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 "radeon/drm/radeon_winsys.h"
  44 #include "r600_pipe_common.h"
  45 #include "radeon_video.h"
  46 #include "radeon_vce.h"
  47
  48 /**
  49  * flush commands to the hardware
  50  */
  51 static void flush(struct rvce_encoder *enc)
  52 {
  53         enc->ws->cs_flush(enc->cs, RADEON_FLUSH_ASYNC, NULL, 0);
  54 }
  55
  56 #if 0
  57 static void dump_feedback(struct rvce_encoder *enc, struct rvid_buffer *fb)
  58 {
  59         uint32_t *ptr = enc->ws->buffer_map(fb->cs_handle, enc->cs, PIPE_TRANSFER_READ_WRITE);
  60         unsigned i = 0;
  61         fprintf(stderr, "\n");
  62         fprintf(stderr, "encStatus:\t\t\t%08x\n", ptr[i++]);
  63         fprintf(stderr, "encHasBitstream:\t\t%08x\n", ptr[i++]);
  64         fprintf(stderr, "encHasAudioBitstream:\t\t%08x\n", ptr[i++]);
  65         fprintf(stderr, "encBitstreamOffset:\t\t%08x\n", ptr[i++]);
  66         fprintf(stderr, "encBitstreamSize:\t\t%08x\n", ptr[i++]);
  67         fprintf(stderr, "encAudioBitstreamOffset:\t%08x\n", ptr[i++]);
  68         fprintf(stderr, "encAudioBitstreamSize:\t\t%08x\n", ptr[i++]);
  69         fprintf(stderr, "encExtrabytes:\t\t\t%08x\n", ptr[i++]);
  70         fprintf(stderr, "encAudioExtrabytes:\t\t%08x\n", ptr[i++]);
  71         fprintf(stderr, "videoTimeStamp:\t\t\t%08x\n", ptr[i++]);
  72         fprintf(stderr, "audioTimeStamp:\t\t\t%08x\n", ptr[i++]);
  73         fprintf(stderr, "videoOutputType:\t\t%08x\n", ptr[i++]);
  74         fprintf(stderr, "attributeFlags:\t\t\t%08x\n", ptr[i++]);
  75         fprintf(stderr, "seiPrivatePackageOffset:\t%08x\n", ptr[i++]);
  76         fprintf(stderr, "seiPrivatePackageSize:\t\t%08x\n", ptr[i++]);
  77         fprintf(stderr, "\n");
  78         enc->ws->buffer_unmap(fb->cs_handle);
  79 }
  80 #endif
  81
  82 /**
  83  * reset the CPB handling
  84  */
  85 static void reset_cpb(struct rvce_encoder *enc)
  86 {
  87         unsigned i;
  88
  89         LIST_INITHEAD(&enc->cpb_slots);
  90         for (i = 0; i < enc->cpb_num; ++i) {
  91                 struct rvce_cpb_slot *slot = &enc->cpb_array[i];
  92                 slot->index = i;
  93                 slot->picture_type = PIPE_H264_ENC_PICTURE_TYPE_SKIP;
  94                 slot->frame_num = 0;
  95                 slot->pic_order_cnt = 0;
  96                 LIST_ADDTAIL(&slot->list, &enc->cpb_slots);
  97         }
  98 }
  99
 100 /**
 101  * sort l0 and l1 to the top of the list
 102  */
 103 static void sort_cpb(struct rvce_encoder *enc)
 104 {
 105         struct rvce_cpb_slot *i, *l0 = NULL, *l1 = NULL;
 106
 107         LIST_FOR_EACH_ENTRY(i, &enc->cpb_slots, list) {
 108                 if (i->frame_num == enc->pic.ref_idx_l0)
 109                         l0 = i;
 110
 111                 if (i->frame_num == enc->pic.ref_idx_l1)
 112                         l1 = i;
 113
 114                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P && l0)
 115                         break;
 116
 117                 if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_B &&
 118                     l0 && l1)
 119                         break;
 120         }
 121
 122         if (l1) {
 123                 LIST_DEL(&l1->list);
 124                 LIST_ADD(&l1->list, &enc->cpb_slots);
 125         }
 126
 127         if (l0) {
 128                 LIST_DEL(&l0->list);
 129                 LIST_ADD(&l0->list, &enc->cpb_slots);
 130         }
 131 }
 132
 133 /**
 134  * get number of cpbs based on dpb
 135  */
 136 static unsigned get_cpb_num(struct rvce_encoder *enc)
 137 {
 138         unsigned w = align(enc->base.width, 16) / 16;
 139         unsigned h = align(enc->base.height, 16) / 16;
 140         unsigned dpb;
 141
 142         switch (enc->base.level) {
 143         case 10:
 144                 dpb = 396;
 145                 break;
 146         case 11:
 147                 dpb = 900;
 148                 break;
 149         case 12:
 150         case 13:
 151         case 20:
 152                 dpb = 2376;
 153                 break;
 154         case 21:
 155                 dpb = 4752;
 156                 break;
 157         case 22:
 158         case 30:
 159                 dpb = 8100;
 160                 break;
 161         case 31:
 162                 dpb = 18000;
 163                 break;
 164         case 32:
 165                 dpb = 20480;
 166                 break;
 167         case 40:
 168         case 41:
 169                 dpb = 32768;
 170                 break;
 171         default:
 172         case 42:
 173                 dpb = 34816;
 174                 break;
 175         case 50:
 176                 dpb = 110400;
 177                 break;
 178         case 51:
 179                 dpb = 184320;
 180                 break;
 181         }
 182
 183         return MIN2(dpb / (w * h), 16);
 184 }
 185
 186 /**
 187  * destroy this video encoder
 188  */
 189 static void rvce_destroy(struct pipe_video_codec *encoder)
 190 {
 191         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 192         if (enc->stream_handle) {
 193                 struct rvid_buffer fb;
 194                 rvid_create_buffer(enc->ws, &fb, 512, RADEON_DOMAIN_GTT, 0);
 195                 enc->fb = &fb;
 196                 enc->session(enc);
 197                 enc->feedback(enc);
 198                 enc->destroy(enc);
 199                 flush(enc);
 200                 rvid_destroy_buffer(&fb);
 201         }
 202         rvid_destroy_buffer(&enc->cpb);
 203         enc->ws->cs_destroy(enc->cs);
 204         FREE(enc->cpb_array);
 205         FREE(enc);
 206 }
 207
 208 static void rvce_begin_frame(struct pipe_video_codec *encoder,
 209                              struct pipe_video_buffer *source,
 210                              struct pipe_picture_desc *picture)
 211 {
 212         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 213         struct vl_video_buffer *vid_buf = (struct vl_video_buffer *)source;
 214         struct pipe_h264_enc_picture_desc *pic = (struct pipe_h264_enc_picture_desc *)picture;
 215
 216         bool need_rate_control =
 217                 enc->pic.rate_ctrl.rate_ctrl_method != pic->rate_ctrl.rate_ctrl_method ||
 218                 enc->pic.quant_i_frames != pic->quant_i_frames ||
 219                 enc->pic.quant_p_frames != pic->quant_p_frames ||
 220                 enc->pic.quant_b_frames != pic->quant_b_frames;
 221
 222         enc->pic = *pic;
 223
 224         enc->get_buffer(vid_buf->resources[0], &enc->handle, &enc->luma);
 225         enc->get_buffer(vid_buf->resources[1], NULL, &enc->chroma);
 226
 227         if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR)
 228                 reset_cpb(enc);
 229         else if (pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_P ||
 230                  pic->picture_type == PIPE_H264_ENC_PICTURE_TYPE_B)
 231                 sort_cpb(enc);
 232
 233         if (!enc->stream_handle) {
 234                 struct rvid_buffer fb;
 235                 enc->stream_handle = rvid_alloc_stream_handle();
 236                 rvid_create_buffer(enc->ws, &fb, 512, RADEON_DOMAIN_GTT, 0);
 237                 enc->fb = &fb;
 238                 enc->session(enc);
 239                 enc->create(enc);
 240                 enc->rate_control(enc);
 241                 need_rate_control = false;
 242                 enc->config_extension(enc);
 243                 enc->motion_estimation(enc);
 244                 enc->rdo(enc);
 245                 enc->pic_control(enc);
 246                 enc->feedback(enc);
 247                 flush(enc);
 248                 //dump_feedback(enc, &fb);
 249                 rvid_destroy_buffer(&fb);
 250         }
 251
 252         enc->session(enc);
 253
 254         if (need_rate_control)
 255                 enc->rate_control(enc);
 256 }
 257
 258 static void rvce_encode_bitstream(struct pipe_video_codec *encoder,
 259                                   struct pipe_video_buffer *source,
 260                                   struct pipe_resource *destination,
 261                                   void **fb)
 262 {
 263         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 264         enc->get_buffer(destination, &enc->bs_handle, NULL);
 265         enc->bs_size = destination->width0;
 266
 267         *fb = enc->fb = CALLOC_STRUCT(rvid_buffer);
 268         if (!rvid_create_buffer(enc->ws, enc->fb, 512, RADEON_DOMAIN_GTT, 0)) {
 269                 RVID_ERR("Can't create feedback buffer.\n");
 270                 return;
 271         }
 272         enc->encode(enc);
 273         enc->feedback(enc);
 274 }
 275
 276 static void rvce_end_frame(struct pipe_video_codec *encoder,
 277                            struct pipe_video_buffer *source,
 278                            struct pipe_picture_desc *picture)
 279 {
 280         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 281         struct rvce_cpb_slot *slot = LIST_ENTRY(
 282                 struct rvce_cpb_slot, enc->cpb_slots.prev, list);
 283
 284         flush(enc);
 285
 286         /* update the CPB backtrack with the just encoded frame */
 287         slot->picture_type = enc->pic.picture_type;
 288         slot->frame_num = enc->pic.frame_num;
 289         slot->pic_order_cnt = enc->pic.pic_order_cnt;
 290         if (!enc->pic.not_referenced) {
 291                 LIST_DEL(&slot->list);
 292                 LIST_ADD(&slot->list, &enc->cpb_slots);
 293         }
 294 }
 295
 296 static void rvce_get_feedback(struct pipe_video_codec *encoder,
 297                               void *feedback, unsigned *size)
 298 {
 299         struct rvce_encoder *enc = (struct rvce_encoder*)encoder;
 300         struct rvid_buffer *fb = feedback;
 301
 302         if (size) {
 303                 uint32_t *ptr = enc->ws->buffer_map(fb->cs_handle, enc->cs, PIPE_TRANSFER_READ_WRITE);
 304
 305                 if (ptr[1]) {
 306                         *size = ptr[4] - ptr[9];
 307                 } else {
 308                         *size = 0;
 309                 }
 310
 311                 enc->ws->buffer_unmap(fb->cs_handle);
 312         }
 313         //dump_feedback(enc, fb);
 314         rvid_destroy_buffer(fb);
 315         FREE(fb);
 316 }
 317
 318 /**
 319  * flush any outstanding command buffers to the hardware
 320  */
 321 static void rvce_flush(struct pipe_video_codec *encoder)
 322 {
 323 }
 324
 325 static void rvce_cs_flush(void *ctx, unsigned flags,
 326                           struct pipe_fence_handle **fence)
 327 {
 328         // just ignored
 329 }
 330
 331 struct pipe_video_codec *rvce_create_encoder(struct pipe_context *context,
 332                                              const struct pipe_video_codec *templ,
 333                                              struct radeon_winsys* ws,
 334                                              rvce_get_buffer get_buffer)
 335 {
 336         struct r600_common_screen *rscreen = (struct r600_common_screen *)context->screen;
 337         struct rvce_encoder *enc;
 338         struct pipe_video_buffer *tmp_buf, templat = {};
 339         struct radeon_surface *tmp_surf;
 340         unsigned cpb_size;
 341
 342         if (!rscreen->info.vce_fw_version) {
 343                 RVID_ERR("Kernel doesn't supports VCE!\n");
 344                 return NULL;
 345
 346         } else if (!rvce_is_fw_version_supported(rscreen)) {
 347                 RVID_ERR("Unsupported VCE fw version loaded!\n");
 348                 return NULL;
 349         }
 350
 351         enc = CALLOC_STRUCT(rvce_encoder);
 352         if (!enc)
 353                 return NULL;
 354
 355         enc->base = *templ;
 356         enc->base.context = context;
 357
 358         enc->base.destroy = rvce_destroy;
 359         enc->base.begin_frame = rvce_begin_frame;
 360         enc->base.encode_bitstream = rvce_encode_bitstream;
 361         enc->base.end_frame = rvce_end_frame;
 362         enc->base.flush = rvce_flush;
 363         enc->base.get_feedback = rvce_get_feedback;
 364         enc->get_buffer = get_buffer;
 365
 366         enc->ws = ws;
 367         enc->cs = ws->cs_create(ws, RING_VCE, rvce_cs_flush, enc, NULL);
 368         if (!enc->cs) {
 369                 RVID_ERR("Can't get command submission context.\n");
 370                 goto error;
 371         }
 372
 373         templat.buffer_format = PIPE_FORMAT_NV12;
 374         templat.chroma_format = PIPE_VIDEO_CHROMA_FORMAT_420;
 375         templat.width = enc->base.width;
 376         templat.height = enc->base.height;
 377         templat.interlaced = false;
 378         if (!(tmp_buf = context->create_video_buffer(context, &templat))) {
 379                 RVID_ERR("Can't create video buffer.\n");
 380                 goto error;
 381         }
 382
 383         enc->cpb_num = get_cpb_num(enc);
 384         if (!enc->cpb_num)
 385                 goto error;
 386
 387         get_buffer(((struct vl_video_buffer *)tmp_buf)->resources[0], NULL, &tmp_surf);
 388         cpb_size = align(tmp_surf->level[0].pitch_bytes, 128);
 389         cpb_size = cpb_size * align(tmp_surf->npix_y, 16);
 390         cpb_size = cpb_size * 3 / 2;
 391         cpb_size = cpb_size * enc->cpb_num;
 392         tmp_buf->destroy(tmp_buf);
 393         if (!rvid_create_buffer(enc->ws, &enc->cpb, cpb_size, RADEON_DOMAIN_VRAM, 0)) {
 394                 RVID_ERR("Can't create CPB buffer.\n");
 395                 goto error;
 396         }
 397
 398         enc->cpb_array = CALLOC(enc->cpb_num, sizeof(struct rvce_cpb_slot));
 399         if (!enc->cpb_array)
 400                 goto error;
 401
 402         reset_cpb(enc);
 403
 404         radeon_vce_40_2_2_init(enc);
 405
 406         return &enc->base;
 407
 408 error:
 409         if (enc->cs)
 410                 enc->ws->cs_destroy(enc->cs);
 411
 412         rvid_destroy_buffer(&enc->cpb);
 413
 414         FREE(enc->cpb_array);
 415         FREE(enc);
 416         return NULL;
 417 }
 418
 419 /**
 420  * check if kernel has the right fw version loaded
 421  */
 422 bool rvce_is_fw_version_supported(struct r600_common_screen *rscreen)
 423 {
 424         return rscreen->info.vce_fw_version == ((40 << 24) | (2 << 16) | (2 << 8));
 425 }