libavcodec/nvdec_vp9.c

   1 /*
   2  * VP9 HW decode acceleration through NVDEC
   3  *
   4  * Copyright (c) 2016 Timo Rothenpieler
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include "libavutil/pixdesc.h"
  24
  25 #include "avcodec.h"
  26 #include "nvdec.h"
  27 #include "decode.h"
  28 #include "hwaccel_internal.h"
  29 #include "internal.h"
  30 #include "vp9shared.h"
  31
  32 static int nvdec_vp9_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
  33 {
  34     VP9SharedContext *h = avctx->priv_data;
  35     const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
  36
  37     NVDECContext      *ctx = avctx->internal->hwaccel_priv_data;
  38     CUVIDPICPARAMS     *pp = &ctx->pic_params;
  39     CUVIDVP9PICPARAMS *ppc = &pp->CodecSpecific.vp9;
  40     FrameDecodeData *fdd;
  41     NVDECFrame *cf;
  42     AVFrame *cur_frame = h->frames[CUR_FRAME].tf.f;
  43
  44     int ret, i;
  45
  46     ret = ff_nvdec_start_frame(avctx, cur_frame);
  47     if (ret < 0)
  48         return ret;
  49
  50     fdd = (FrameDecodeData*)cur_frame->private_ref->data;
  51     cf  = (NVDECFrame*)fdd->hwaccel_priv;
  52
  53     *pp = (CUVIDPICPARAMS) {
  54         .PicWidthInMbs     = (cur_frame->width  + 15) / 16,
  55         .FrameHeightInMbs  = (cur_frame->height + 15) / 16,
  56         .CurrPicIdx        = cf->idx,
  57
  58         .CodecSpecific.vp9 = {
  59             .width                    = cur_frame->width,
  60             .height                   = cur_frame->height,
  61
  62             .LastRefIdx               = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[0]].f),
  63             .GoldenRefIdx             = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[1]].f),
  64             .AltRefIdx                = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[2]].f),
  65
  66             .profile                  = h->h.profile,
  67             .frameContextIdx          = h->h.framectxid,
  68             .frameType                = !h->h.keyframe,
  69             .showFrame                = !h->h.invisible,
  70             .errorResilient           = h->h.errorres,
  71             .frameParallelDecoding    = h->h.parallelmode,
  72             .subSamplingX             = pixdesc->log2_chroma_w,
  73             .subSamplingY             = pixdesc->log2_chroma_h,
  74             .intraOnly                = h->h.intraonly,
  75             .allow_high_precision_mv  = h->h.keyframe ? 0 : h->h.highprecisionmvs,
  76             .refreshEntropyProbs      = h->h.refreshctx,
  77
  78             .bitDepthMinus8Luma       = pixdesc->comp[0].depth - 8,
  79             .bitDepthMinus8Chroma     = pixdesc->comp[1].depth - 8,
  80
  81             .loopFilterLevel          = h->h.filter.level,
  82             .loopFilterSharpness      = h->h.filter.sharpness,
  83             .modeRefLfEnabled         = h->h.lf_delta.enabled,
  84
  85             .log2_tile_columns        = h->h.tiling.log2_tile_cols,
  86             .log2_tile_rows           = h->h.tiling.log2_tile_rows,
  87
  88             .segmentEnabled           = h->h.segmentation.enabled,
  89             .segmentMapUpdate         = h->h.segmentation.update_map,
  90             .segmentMapTemporalUpdate = h->h.segmentation.temporal,
  91             .segmentFeatureMode       = h->h.segmentation.absolute_vals,
  92
  93             .qpYAc                    = h->h.yac_qi,
  94             .qpYDc                    = h->h.ydc_qdelta,
  95             .qpChDc                   = h->h.uvdc_qdelta,
  96             .qpChAc                   = h->h.uvac_qdelta,
  97
  98             .resetFrameContext        = h->h.resetctx,
  99             .mcomp_filter_type        = h->h.filtermode ^ (h->h.filtermode <= 1),
 100
 101             .frameTagSize             = h->h.uncompressed_header_size,
 102             .offsetToDctParts         = h->h.compressed_header_size,
 103
 104             .refFrameSignBias[0]      = 0,
 105         }
 106     };
 107
 108     for (i = 0; i < 2; i++)
 109         ppc->mbModeLfDelta[i] = h->h.lf_delta.mode[i];
 110
 111     for (i = 0; i < 4; i++)
 112         ppc->mbRefLfDelta[i] = h->h.lf_delta.ref[i];
 113
 114     for (i = 0; i < 7; i++)
 115         ppc->mb_segment_tree_probs[i] = h->h.segmentation.prob[i];
 116
 117     for (i = 0; i < 3; i++) {
 118         ppc->activeRefIdx[i] = h->h.refidx[i];
 119         ppc->segment_pred_probs[i] = h->h.segmentation.pred_prob[i];
 120         ppc->refFrameSignBias[i + 1] = h->h.signbias[i];
 121     }
 122
 123     for (i = 0; i < 8; i++) {
 124         ppc->segmentFeatureEnable[i][0] = h->h.segmentation.feat[i].q_enabled;
 125         ppc->segmentFeatureEnable[i][1] = h->h.segmentation.feat[i].lf_enabled;
 126         ppc->segmentFeatureEnable[i][2] = h->h.segmentation.feat[i].ref_enabled;
 127         ppc->segmentFeatureEnable[i][3] = h->h.segmentation.feat[i].skip_enabled;
 128
 129         ppc->segmentFeatureData[i][0] = h->h.segmentation.feat[i].q_val;
 130         ppc->segmentFeatureData[i][1] = h->h.segmentation.feat[i].lf_val;
 131         ppc->segmentFeatureData[i][2] = h->h.segmentation.feat[i].ref_val;
 132         ppc->segmentFeatureData[i][3] = 0;
 133     }
 134
 135     switch (avctx->colorspace) {
 136     default:
 137     case AVCOL_SPC_UNSPECIFIED:
 138         ppc->colorSpace = 0;
 139         break;
 140     case AVCOL_SPC_BT470BG:
 141         ppc->colorSpace = 1;
 142         break;
 143     case AVCOL_SPC_BT709:
 144         ppc->colorSpace = 2;
 145         break;
 146     case AVCOL_SPC_SMPTE170M:
 147         ppc->colorSpace = 3;
 148         break;
 149     case AVCOL_SPC_SMPTE240M:
 150         ppc->colorSpace = 4;
 151         break;
 152     case AVCOL_SPC_BT2020_NCL:
 153         ppc->colorSpace = 5;
 154         break;
 155     case AVCOL_SPC_RESERVED:
 156         ppc->colorSpace = 6;
 157         break;
 158     case AVCOL_SPC_RGB:
 159         ppc->colorSpace = 7;
 160         break;
 161     }
 162
 163     return 0;
 164 }
 165
 166 static int nvdec_vp9_frame_params(AVCodecContext *avctx,
 167                                   AVBufferRef *hw_frames_ctx)
 168 {
 169     // VP9 uses a fixed size pool of 8 possible reference frames
 170     return ff_nvdec_frame_params(avctx, hw_frames_ctx, 8, 0);
 171 }
 172
 173 const FFHWAccel ff_vp9_nvdec_hwaccel = {
 174     .p.name               = "vp9_nvdec",
 175     .p.type               = AVMEDIA_TYPE_VIDEO,
 176     .p.id                 = AV_CODEC_ID_VP9,
 177     .p.pix_fmt            = AV_PIX_FMT_CUDA,
 178     .start_frame          = nvdec_vp9_start_frame,
 179     .end_frame            = ff_nvdec_simple_end_frame,
 180     .decode_slice         = ff_nvdec_simple_decode_slice,
 181     .frame_params         = nvdec_vp9_frame_params,
 182     .init                 = ff_nvdec_decode_init,
 183     .uninit               = ff_nvdec_decode_uninit,
 184     .priv_data_size       = sizeof(NVDECContext),
 185 };