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