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