hardware
  57  */
  58 static void flush(struct rvce_encoder *enc)
  59 {
  60         enc->ws->cs_flush(enc->cs, RADEON_FLUSH_ASYNC, NULL);
  61         enc->task_info_idx = 0;
  62         enc->bs_idx = 0;
  63 }
  64
  65 #if 0
  66 static void dump_feedback(struct rvce_encoder *enc, struct rvid_buffer *fb)
  67 {
  68         uint32_t *ptr = enc->ws->buffer_map(fb->res->buf, enc->cs, PIPE_TRANSFER_READ_WRITE);
  69         unsigned i = 0;
  70         fprintf(stderr, "\n");
  71         fprintf(stderr, "encStatus:\t\t\t%08x\n", ptr[i++]);
  72         fprintf(stderr, "encHasBitstream:\t\t%08x\n", ptr[i++]);
  73         fprintf(stderr, "encHasAudioBitstream:\t\t%08x\n", ptr[i++]);
  74         fprintf(stderr, "encBitstreamOffset:\t\t%08x\n", ptr[i++]);
  75         fprintf(stderr, "encBitstreamSize:\t\t%08x\n", ptr[i++]);
  76         fprintf(stderr, "encAudioBitstreamOffset:\t%08x\n", ptr[i++]);
  77         fprintf(stderr, "encAudioBitstreamSize:\t\t%08x\n", ptr[i++]);
  78         fprintf(stderr, "encExtrabytes:\t\t\t%08x\n", ptr[i++]);
  79         fprintf(stderr, "encAudioExtrabytes:\t\t%08x\n", ptr[i++]);
  80         fprintf(stderr, "videoTimeStamp:\t\t\t%08x\n", ptr[i++]);
  81         fprintf(stderr, "audioTimeStamp:\t\t\t%08x\n", ptr[i++]);
  82         fprintf(stderr, "videoOutputType:\t\t%08x\n", ptr[i++]);
  83         fprintf(stderr, "attributeFlags:\t\t\t%08x\n", ptr[i++]);
  84         fprintf(stderr, "seiPrivatePackageOffset:\t%08x\n", ptr[i++]);
  85         fprintf(stderr, "seiPrivatePackageSize:\t\t%08x\n", ptr[i++]);
  86         fprintf(stderr, "\n");
  87         enc->ws->buffer_unmap(fb->res->buf);
  88 }
  89 #endif
  90
  91 /**
  92  * reset the CPB handling
  93  */
  94 static void reset_cpb(struct rvce_encoder *enc)
  95 {
  96         unsigned i;
  97
  98         LIST_INITHEAD(&enc->cpb_slots);
  99         for (i = 0; i < enc->cpb_num; ++i) {
 100                 struct rvce_cpb_slot *slot = &enc->cpb_array[i];
 101                 slot->index = i;
 102                 slot->picture_type = PIPE_H264_ENC_PICTURE_TYPE_SKIP;
 103                 slot->frame_num = 0;
 104                 slot->pic_order_cnt = 0;
 105                 LIST_ADDTAIL(&slot->list, &enc->cpb_slots);
 106         }
 107 }
 108
 109 /**
 110  * sort l0 and l1 to the top of the list
 111  */
 112 static void sort_cpb(struct rvce_encoder *enc)
 113 {
 114         struct rvce_cpb_slot *i, *l0 = NULL, *l1 = NULL;
 115
 116         LIST_FOR_EACH_ENTRY(i, &enc->cpb_slots, list) {
 117                 if (i->frame_num == enc->pic.ref_idx_l0)
 118                         l0 = i;
 119
 120                 if (i->frame_num == enc->pic.ref_idx_l1)
 121                         l1 = i;
 122
 123                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P && l0)
 124                         break;
 125
 126                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_B &&
 127                     l0 && l1)
 128                         break;
 129         }
 130
 131         if (l1) {
 132                 LIST_DEL(&l1->list);
 133                 LIST_ADD(&l1->list, &enc->cpb_slots);
 134         }
 135
 136         if (l0) {
 137                 LIST_DEL(&l0->list);
 138                 LIST_ADD(&l0->list, &enc->cpb_slots);
 139         }
 140 }
 141
 142 /**
 143  * get number of cpbs based on dpb
 144  */
 145 static unsigned get_cpb_num(struct rvce_encoder *enc)
 146 {
 147         unsigned w = align(enc->base.width, 16) / 16;
 148         unsigned h = align(enc->base.height, 16) / 16;
 149         unsigned dpb;
 150
 151         switch (enc->base.level) {
 152         case 10:
 153                 dpb = 396;
 154                 break;
 155         case 11:
 156                 dpb = 900;
 157                 break;
 158         case 12:
 159         case 13:
 160         case 20:
 161                 dpb = 2376;
 162                 break;
 163         case 21:
 164                 dpb = 4752;
 165                 break;
 166         case 22:
 167         case 30:
 168                 dpb = 8100;
 169                 break;
 170         case 31:
 171                 dpb = 18000;
 172                 break;
 173         case 32:
 174                 dpb = 20480;
 175                 break;
 176         case 40:
 177         case 41:
 178                 dpb = 32768;
 179                 break;
 180         default:
 181         case 42:
 182                 dpb = 34816;
 183                 break;
 184         case 50:
 185                 dpb = 110400;
 186                 break;
 187         case 51:
 188                 dpb = 184320;
 189                 break;
 190         }
 191
 192         return MIN2(dpb / (w * h), 16);
 193 }
 194
 195 /**
 196  * Get the slot for the currently encoded frame
 197  */
 198 struct rvce_cpb_slot *current_slot(struct rvce_encoder *enc)
 199 {
 200         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 201 }
 202
 203 /**
 204  * Get the slot for L0
 205  */
 206 struct rvce_cpb_slot *l0_slot(struct rvce_encoder *enc)
 207 {
 208         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next, list);
 209 }
 210
 211 /**
 212  * Get the slot for L1
 213  */
 214 struct rvce_cpb_slot *l1_slot(struct rvce_encoder *enc)
 215 {
 216         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next->next, list);
 217 }
 218
 219 /**
 220  * Calculate the offsets into the CPB
 221  */
 222 void rvce_frame_offset(struct rvce_encoder *enc, struct rvce_cpb_slot *slot,
 223                        signed *luma_offset, signed *chroma_offset)
 224 {
 225         unsigned pitch = align(enc->luma->level[0].pitch_bytes, 128);
 226         unsigned vpitch = align(enc->luma->npix_y, 16);
 227         unsigned fsize = pitch * (vpitch + vpitch / 2);
 228
 229         *luma_offset = slot->index * fsize;
 230         *chroma_offset = *luma_offset + pitch * vpitch;
 231 }
 232
 233 /**
 234  * destroy this video encoder
 235  */
 236 static void rvce_destroy(struct pipe_video_codec *encoder)
 237 {
 238         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 239         if (enc->stream_handle) {
 240                 struct rvid_buffer fb;
 241                 rvid_create_buffer(enc->screen, &fb, 512, PIPE_USAGE_STAGING);
 242                 enc->fb = &fb;
 243                 enc->session(enc);
 244                 enc->feedback(enc);
 245                 enc->destroy(enc);
 246                 flush(enc);
 247                 rvid_destroy_buffer(&fb);
 248         }
 249         rvid_destroy_buffer(&enc->cpb);
 250         enc->ws->cs_destroy(enc->cs);
 251         FREE(enc->cpb_array);
 252         FREE(enc);
 253 }
 254
 255 static void rvce_begin_frame(struct pipe_video_codec *encoder,
 256                              struct pipe_video_buffer *source,
 257                              struct pipe_picture_desc *picture)
 258 {
 259         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 260         struct vl_video_buffer *vid_buf = (struct vl_video_buffer *)source;
 261         struct pipe_h264_enc_picture_desc *pic = (struct pipe_h264_enc_picture_desc *)picture;
 262
 263         bool need_rate_control =
 264                 enc->pic.rate_ctrl.rate_ctrl_method != pic->rate_ctrl.rate_ctrl_method ||
 265                 enc->pic.quant_i_frames != pic->quant_i_frames ||
 266                 enc->pic.quant_p_frames != pic->quant_p_frames ||
 267                 enc->pic.quant_b_frames != pic->quant_b_frames;
 268
 269         enc->pic = *pic;
 270         get_pic_param(enc, pic);
 271
 272         enc->get_buffer(vid_buf->resources[0], &enc->handle, &enc->luma);
 273         enc->get_buffer(vid_buf->resources[1], NULL, &enc->chroma);
 274
 275         if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR)
 276                 reset_cpb(enc);
 277         else if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_P ||
 278                  pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_B)
 279                 sort_cpb(enc);
 280
 281         if (!enc->stream_handle) {
 282                 struct rvid_buffer fb;
 283                 enc->stream_handle = rvid_alloc_stream_handle();
 284                 rvid_create_buffer(enc->screen, &fb, 512, PIPE_USAGE_STAGING);
 285                 enc->fb = &fb;
 286                 enc->session(enc);
 287                 enc->create(enc);
 288                 enc->config(enc);
 289                 enc->feedback(enc);
 290                 flush(enc);
 291                 //dump_feedback(enc, &fb);
 292                 rvid_destroy_buffer(&fb);
 293                 need_rate_control = false;
 294         }
 295
 296         if (need_rate_control) {
 297                 enc->session(enc);
 298                 enc->config(enc);
 299                 flush(enc);
 300         }
 301 }
 302
 303 static void rvce_encode_bitstream(struct pipe_video_codec *encoder,
 304                                   struct pipe_video_buffer *source,
 305                                   struct pipe_resource *destination,
 306                                   void **fb)
 307 {
 308         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 309         enc->get_buffer(destination, &enc->bs_handle, NULL);
 310         enc->bs_size = destination->width0;
 311
 312         *fb = enc->fb = CALLOC_STRUCT(rvid_buffer);
 313         if (!rvid_create_buffer(enc->screen, enc->fb, 512, PIPE_USAGE_STAGING)) {
 314                 RVID_ERR("Can't create feedback buffer.\n");
 315                 return;
 316         }
 317         if (!radeon_emitted(enc->cs, 0))
 318                 enc->session(enc);
 319         enc->encode(enc);
 320         enc->feedback(enc);
 321 }
 322
 323 static void rvce_end_frame(struct pipe_video_codec *encoder,
 324                            struct pipe_video_buffer *source,
 325                            struct pipe_picture_desc *picture)
 326 {
 327         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 328         struct rvce_cpb_slot *slot = LIST_ENTRY(
 329                 struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 330
 331         if (!enc->dual_inst || enc->bs_idx > 1)
 332                 flush(enc);
 333
 334         /* update the CPB backtrack with the just encoded frame */
 335         slot->picture_type = enc->pic.picture_type;
 336         slot->frame_num = enc->pic.frame_num;
 337         slot->pic_order_cnt = enc->pic.pic_order_cnt;
 338         if (!enc->pic.not_referenced) {
 339                 LIST_DEL(&slot->list);
 340                 LIST_ADD(&slot->list, &enc->cpb_slots);
 341         }
 342 }
 343
 344 static void rvce_get_feedback(struct pipe_video_codec *encoder,
 345                               void *feedback, unsigned *size)
 346 {
 347         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 348         struct rvid_buffer *fb = feedback;
 349
 350         if (size) {
 351                 uint32_t *ptr = enc->ws->buffer_map(fb->res->buf, enc->cs, PIPE_TRANSFER_READ_WRITE);
 352
 353                 if (ptr[1]) {
 354                         *size = ptr[4] - ptr[9];
 355                 } else {
 356                         *size = 0;
 357                 }
 358
 359                 enc->ws->buffer_unmap(fb->res->buf);
 360         }
 361         //dump_feedback(enc, fb);
 362         rvid_destroy_buffer(fb);
 363         FREE(fb);
 364 }
 365
 366 /**
 367  * flush any outstanding command buffers to the hardware
 368  */
 369 static void rvce_flush(struct pipe_video_codec *encoder)
 370 {
 371         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 372
 373         flush(enc);
 374 }
 375
 376 static void rvce_cs_flush(void *ctx, unsigned flags,
 377                           struct pipe_fence_handle **fence)
 378 {
 379         // just ignored
 380 }
 381
 382 struct pipe_video_codec *rvce_create_encoder(struct pipe_context *context,
 383                                              const struct pipe_video_codec *templ,
 384                                              struct radeon_winsys* ws,
 385                                              rvce_get_buffer get_buffer)
 386 {
 387         struct r600_common_screen *rscreen = (struct r600_common_screen *)context->screen;
 388         struct r600_common_context *rctx = (struct r600_common_context*)context;
 389         struct rvce_encoder *enc;
 390         struct pipe_video_buffer *tmp_buf, templat = {};
 391         struct radeon_surf *tmp_surf;
 392         unsigned cpb_size;
 393
 394         if (!rscreen->info.vce_fw_version) {
 395                 RVID_ERR("Kernel doesn't supports VCE!\n");
 396                 return NULL;
 397
 398         } else if (!rvce_is_fw_version_supported(rscreen)) {
 399                 RVID_ERR("Unsupported VCE fw version loaded!\n");
 400                 return NULL;
 401         }
 402
 403         enc = CALLOC_STRUCT(rvce_encoder);
 404         if (!enc)
 405                 return NULL;
 406
 407         if (rscreen->info.drm_major == 3)
 408                 enc->use_vm = true;
 409         if ((rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 42) ||
 410             rscreen->info.drm_major == 3)
 411                 enc->use_vui = true;
 412         if (rscreen->info.family >= CHIP_TONGA &&
 413             rscreen->info.family != CHIP_STONEY &&
 414             rscreen->info.family != CHIP_POLARIS11)
 415                 enc->dual_pipe = true;
 416         /* TODO enable B frame with dual instance */
 417         if ((rscreen->info.family >= CHIP_TONGA) &&
 418                 (templ->max_references == 1) &&
 419                 (rscreen->info.vce_harvest_config == 0))
 420                 enc->dual_inst = true;
 421
 422         enc->base = *templ;
 423         enc->base.context = context;
 424
 425         enc->base.destroy = rvce_destroy;
 426         enc->base.begin_frame = rvce_begin_frame;
 427         enc->base.encode_bitstream = rvce_encode_bitstream;
 428         enc->base.end_frame = rvce_end_frame;
 429         enc->base.flush = rvce_flush;
 430         enc->base.get_feedback = rvce_get_feedback;
 431         enc->get_buffer = get_buffer;
 432
 433         enc->screen = context->screen;
 434         enc->ws = ws;
 435         enc->cs = ws->cs_create(rctx->ctx, RING_VCE, rvce_cs_flush, enc);
 436         if (!enc->cs) {
 437                 RVID_ERR("Can't get command submission context.\n");
 438                 goto error;
 439         }
 440
 441         templat.buffer_format = PIPE_FORMAT_NV12;
 442         templat.chroma_format = PIPE_VIDEO_CHROMA_FORMAT_420;
 443         templat.width = enc->base.width;
 444         templat.height = enc->base.height;
 445         templat.interlaced = false;
 446         if (!(tmp_buf = context->create_video_buffer(context, &templat))) {
 447                 RVID_ERR("Can't create video buffer.\n");
 448                 goto error;
 449         }
 450
 451         enc->cpb_num = get_cpb_num(enc);
 452         if (!enc->cpb_num)
 453                 goto error;
 454
 455         get_buffer(((struct vl_video_buffer *)tmp_buf)->resources[0], NULL, &tmp_surf);
 456         cpb_size = align(tmp_surf->level[0].pitch_bytes, 128);
 457         cpb_size = cpb_size * align(tmp_surf->npix_y, 32);
 458         cpb_size = cpb_size * 3 / 2;
 459         cpb_size = cpb_size * enc->cpb_num;
 460         if (enc->dual_pipe)
 461                 cpb_size +=  RVCE_MAX_AUX_BUFFER_NUM *
 462                         RVCE_MAX_BITSTREAM_OUTPUT_ROW_SIZE * 2;
 463         tmp_buf->destroy(tmp_buf);
 464         if (!rvid_create_buffer(enc->screen, &enc->cpb, cpb_size, PIPE_USAGE_DEFAULT)) {
 465                 RVID_ERR("Can't create CPB buffer.\n");
 466                 goto error;
 467         }
 468
 469         enc->cpb_array = CALLOC(enc->cpb_num, sizeof(struct rvce_cpb_slot));
 470         if (!enc->cpb_array)
 471                 goto error;
 472
 473         reset_cpb(enc);
 474
 475         switch (rscreen->info.vce_fw_version) {
 476         case FW_40_2_2:
 477                 radeon_vce_40_2_2_init(enc);
 478                 get_pic_param = radeon_vce_40_2_2_get_param;
 479                 break;
 480
 481         case FW_50_0_1:
 482         case FW_50_1_2:
 483         case FW_50_10_2:
 484         case FW_50_17_3:
 485                 radeon_vce_50_init(enc);
 486                 get_pic_param = radeon_vce_50_get_param;
 487                 break;
 488
 489         case FW_52_0_3:
 490         case FW_52_4_3:
 491                 radeon_vce_52_init(enc);
 492                 get_pic_param = radeon_vce_52_get_param;
 493                 break;
 494
 495         default:
 496                 goto error;
 497         }
 498
 499         return &enc->base;
 500
 501 error:
 502         if (enc->cs)
 503                 enc->ws->cs_destroy(enc->cs);
 504
 505         rvid_destroy_buffer(&enc->cpb);
 506
 507         FREE(enc->cpb_array);
 508         FREE(enc);
 509         return NULL;
 510 }
 511
 512 /**
 513  * check if kernel has the right fw version loaded
 514  */
 515 bool rvce_is_fw_version_supported(struct r600_common_screen *rscreen)
 516 {
 517         switch (rscreen->info.vce_fw_version) {
 518         case FW_40_2_2:
 519         case FW_50_0_1:
 520         case FW_50_1_2:
 521         case FW_50_10_2:
 522         case FW_50_17_3:
 523         case FW_52_0_3:
 524         case FW_52_4_3:
 525                 return true;
 526         default:
 527                 return false;
 528         }
 529 }
 530
 531 /**
 532  * Add the buffer as relocation to the current command submission
 533  */
 534 void rvce_add_buffer(struct rvce_encoder *enc, struct pb_buffer *buf,
 535                      enum radeon_bo_usage usage, enum radeon_bo_domain domain,
 536                      signed offset)
 537 {
 538         int reloc_idx;
 539
 540         reloc_idx = enc->ws->cs_add_buffer(enc->cs, buf, usage, domain, RADEON_PRIO_VCE);
 541         if (enc->use_vm) {
 542                 uint64_t addr;
 543                 addr = enc->ws->buffer_get_virtual_address(buf);
 544                 addr = addr + offset;
 545                 RVCE_CS(addr >> 32);
 546                 RVCE_CS(addr);
 547         } else {
 548                 RVCE_CS(reloc_idx * 4);
 549                 RVCE_CS(offset);
 550         }
 551 }