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
 271         enc->get_buffer(vid_buf->resources[0], &enc->handle, &enc->luma);
 272         enc->get_buffer(vid_buf->resources[1], NULL, &enc->chroma);
 273
 274         if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR)
 275                 reset_cpb(enc);
 276         else if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_P ||
 277                  pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_B)
 278                 sort_cpb(enc);
 279
 280         if (!enc->stream_handle) {
 281                 struct rvid_buffer fb;
 282                 enc->stream_handle = rvid_alloc_stream_handle();
 283                 rvid_create_buffer(enc->screen, &fb, 512, PIPE_USAGE_STAGING);
 284                 enc->fb = &fb;
 285                 enc->session(enc);
 286                 enc->create(enc);
 287                 enc->config(enc);
 288                 enc->feedback(enc);
 289                 flush(enc);
 290                 //dump_feedback(enc, &fb);
 291                 rvid_destroy_buffer(&fb);
 292                 need_rate_control = false;
 293         }
 294
 295         if (need_rate_control) {
 296                 enc->session(enc);
 297                 enc->config(enc);
 298                 flush(enc);
 299         }
 300 }
 301
 302 static void rvce_encode_bitstream(struct pipe_video_codec *encoder,
 303                                   struct pipe_video_buffer *source,
 304                                   struct pipe_resource *destination,
 305                                   void **fb)
 306 {
 307         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 308         enc->get_buffer(destination, &enc->bs_handle, NULL);
 309         enc->bs_size = destination->width0;
 310
 311         *fb = enc->fb = CALLOC_STRUCT(rvid_buffer);
 312         if (!rvid_create_buffer(enc->screen, enc->fb, 512, PIPE_USAGE_STAGING)) {
 313                 RVID_ERR("Can't create feedback buffer.\n");
 314                 return;
 315         }
 316         if (!enc->cs->cdw)
 317                 enc->session(enc);
 318         enc->encode(enc);
 319         enc->feedback(enc);
 320 }
 321
 322 static void rvce_end_frame(struct pipe_video_codec *encoder,
 323                            struct pipe_video_buffer *source,
 324                            struct pipe_picture_desc *picture)
 325 {
 326         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 327         struct rvce_cpb_slot *slot = LIST_ENTRY(
 328                 struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 329
 330         if (!enc->dual_inst || enc->bs_idx > 1)
 331                 flush(enc);
 332
 333         /* update the CPB backtrack with the just encoded frame */
 334         slot->picture_type = enc->pic.picture_type;
 335         slot->frame_num = enc->pic.frame_num;
 336         slot->pic_order_cnt = enc->pic.pic_order_cnt;
 337         if (!enc->pic.not_referenced) {
 338                 LIST_DEL(&slot->list);
 339                 LIST_ADD(&slot->list, &enc->cpb_slots);
 340         }
 341 }
 342
 343 static void rvce_get_feedback(struct pipe_video_codec *encoder,
 344                               void *feedback, unsigned *size)
 345 {
 346         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 347         struct rvid_buffer *fb = feedback;
 348
 349         if (size) {
 350                 uint32_t *ptr = enc->ws->buffer_map(fb->res->buf, enc->cs, PIPE_TRANSFER_READ_WRITE);
 351
 352                 if (ptr[1]) {
 353                         *size = ptr[4] - ptr[9];
 354                 } else {
 355                         *size = 0;
 356                 }
 357
 358                 enc->ws->buffer_unmap(fb->res->buf);
 359         }
 360         //dump_feedback(enc, fb);
 361         rvid_destroy_buffer(fb);
 362         FREE(fb);
 363 }
 364
 365 /**
 366  * flush any outstanding command buffers to the hardware
 367  */
 368 static void rvce_flush(struct pipe_video_codec *encoder)
 369 {
 370         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 371
 372         flush(enc);
 373 }
 374
 375 static void rvce_cs_flush(void *ctx, unsigned flags,
 376                           struct pipe_fence_handle **fence)
 377 {
 378         // just ignored
 379 }
 380
 381 struct pipe_video_codec *rvce_create_encoder(struct pipe_context *context,
 382                                              const struct pipe_video_codec *templ,
 383                                              struct radeon_winsys* ws,
 384                                              rvce_get_buffer get_buffer)
 385 {
 386         struct r600_common_screen *rscreen = (struct r600_common_screen *)context->screen;
 387         struct r600_common_context *rctx = (struct r600_common_context*)context;
 388         struct rvce_encoder *enc;
 389         struct pipe_video_buffer *tmp_buf, templat = {};
 390         struct radeon_surf *tmp_surf;
 391         unsigned cpb_size;
 392
 393         if (!rscreen->info.vce_fw_version) {
 394                 RVID_ERR("Kernel doesn't supports VCE!\n");
 395                 return NULL;
 396
 397         } else if (!rvce_is_fw_version_supported(rscreen)) {
 398                 RVID_ERR("Unsupported VCE fw version loaded!\n");
 399                 return NULL;
 400         }
 401
 402         enc = CALLOC_STRUCT(rvce_encoder);
 403         if (!enc)
 404                 return NULL;
 405
 406         if (rscreen->info.drm_major == 3)
 407                 enc->use_vm = true;
 408         if ((rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 42) ||
 409             rscreen->info.drm_major == 3)
 410                 enc->use_vui = true;
 411         if (rscreen->info.family >= CHIP_TONGA &&
 412             rscreen->info.family != CHIP_STONEY &&
 413             rscreen->info.family != CHIP_POLARIS11)
 414                 enc->dual_pipe = true;
 415         /* TODO enable B frame with dual instance */
 416         if ((rscreen->info.family >= CHIP_TONGA) &&
 417                 (templ->max_references == 1) &&
 418                 (rscreen->info.vce_harvest_config == 0))
 419                 enc->dual_inst = true;
 420
 421         enc->base = *templ;
 422         enc->base.context = context;
 423
 424         enc->base.destroy = rvce_destroy;
 425         enc->base.begin_frame = rvce_begin_frame;
 426         enc->base.encode_bitstream = rvce_encode_bitstream;
 427         enc->base.end_frame = rvce_end_frame;
 428         enc->base.flush = rvce_flush;
 429         enc->base.get_feedback = rvce_get_feedback;
 430         enc->get_buffer = get_buffer;
 431
 432         enc->screen = context->screen;
 433         enc->ws = ws;
 434         enc->cs = ws->cs_create(rctx->ctx, RING_VCE, rvce_cs_flush, enc);
 435         if (!enc->cs) {
 436                 RVID_ERR("Can't get command submission context.\n");
 437                 goto error;
 438         }
 439
 440         templat.buffer_format = PIPE_FORMAT_NV12;
 441         templat.chroma_format = PIPE_VIDEO_CHROMA_FORMAT_420;
 442         templat.width = enc->base.width;
 443         templat.height = enc->base.height;
 444         templat.interlaced = false;
 445         if (!(tmp_buf = context->create_video_buffer(context, &templat))) {
 446                 RVID_ERR("Can't create video buffer.\n");
 447                 goto error;
 448         }
 449
 450         enc->cpb_num = get_cpb_num(enc);
 451         if (!enc->cpb_num)
 452                 goto error;
 453
 454         get_buffer(((struct vl_video_buffer *)tmp_buf)->resources[0], NULL, &tmp_surf);
 455         cpb_size = align(tmp_surf->level[0].pitch_bytes, 128);
 456         cpb_size = cpb_size * align(tmp_surf->npix_y, 16);
 457         cpb_size = cpb_size * 3 / 2;
 458         cpb_size = cpb_size * enc->cpb_num;
 459         if (enc->dual_pipe)
 460                 cpb_size +=  RVCE_MAX_AUX_BUFFER_NUM *
 461                         RVCE_MAX_BITSTREAM_OUTPUT_ROW_SIZE * 2;
 462         tmp_buf->destroy(tmp_buf);
 463         if (!rvid_create_buffer(enc->screen, &enc->cpb, cpb_size, PIPE_USAGE_DEFAULT)) {
 464                 RVID_ERR("Can't create CPB buffer.\n");
 465                 goto error;
 466         }
 467
 468         enc->cpb_array = CALLOC(enc->cpb_num, sizeof(struct rvce_cpb_slot));
 469         if (!enc->cpb_array)
 470                 goto error;
 471
 472         reset_cpb(enc);
 473
 474         switch (rscreen->info.vce_fw_version) {
 475         case FW_40_2_2:
 476                 radeon_vce_40_2_2_init(enc);
 477                 break;
 478
 479         case FW_50_0_1:
 480         case FW_50_1_2:
 481         case FW_50_10_2:
 482         case FW_50_17_3:
 483                 radeon_vce_50_init(enc);
 484                 break;
 485
 486         case FW_52_0_3:
 487         case FW_52_4_3:
 488                 radeon_vce_52_init(enc);
 489                 break;
 490
 491         default:
 492                 goto error;
 493         }
 494
 495         return &enc->base;
 496
 497 error:
 498         if (enc->cs)
 499                 enc->ws->cs_destroy(enc->cs);
 500
 501         rvid_destroy_buffer(&enc->cpb);
 502
 503         FREE(enc->cpb_array);
 504         FREE(enc);
 505         return NULL;
 506 }
 507
 508 /**
 509  * check if kernel has the right fw version loaded
 510  */
 511 bool rvce_is_fw_version_supported(struct r600_common_screen *rscreen)
 512 {
 513         switch (rscreen->info.vce_fw_version) {
 514         case FW_40_2_2:
 515         case FW_50_0_1:
 516         case FW_50_1_2:
 517         case FW_50_10_2:
 518         case FW_50_17_3:
 519         case FW_52_0_3:
 520         case FW_52_4_3:
 521                 return true;
 522         default:
 523                 return false;
 524         }
 525 }
 526
 527 /**
 528  * Add the buffer as relocation to the current command submission
 529  */
 530 void rvce_add_buffer(struct rvce_encoder *enc, struct pb_buffer *buf,
 531                      enum radeon_bo_usage usage, enum radeon_bo_domain domain,
 532                      signed offset)
 533 {
 534         int reloc_idx;
 535
 536         reloc_idx = enc->ws->cs_add_buffer(enc->cs, buf, usage, domain, RADEON_PRIO_VCE);
 537         if (enc->use_vm) {
 538                 uint64_t addr;
 539                 addr = enc->ws->buffer_get_virtual_address(buf);
 540                 addr = addr + offset;
 541                 RVCE_CS(addr >> 32);
 542                 RVCE_CS(addr);
 543         } else {
 544                 RVCE_CS(reloc_idx * 4);
 545                 RVCE_CS(offset);
 546         }
 547 }