src/gallium/drivers/r600/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 #define FW_52_8_3 ((52 << 24) | (8 << 16) | (3 << 8))
  55 #define FW_53 (53 << 24)
  56
  57 /**
  58  * flush commands to the hardware
  59  */
  60 static void flush(struct rvce_encoder *enc)
  61 {
  62         enc->ws->cs_flush(enc->cs, PIPE_FLUSH_ASYNC, NULL);
  63         enc->task_info_idx = 0;
  64         enc->bs_idx = 0;
  65 }
  66
  67 #if 0
  68 static void dump_feedback(struct rvce_encoder *enc, struct rvid_buffer *fb)
  69 {
  70         uint32_t *ptr = enc->ws->buffer_map(fb->res->buf, enc->cs, PIPE_TRANSFER_READ_WRITE);
  71         unsigned i = 0;
  72         fprintf(stderr, "\n");
  73         fprintf(stderr, "encStatus:\t\t\t%08x\n", ptr[i++]);
  74         fprintf(stderr, "encHasBitstream:\t\t%08x\n", ptr[i++]);
  75         fprintf(stderr, "encHasAudioBitstream:\t\t%08x\n", ptr[i++]);
  76         fprintf(stderr, "encBitstreamOffset:\t\t%08x\n", ptr[i++]);
  77         fprintf(stderr, "encBitstreamSize:\t\t%08x\n", ptr[i++]);
  78         fprintf(stderr, "encAudioBitstreamOffset:\t%08x\n", ptr[i++]);
  79         fprintf(stderr, "encAudioBitstreamSize:\t\t%08x\n", ptr[i++]);
  80         fprintf(stderr, "encExtrabytes:\t\t\t%08x\n", ptr[i++]);
  81         fprintf(stderr, "encAudioExtrabytes:\t\t%08x\n", ptr[i++]);
  82         fprintf(stderr, "videoTimeStamp:\t\t\t%08x\n", ptr[i++]);
  83         fprintf(stderr, "audioTimeStamp:\t\t\t%08x\n", ptr[i++]);
  84         fprintf(stderr, "videoOutputType:\t\t%08x\n", ptr[i++]);
  85         fprintf(stderr, "attributeFlags:\t\t\t%08x\n", ptr[i++]);
  86         fprintf(stderr, "seiPrivatePackageOffset:\t%08x\n", ptr[i++]);
  87         fprintf(stderr, "seiPrivatePackageSize:\t\t%08x\n", ptr[i++]);
  88         fprintf(stderr, "\n");
  89         enc->ws->buffer_unmap(fb->res->buf);
  90 }
  91 #endif
  92
  93 /**
  94  * reset the CPB handling
  95  */
  96 static void reset_cpb(struct rvce_encoder *enc)
  97 {
  98         unsigned i;
  99
 100         LIST_INITHEAD(&enc->cpb_slots);
 101         for (i = 0; i < enc->cpb_num; ++i) {
 102                 struct rvce_cpb_slot *slot = &enc->cpb_array[i];
 103                 slot->index = i;
 104                 slot->picture_type = PIPE_H264_ENC_PICTURE_TYPE_SKIP;
 105                 slot->frame_num = 0;
 106                 slot->pic_order_cnt = 0;
 107                 LIST_ADDTAIL(&slot->list, &enc->cpb_slots);
 108         }
 109 }
 110
 111 /**
 112  * sort l0 and l1 to the top of the list
 113  */
 114 static void sort_cpb(struct rvce_encoder *enc)
 115 {
 116         struct rvce_cpb_slot *i, *l0 = NULL, *l1 = NULL;
 117
 118         LIST_FOR_EACH_ENTRY(i, &enc->cpb_slots, list) {
 119                 if (i->frame_num == enc->pic.ref_idx_l0)
 120                         l0 = i;
 121
 122                 if (i->frame_num == enc->pic.ref_idx_l1)
 123                         l1 = i;
 124
 125                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P && l0)
 126                         break;
 127
 128                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_B &&
 129                     l0 && l1)
 130                         break;
 131         }
 132
 133         if (l1) {
 134                 LIST_DEL(&l1->list);
 135                 LIST_ADD(&l1->list, &enc->cpb_slots);
 136         }
 137
 138         if (l0) {
 139                 LIST_DEL(&l0->list);
 140                 LIST_ADD(&l0->list, &enc->cpb_slots);
 141         }
 142 }
 143
 144 /**
 145  * get number of cpbs based on dpb
 146  */
 147 static unsigned get_cpb_num(struct rvce_encoder *enc)
 148 {
 149         unsigned w = align(enc->base.width, 16) / 16;
 150         unsigned h = align(enc->base.height, 16) / 16;
 151         unsigned dpb;
 152
 153         switch (enc->base.level) {
 154         case 10:
 155                 dpb = 396;
 156                 break;
 157         case 11:
 158                 dpb = 900;
 159                 break;
 160         case 12:
 161         case 13:
 162         case 20:
 163                 dpb = 2376;
 164                 break;
 165         case 21:
 166                 dpb = 4752;
 167                 break;
 168         case 22:
 169         case 30:
 170                 dpb = 8100;
 171                 break;
 172         case 31:
 173                 dpb = 18000;
 174                 break;
 175         case 32:
 176                 dpb = 20480;
 177                 break;
 178         case 40:
 179         case 41:
 180                 dpb = 32768;
 181                 break;
 182         case 42:
 183                 dpb = 34816;
 184                 break;
 185         case 50:
 186                 dpb = 110400;
 187                 break;
 188         default:
 189         case 51:
 190         case 52:
 191                 dpb = 184320;
 192                 break;
 193         }
 194
 195         return MIN2(dpb / (w * h), 16);
 196 }
 197
 198 /**
 199  * Get the slot for the currently encoded frame
 200  */
 201 struct rvce_cpb_slot *current_slot(struct rvce_encoder *enc)
 202 {
 203         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 204 }
 205
 206 /**
 207  * Get the slot for L0
 208  */
 209 struct rvce_cpb_slot *l0_slot(struct rvce_encoder *enc)
 210 {
 211         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next, list);
 212 }
 213
 214 /**
 215  * Get the slot for L1
 216  */
 217 struct rvce_cpb_slot *l1_slot(struct rvce_encoder *enc)
 218 {
 219         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next->next, list);
 220 }
 221
 222 /**
 223  * Calculate the offsets into the CPB
 224  */
 225 void rvce_frame_offset(struct rvce_encoder *enc, struct rvce_cpb_slot *slot,
 226                        signed *luma_offset, signed *chroma_offset)
 227 {
 228         unsigned pitch, vpitch, fsize;
 229
 230         pitch = align(enc->luma->u.legacy.level[0].nblk_x * enc->luma->bpe, 128);
 231         vpitch = align(enc->luma->u.legacy.level[0].nblk_y, 16);
 232         fsize = pitch * (vpitch + vpitch / 2);
 233
 234         *luma_offset = slot->index * fsize;
 235         *chroma_offset = *luma_offset + pitch * vpitch;
 236 }
 237
 238 /**
 239  * destroy this video encoder
 240  */
 241 static void rvce_destroy(struct pipe_video_codec *encoder)
 242 {
 243         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 244         if (enc->stream_handle) {
 245                 struct rvid_buffer fb;
 246                 rvid_create_buffer(enc->screen, &fb, 512, PIPE_USAGE_STAGING);
 247                 enc->fb = &fb;
 248                 enc->session(enc);
 249                 enc->feedback(enc);
 250                 enc->destroy(enc);
 251                 flush(enc);
 252                 rvid_destroy_buffer(&fb);
 253         }
 254         rvid_destroy_buffer(&enc->cpb);
 255         enc->ws->cs_destroy(enc->cs);
 256         FREE(enc->cpb_array);
 257         FREE(enc);
 258 }
 259
 260 static void rvce_begin_frame(struct pipe_video_codec *encoder,
 261                              struct pipe_video_buffer *source,
 262                              struct pipe_picture_desc *picture)
 263 {
 264         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 265         struct vl_video_buffer *vid_buf = (struct vl_video_buffer *)source;
 266         struct pipe_h264_enc_picture_desc *pic = (struct pipe_h264_enc_picture_desc *)picture;
 267
 268         bool need_rate_control =
 269                 enc->pic.rate_ctrl.rate_ctrl_method != pic->rate_ctrl.rate_ctrl_method ||
 270                 enc->pic.quant_i_frames != pic->quant_i_frames ||
 271                 enc->pic.quant_p_frames != pic->quant_p_frames ||
 272                 enc->pic.quant_b_frames != pic->quant_b_frames;
 273
 274         enc->pic = *pic;
 275         get_pic_param(enc, pic);
 276
 277         enc->get_buffer(vid_buf->resources[0], &enc->handle, &enc->luma);
 278         enc->get_buffer(vid_buf->resources[1], NULL, &enc->chroma);
 279
 280         if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR)
 281                 reset_cpb(enc);
 282         else if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_P ||
 283                  pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_B)
 284                 sort_cpb(enc);
 285
 286         if (!enc->stream_handle) {
 287                 struct rvid_buffer fb;
 288                 enc->stream_handle = rvid_alloc_stream_handle();
 289                 rvid_create_buffer(enc->screen, &fb, 512, PIPE_USAGE_STAGING);
 290                 enc->fb = &fb;
 291                 enc->session(enc);
 292                 enc->create(enc);
 293                 enc->config(enc);
 294                 enc->feedback(enc);
 295                 flush(enc);
 296                 //dump_feedback(enc, &fb);
 297                 rvid_destroy_buffer(&fb);
 298                 need_rate_control = false;
 299         }
 300
 301         if (need_rate_control) {
 302                 enc->session(enc);
 303                 enc->config(enc);
 304                 flush(enc);
 305         }
 306 }
 307
 308 static void rvce_encode_bitstream(struct pipe_video_codec *encoder,
 309                                   struct pipe_video_buffer *source,
 310                                   struct pipe_resource *destination,
 311                                   void **fb)
 312 {
 313         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 314         enc->get_buffer(destination, &enc->bs_handle, NULL);
 315         enc->bs_size = destination->width0;
 316
 317         *fb = enc->fb = CALLOC_STRUCT(rvid_buffer);
 318         if (!rvid_create_buffer(enc->screen, enc->fb, 512, PIPE_USAGE_STAGING)) {
 319                 RVID_ERR("Can't create feedback buffer.\n");
 320                 return;
 321         }
 322         if (!radeon_emitted(enc->cs, 0))
 323                 enc->session(enc);
 324         enc->encode(enc);
 325         enc->feedback(enc);
 326 }
 327
 328 static void rvce_end_frame(struct pipe_video_codec *encoder,
 329                            struct pipe_video_buffer *source,
 330                            struct pipe_picture_desc *picture)
 331 {
 332         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 333         struct rvce_cpb_slot *slot = LIST_ENTRY(
 334                 struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 335
 336         if (!enc->dual_inst || enc->bs_idx > 1)
 337                 flush(enc);
 338
 339         /* update the CPB backtrack with the just encoded frame */
 340         slot->picture_type = enc->pic.picture_type;
 341         slot->frame_num = enc->pic.frame_num;
 342         slot->pic_order_cnt = enc->pic.pic_order_cnt;
 343         if (!enc->pic.not_referenced) {
 344                 LIST_DEL(&slot->list);
 345                 LIST_ADD(&slot->list, &enc->cpb_slots);
 346         }
 347 }
 348
 349 static void rvce_get_feedback(struct pipe_video_codec *encoder,
 350                               void *feedback, unsigned *size)
 351 {
 352         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 353         struct rvid_buffer *fb = feedback;
 354
 355         if (size) {
 356                 uint32_t *ptr = enc->ws->buffer_map(
 357                         fb->res->buf, enc->cs,
 358                         PIPE_TRANSFER_READ_WRITE | RADEON_TRANSFER_TEMPORARY);
 359
 360                 if (ptr[1]) {
 361                         *size = ptr[4] - ptr[9];
 362                 } else {
 363                         *size = 0;
 364                 }
 365
 366                 enc->ws->buffer_unmap(fb->res->buf);
 367         }
 368         //dump_feedback(enc, fb);
 369         rvid_destroy_buffer(fb);
 370         FREE(fb);
 371 }
 372
 373 /**
 374  * flush any outstanding command buffers to the hardware
 375  */
 376 static void rvce_flush(struct pipe_video_codec *encoder)
 377 {
 378         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 379
 380         flush(enc);
 381 }
 382
 383 static void rvce_cs_flush(void *ctx, unsigned flags,
 384                           struct pipe_fence_handle **fence)
 385 {
 386         // just ignored
 387 }
 388
 389 struct pipe_video_codec *rvce_create_encoder(struct pipe_context *context,
 390                                              const struct pipe_video_codec *templ,
 391                                              struct radeon_winsys* ws,
 392                                              rvce_get_buffer get_buffer)
 393 {
 394         struct r600_common_screen *rscreen = (struct r600_common_screen *)context->screen;
 395         struct r600_common_context *rctx = (struct r600_common_context*)context;
 396         struct rvce_encoder *enc;
 397         struct pipe_video_buffer *tmp_buf, templat = {};
 398         struct radeon_surf *tmp_surf;
 399         unsigned cpb_size;
 400
 401         if (!rscreen->info.vce_fw_version) {
 402                 RVID_ERR("Kernel doesn't supports VCE!\n");
 403                 return NULL;
 404
 405         } else if (!rvce_is_fw_version_supported(rscreen)) {
 406                 RVID_ERR("Unsupported VCE fw version loaded!\n");
 407                 return NULL;
 408         }
 409
 410         enc = CALLOC_STRUCT(rvce_encoder);
 411         if (!enc)
 412                 return NULL;
 413
 414         if (rscreen->info.drm_major == 3)
 415                 enc->use_vm = true;
 416         if ((rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 42) ||
 417             rscreen->info.drm_major == 3)
 418                 enc->use_vui = true;
 419
 420         enc->base = *templ;
 421         enc->base.context = context;
 422
 423         enc->base.destroy = rvce_destroy;
 424         enc->base.begin_frame = rvce_begin_frame;
 425         enc->base.encode_bitstream = rvce_encode_bitstream;
 426         enc->base.end_frame = rvce_end_frame;
 427         enc->base.flush = rvce_flush;
 428         enc->base.get_feedback = rvce_get_feedback;
 429         enc->get_buffer = get_buffer;
 430
 431         enc->screen = context->screen;
 432         enc->ws = ws;
 433         enc->cs = ws->cs_create(rctx->ctx, RING_VCE, rvce_cs_flush, enc, false);
 434         if (!enc->cs) {
 435                 RVID_ERR("Can't get command submission context.\n");
 436                 goto error;
 437         }
 438
 439         templat.buffer_format = PIPE_FORMAT_NV12;
 440         templat.chroma_format = PIPE_VIDEO_CHROMA_FORMAT_420;
 441         templat.width = enc->base.width;
 442         templat.height = enc->base.height;
 443         templat.interlaced = false;
 444         if (!(tmp_buf = context->create_video_buffer(context, &templat))) {
 445                 RVID_ERR("Can't create video buffer.\n");
 446                 goto error;
 447         }
 448
 449         enc->cpb_num = get_cpb_num(enc);
 450         if (!enc->cpb_num)
 451                 goto error;
 452
 453         get_buffer(((struct vl_video_buffer *)tmp_buf)->resources[0], NULL, &tmp_surf);
 454
 455         cpb_size = align(tmp_surf->u.legacy.level[0].nblk_x * tmp_surf->bpe, 128) *
 456           align(tmp_surf->u.legacy.level[0].nblk_y, 32);
 457
 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         goto error;
 476
 477         return &enc->base;
 478
 479 error:
 480         if (enc->cs)
 481                 enc->ws->cs_destroy(enc->cs);
 482
 483         rvid_destroy_buffer(&enc->cpb);
 484
 485         FREE(enc->cpb_array);
 486         FREE(enc);
 487         return NULL;
 488 }
 489
 490 /**
 491  * check if kernel has the right fw version loaded
 492  */
 493 bool rvce_is_fw_version_supported(struct r600_common_screen *rscreen)
 494 {
 495         switch (rscreen->info.vce_fw_version) {
 496         case FW_40_2_2:
 497         case FW_50_0_1:
 498         case FW_50_1_2:
 499         case FW_50_10_2:
 500         case FW_50_17_3:
 501         case FW_52_0_3:
 502         case FW_52_4_3:
 503         case FW_52_8_3:
 504                 return true;
 505         default:
 506                 if ((rscreen->info.vce_fw_version & (0xff << 24)) == FW_53)
 507                         return true;
 508                 else
 509                         return false;
 510         }
 511 }
 512
 513 /**
 514  * Add the buffer as relocation to the current command submission
 515  */
 516 void rvce_add_buffer(struct rvce_encoder *enc, struct pb_buffer *buf,
 517                      enum radeon_bo_usage usage, enum radeon_bo_domain domain,
 518                      signed offset)
 519 {
 520         int reloc_idx;
 521
 522         reloc_idx = enc->ws->cs_add_buffer(enc->cs, buf, usage | RADEON_USAGE_SYNCHRONIZED,
 523                                            domain, 0);
 524         if (enc->use_vm) {
 525                 uint64_t addr;
 526                 addr = enc->ws->buffer_get_virtual_address(buf);
 527                 addr = addr + offset;
 528                 RVCE_CS(addr >> 32);
 529                 RVCE_CS(addr);
 530         } else {
 531                 offset += enc->ws->buffer_get_reloc_offset(buf);
 532                 RVCE_CS(reloc_idx * 4);
 533                 RVCE_CS(offset);
 534         }
 535 }