git.sesse.net Git - ffmpeg/blob - libavcodec/vdpau.c

   1 /*
   2  * Video Decode and Presentation API for UNIX (VDPAU) is used for
   3  * HW decode acceleration for MPEG-1/2, MPEG-4 ASP, H.264 and VC-1.
   4  *
   5  * Copyright (c) 2008 NVIDIA
   6  *
   7  * This file is part of Libav.
   8  *
   9  * Libav is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * Libav is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with Libav; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include <limits.h>
  25 #include "avcodec.h"
  26 #include "h264.h"
  27 #include "vc1.h"
  28
  29 #undef NDEBUG
  30 #include <assert.h>
  31
  32 #include "vdpau.h"
  33 #include "vdpau_internal.h"
  34
  35 /**
  36  * @addtogroup VDPAU_Decoding
  37  *
  38  * @{
  39  */
  40
  41 int ff_vdpau_common_start_frame(AVCodecContext *avctx,
  42                                 av_unused const uint8_t *buffer,
  43                                 av_unused uint32_t size)
  44 {
  45     AVVDPAUContext *hwctx = avctx->hwaccel_context;
  46
  47     hwctx->bitstream_buffers_used = 0;
  48     return 0;
  49 }
  50
  51 int ff_vdpau_common_end_frame(AVCodecContext *avctx)
  52 {
  53     MpegEncContext * const s = avctx->priv_data;
  54     AVVDPAUContext *hwctx = avctx->hwaccel_context;
  55
  56     if (hwctx->bitstream_buffers_used) {
  57         VdpVideoSurface surf = ff_vdpau_get_surface_id(s->current_picture_ptr);
  58
  59         hwctx->render(hwctx->decoder, surf, (void *)&hwctx->info,
  60                       hwctx->bitstream_buffers_used, hwctx->bitstream_buffers);
  61
  62         ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
  63         hwctx->bitstream_buffers_used = 0;
  64     }
  65     return 0;
  66 }
  67
  68 int ff_vdpau_add_buffer(AVCodecContext *avctx,
  69                         const uint8_t *buf, uint32_t size)
  70 {
  71     AVVDPAUContext *hwctx = avctx->hwaccel_context;
  72     VdpBitstreamBuffer *buffers = hwctx->bitstream_buffers;
  73
  74     buffers = av_fast_realloc(buffers, &hwctx->bitstream_buffers_allocated,
  75                               (hwctx->bitstream_buffers_used + 1) * sizeof(*buffers));
  76     if (!buffers)
  77         return AVERROR(ENOMEM);
  78
  79     hwctx->bitstream_buffers = buffers;
  80     buffers += hwctx->bitstream_buffers_used++;
  81
  82     buffers->struct_version  = VDP_BITSTREAM_BUFFER_VERSION;
  83     buffers->bitstream       = buf;
  84     buffers->bitstream_bytes = size;
  85     return 0;
  86 }
  87
  88 /* Obsolete non-hwaccel VDPAU support below... */
  89
  90 void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
  91 {
  92     H264Context *h = s->avctx->priv_data;
  93     struct vdpau_render_state *render, *render_ref;
  94     VdpReferenceFrameH264 *rf, *rf2;
  95     Picture *pic;
  96     int i, list, pic_frame_idx;
  97
  98     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
  99     assert(render);
 100
 101     rf = &render->info.h264.referenceFrames[0];
 102 #define H264_RF_COUNT FF_ARRAY_ELEMS(render->info.h264.referenceFrames)
 103
 104     for (list = 0; list < 2; ++list) {
 105         Picture **lp = list ? h->long_ref : h->short_ref;
 106         int ls = list ? 16 : h->short_ref_count;
 107
 108         for (i = 0; i < ls; ++i) {
 109             pic = lp[i];
 110             if (!pic || !pic->f.reference)
 111                 continue;
 112             pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
 113
 114             render_ref = (struct vdpau_render_state *)pic->f.data[0];
 115             assert(render_ref);
 116
 117             rf2 = &render->info.h264.referenceFrames[0];
 118             while (rf2 != rf) {
 119                 if (
 120                     (rf2->surface == render_ref->surface)
 121                     && (rf2->is_long_term == pic->long_ref)
 122                     && (rf2->frame_idx == pic_frame_idx)
 123                 )
 124                     break;
 125                 ++rf2;
 126             }
 127             if (rf2 != rf) {
 128                 rf2->top_is_reference    |= (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
 129                 rf2->bottom_is_reference |= (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
 130                 continue;
 131             }
 132
 133             if (rf >= &render->info.h264.referenceFrames[H264_RF_COUNT])
 134                 continue;
 135
 136             rf->surface             = render_ref->surface;
 137             rf->is_long_term        = pic->long_ref;
 138             rf->top_is_reference    = (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
 139             rf->bottom_is_reference = (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
 140             rf->field_order_cnt[0]  = pic->field_poc[0];
 141             rf->field_order_cnt[1]  = pic->field_poc[1];
 142             rf->frame_idx           = pic_frame_idx;
 143
 144             ++rf;
 145         }
 146     }
 147
 148     for (; rf < &render->info.h264.referenceFrames[H264_RF_COUNT]; ++rf) {
 149         rf->surface             = VDP_INVALID_HANDLE;
 150         rf->is_long_term        = 0;
 151         rf->top_is_reference    = 0;
 152         rf->bottom_is_reference = 0;
 153         rf->field_order_cnt[0]  = 0;
 154         rf->field_order_cnt[1]  = 0;
 155         rf->frame_idx           = 0;
 156     }
 157 }
 158
 159 void ff_vdpau_add_data_chunk(MpegEncContext *s,
 160                              const uint8_t *buf, int buf_size)
 161 {
 162     struct vdpau_render_state *render;
 163
 164     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 165     assert(render);
 166
 167     render->bitstream_buffers= av_fast_realloc(
 168         render->bitstream_buffers,
 169         &render->bitstream_buffers_allocated,
 170         sizeof(*render->bitstream_buffers)*(render->bitstream_buffers_used + 1)
 171     );
 172
 173     render->bitstream_buffers[render->bitstream_buffers_used].struct_version  = VDP_BITSTREAM_BUFFER_VERSION;
 174     render->bitstream_buffers[render->bitstream_buffers_used].bitstream       = buf;
 175     render->bitstream_buffers[render->bitstream_buffers_used].bitstream_bytes = buf_size;
 176     render->bitstream_buffers_used++;
 177 }
 178
 179 void ff_vdpau_h264_picture_start(MpegEncContext *s)
 180 {
 181     H264Context *h = s->avctx->priv_data;
 182     struct vdpau_render_state *render;
 183     int i;
 184
 185     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 186     assert(render);
 187
 188     for (i = 0; i < 2; ++i) {
 189         int foc = s->current_picture_ptr->field_poc[i];
 190         if (foc == INT_MAX)
 191             foc = 0;
 192         render->info.h264.field_order_cnt[i] = foc;
 193     }
 194
 195     render->info.h264.frame_num = h->frame_num;
 196 }
 197
 198 void ff_vdpau_h264_picture_complete(MpegEncContext *s)
 199 {
 200     H264Context *h = s->avctx->priv_data;
 201     struct vdpau_render_state *render;
 202
 203     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 204     assert(render);
 205
 206     render->info.h264.slice_count = h->slice_num;
 207     if (render->info.h264.slice_count < 1)
 208         return;
 209
 210     render->info.h264.is_reference                           = (s->current_picture_ptr->f.reference & 3) ? VDP_TRUE : VDP_FALSE;
 211     render->info.h264.field_pic_flag                         = s->picture_structure != PICT_FRAME;
 212     render->info.h264.bottom_field_flag                      = s->picture_structure == PICT_BOTTOM_FIELD;
 213     render->info.h264.num_ref_frames                         = h->sps.ref_frame_count;
 214     render->info.h264.mb_adaptive_frame_field_flag           = h->sps.mb_aff && !render->info.h264.field_pic_flag;
 215     render->info.h264.constrained_intra_pred_flag            = h->pps.constrained_intra_pred;
 216     render->info.h264.weighted_pred_flag                     = h->pps.weighted_pred;
 217     render->info.h264.weighted_bipred_idc                    = h->pps.weighted_bipred_idc;
 218     render->info.h264.frame_mbs_only_flag                    = h->sps.frame_mbs_only_flag;
 219     render->info.h264.transform_8x8_mode_flag                = h->pps.transform_8x8_mode;
 220     render->info.h264.chroma_qp_index_offset                 = h->pps.chroma_qp_index_offset[0];
 221     render->info.h264.second_chroma_qp_index_offset          = h->pps.chroma_qp_index_offset[1];
 222     render->info.h264.pic_init_qp_minus26                    = h->pps.init_qp - 26;
 223     render->info.h264.num_ref_idx_l0_active_minus1           = h->pps.ref_count[0] - 1;
 224     render->info.h264.num_ref_idx_l1_active_minus1           = h->pps.ref_count[1] - 1;
 225     render->info.h264.log2_max_frame_num_minus4              = h->sps.log2_max_frame_num - 4;
 226     render->info.h264.pic_order_cnt_type                     = h->sps.poc_type;
 227     render->info.h264.log2_max_pic_order_cnt_lsb_minus4      = h->sps.poc_type ? 0 : h->sps.log2_max_poc_lsb - 4;
 228     render->info.h264.delta_pic_order_always_zero_flag       = h->sps.delta_pic_order_always_zero_flag;
 229     render->info.h264.direct_8x8_inference_flag              = h->sps.direct_8x8_inference_flag;
 230     render->info.h264.entropy_coding_mode_flag               = h->pps.cabac;
 231     render->info.h264.pic_order_present_flag                 = h->pps.pic_order_present;
 232     render->info.h264.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
 233     render->info.h264.redundant_pic_cnt_present_flag         = h->pps.redundant_pic_cnt_present;
 234     memcpy(render->info.h264.scaling_lists_4x4, h->pps.scaling_matrix4, sizeof(render->info.h264.scaling_lists_4x4));
 235     memcpy(render->info.h264.scaling_lists_8x8[0], h->pps.scaling_matrix8[0], sizeof(render->info.h264.scaling_lists_8x8[0]));
 236     memcpy(render->info.h264.scaling_lists_8x8[1], h->pps.scaling_matrix8[3], sizeof(render->info.h264.scaling_lists_8x8[0]));
 237
 238     ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 239     render->bitstream_buffers_used = 0;
 240 }
 241
 242 void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
 243                                     int buf_size, int slice_count)
 244 {
 245     struct vdpau_render_state *render, *last, *next;
 246     int i;
 247
 248     if (!s->current_picture_ptr) return;
 249
 250     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 251     assert(render);
 252
 253     /* fill VdpPictureInfoMPEG1Or2 struct */
 254     render->info.mpeg.picture_structure          = s->picture_structure;
 255     render->info.mpeg.picture_coding_type        = s->pict_type;
 256     render->info.mpeg.intra_dc_precision         = s->intra_dc_precision;
 257     render->info.mpeg.frame_pred_frame_dct       = s->frame_pred_frame_dct;
 258     render->info.mpeg.concealment_motion_vectors = s->concealment_motion_vectors;
 259     render->info.mpeg.intra_vlc_format           = s->intra_vlc_format;
 260     render->info.mpeg.alternate_scan             = s->alternate_scan;
 261     render->info.mpeg.q_scale_type               = s->q_scale_type;
 262     render->info.mpeg.top_field_first            = s->top_field_first;
 263     render->info.mpeg.full_pel_forward_vector    = s->full_pel[0]; // MPEG-1 only.  Set 0 for MPEG-2
 264     render->info.mpeg.full_pel_backward_vector   = s->full_pel[1]; // MPEG-1 only.  Set 0 for MPEG-2
 265     render->info.mpeg.f_code[0][0]               = s->mpeg_f_code[0][0]; // For MPEG-1 fill both horiz. & vert.
 266     render->info.mpeg.f_code[0][1]               = s->mpeg_f_code[0][1];
 267     render->info.mpeg.f_code[1][0]               = s->mpeg_f_code[1][0];
 268     render->info.mpeg.f_code[1][1]               = s->mpeg_f_code[1][1];
 269     for (i = 0; i < 64; ++i) {
 270         render->info.mpeg.intra_quantizer_matrix[i]     = s->intra_matrix[i];
 271         render->info.mpeg.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
 272     }
 273
 274     render->info.mpeg.forward_reference          = VDP_INVALID_HANDLE;
 275     render->info.mpeg.backward_reference         = VDP_INVALID_HANDLE;
 276
 277     switch(s->pict_type){
 278     case  AV_PICTURE_TYPE_B:
 279         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
 280         assert(next);
 281         render->info.mpeg.backward_reference     = next->surface;
 282         // no return here, going to set forward prediction
 283     case  AV_PICTURE_TYPE_P:
 284         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
 285         if (!last) // FIXME: Does this test make sense?
 286             last = render; // predict second field from the first
 287         render->info.mpeg.forward_reference      = last->surface;
 288     }
 289
 290     ff_vdpau_add_data_chunk(s, buf, buf_size);
 291
 292     render->info.mpeg.slice_count                = slice_count;
 293
 294     if (slice_count)
 295         ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 296     render->bitstream_buffers_used               = 0;
 297 }
 298
 299 void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
 300                                  int buf_size)
 301 {
 302     VC1Context *v = s->avctx->priv_data;
 303     struct vdpau_render_state *render, *last, *next;
 304
 305     render = (struct vdpau_render_state *)s->current_picture.f.data[0];
 306     assert(render);
 307
 308     /*  fill LvPictureInfoVC1 struct */
 309     render->info.vc1.frame_coding_mode  = v->fcm;
 310     render->info.vc1.postprocflag       = v->postprocflag;
 311     render->info.vc1.pulldown           = v->broadcast;
 312     render->info.vc1.interlace          = v->interlace;
 313     render->info.vc1.tfcntrflag         = v->tfcntrflag;
 314     render->info.vc1.finterpflag        = v->finterpflag;
 315     render->info.vc1.psf                = v->psf;
 316     render->info.vc1.dquant             = v->dquant;
 317     render->info.vc1.panscan_flag       = v->panscanflag;
 318     render->info.vc1.refdist_flag       = v->refdist_flag;
 319     render->info.vc1.quantizer          = v->quantizer_mode;
 320     render->info.vc1.extended_mv        = v->extended_mv;
 321     render->info.vc1.extended_dmv       = v->extended_dmv;
 322     render->info.vc1.overlap            = v->overlap;
 323     render->info.vc1.vstransform        = v->vstransform;
 324     render->info.vc1.loopfilter         = v->s.loop_filter;
 325     render->info.vc1.fastuvmc           = v->fastuvmc;
 326     render->info.vc1.range_mapy_flag    = v->range_mapy_flag;
 327     render->info.vc1.range_mapy         = v->range_mapy;
 328     render->info.vc1.range_mapuv_flag   = v->range_mapuv_flag;
 329     render->info.vc1.range_mapuv        = v->range_mapuv;
 330     /* Specific to simple/main profile only */
 331     render->info.vc1.multires           = v->multires;
 332     render->info.vc1.syncmarker         = v->s.resync_marker;
 333     render->info.vc1.rangered           = v->rangered | (v->rangeredfrm << 1);
 334     render->info.vc1.maxbframes         = v->s.max_b_frames;
 335
 336     render->info.vc1.deblockEnable      = v->postprocflag & 1;
 337     render->info.vc1.pquant             = v->pq;
 338
 339     render->info.vc1.forward_reference  = VDP_INVALID_HANDLE;
 340     render->info.vc1.backward_reference = VDP_INVALID_HANDLE;
 341
 342     if (v->bi_type)
 343         render->info.vc1.picture_type = 4;
 344     else
 345         render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
 346
 347     switch(s->pict_type){
 348     case  AV_PICTURE_TYPE_B:
 349         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
 350         assert(next);
 351         render->info.vc1.backward_reference = next->surface;
 352         // no break here, going to set forward prediction
 353     case  AV_PICTURE_TYPE_P:
 354         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
 355         if (!last) // FIXME: Does this test make sense?
 356             last = render; // predict second field from the first
 357         render->info.vc1.forward_reference = last->surface;
 358     }
 359
 360     ff_vdpau_add_data_chunk(s, buf, buf_size);
 361
 362     render->info.vc1.slice_count          = 1;
 363
 364     ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 365     render->bitstream_buffers_used        = 0;
 366 }
 367
 368 void ff_vdpau_mpeg4_decode_picture(MpegEncContext *s, const uint8_t *buf,
 369                                    int buf_size)
 370 {
 371     struct vdpau_render_state *render, *last, *next;
 372     int i;
 373
 374     if (!s->current_picture_ptr) return;
 375
 376     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 377     assert(render);
 378
 379     /* fill VdpPictureInfoMPEG4Part2 struct */
 380     render->info.mpeg4.trd[0]                            = s->pp_time;
 381     render->info.mpeg4.trb[0]                            = s->pb_time;
 382     render->info.mpeg4.trd[1]                            = s->pp_field_time >> 1;
 383     render->info.mpeg4.trb[1]                            = s->pb_field_time >> 1;
 384     render->info.mpeg4.vop_time_increment_resolution     = s->avctx->time_base.den;
 385     render->info.mpeg4.vop_coding_type                   = 0;
 386     render->info.mpeg4.vop_fcode_forward                 = s->f_code;
 387     render->info.mpeg4.vop_fcode_backward                = s->b_code;
 388     render->info.mpeg4.resync_marker_disable             = !s->resync_marker;
 389     render->info.mpeg4.interlaced                        = !s->progressive_sequence;
 390     render->info.mpeg4.quant_type                        = s->mpeg_quant;
 391     render->info.mpeg4.quarter_sample                    = s->quarter_sample;
 392     render->info.mpeg4.short_video_header                = s->avctx->codec->id == AV_CODEC_ID_H263;
 393     render->info.mpeg4.rounding_control                  = s->no_rounding;
 394     render->info.mpeg4.alternate_vertical_scan_flag      = s->alternate_scan;
 395     render->info.mpeg4.top_field_first                   = s->top_field_first;
 396     for (i = 0; i < 64; ++i) {
 397         render->info.mpeg4.intra_quantizer_matrix[i]     = s->intra_matrix[i];
 398         render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
 399     }
 400     render->info.mpeg4.forward_reference                 = VDP_INVALID_HANDLE;
 401     render->info.mpeg4.backward_reference                = VDP_INVALID_HANDLE;
 402
 403     switch (s->pict_type) {
 404     case AV_PICTURE_TYPE_B:
 405         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
 406         assert(next);
 407         render->info.mpeg4.backward_reference     = next->surface;
 408         render->info.mpeg4.vop_coding_type        = 2;
 409         // no break here, going to set forward prediction
 410     case AV_PICTURE_TYPE_P:
 411         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
 412         assert(last);
 413         render->info.mpeg4.forward_reference      = last->surface;
 414     }
 415
 416     ff_vdpau_add_data_chunk(s, buf, buf_size);
 417
 418     ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 419     render->bitstream_buffers_used = 0;
 420 }
 421
 422 /* @}*/