FFmpeg
nvdec_vp9.c
Go to the documentation of this file.
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 "internal.h"
29 #include "vp9shared.h"
30 
31 static int nvdec_vp9_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
32 {
33  VP9SharedContext *h = avctx->priv_data;
34  const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
35 
37  CUVIDPICPARAMS *pp = &ctx->pic_params;
38  CUVIDVP9PICPARAMS *ppc = &pp->CodecSpecific.vp9;
39  FrameDecodeData *fdd;
40  NVDECFrame *cf;
41  AVFrame *cur_frame = h->frames[CUR_FRAME].tf.f;
42 
43  int ret, i;
44 
45  ret = ff_nvdec_start_frame(avctx, cur_frame);
46  if (ret < 0)
47  return ret;
48 
49  fdd = (FrameDecodeData*)cur_frame->private_ref->data;
50  cf = (NVDECFrame*)fdd->hwaccel_priv;
51 
52  *pp = (CUVIDPICPARAMS) {
53  .PicWidthInMbs = (cur_frame->width + 15) / 16,
54  .FrameHeightInMbs = (cur_frame->height + 15) / 16,
55  .CurrPicIdx = cf->idx,
56 
57  .CodecSpecific.vp9 = {
58  .width = cur_frame->width,
59  .height = cur_frame->height,
60 
61  .LastRefIdx = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[0]].f),
62  .GoldenRefIdx = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[1]].f),
63  .AltRefIdx = ff_nvdec_get_ref_idx(h->refs[h->h.refidx[2]].f),
64 
65  .profile = h->h.profile,
66  .frameContextIdx = h->h.framectxid,
67  .frameType = !h->h.keyframe,
68  .showFrame = !h->h.invisible,
69  .errorResilient = h->h.errorres,
70  .frameParallelDecoding = h->h.parallelmode,
71  .subSamplingX = pixdesc->log2_chroma_w,
72  .subSamplingY = pixdesc->log2_chroma_h,
73  .intraOnly = h->h.intraonly,
74  .allow_high_precision_mv = h->h.keyframe ? 0 : h->h.highprecisionmvs,
75  .refreshEntropyProbs = h->h.refreshctx,
76 
77  .bitDepthMinus8Luma = pixdesc->comp[0].depth - 8,
78  .bitDepthMinus8Chroma = pixdesc->comp[1].depth - 8,
79 
80  .loopFilterLevel = h->h.filter.level,
81  .loopFilterSharpness = h->h.filter.sharpness,
82  .modeRefLfEnabled = h->h.lf_delta.enabled,
83 
84  .log2_tile_columns = h->h.tiling.log2_tile_cols,
85  .log2_tile_rows = h->h.tiling.log2_tile_rows,
86 
87  .segmentEnabled = h->h.segmentation.enabled,
88  .segmentMapUpdate = h->h.segmentation.update_map,
89  .segmentMapTemporalUpdate = h->h.segmentation.temporal,
90  .segmentFeatureMode = h->h.segmentation.absolute_vals,
91 
92  .qpYAc = h->h.yac_qi,
93  .qpYDc = h->h.ydc_qdelta,
94  .qpChDc = h->h.uvdc_qdelta,
95  .qpChAc = h->h.uvac_qdelta,
96 
97  .resetFrameContext = h->h.resetctx,
98  .mcomp_filter_type = h->h.filtermode ^ (h->h.filtermode <= 1),
99 
100  .frameTagSize = h->h.uncompressed_header_size,
101  .offsetToDctParts = h->h.compressed_header_size,
102 
103  .refFrameSignBias[0] = 0,
104  }
105  };
106 
107  for (i = 0; i < 2; i++)
108  ppc->mbModeLfDelta[i] = h->h.lf_delta.mode[i];
109 
110  for (i = 0; i < 4; i++)
111  ppc->mbRefLfDelta[i] = h->h.lf_delta.ref[i];
112 
113  for (i = 0; i < 7; i++)
114  ppc->mb_segment_tree_probs[i] = h->h.segmentation.prob[i];
115 
116  for (i = 0; i < 3; i++) {
117  ppc->activeRefIdx[i] = h->h.refidx[i];
118  ppc->segment_pred_probs[i] = h->h.segmentation.pred_prob[i];
119  ppc->refFrameSignBias[i + 1] = h->h.signbias[i];
120  }
121 
122  for (i = 0; i < 8; i++) {
123  ppc->segmentFeatureEnable[i][0] = h->h.segmentation.feat[i].q_enabled;
124  ppc->segmentFeatureEnable[i][1] = h->h.segmentation.feat[i].lf_enabled;
125  ppc->segmentFeatureEnable[i][2] = h->h.segmentation.feat[i].ref_enabled;
126  ppc->segmentFeatureEnable[i][3] = h->h.segmentation.feat[i].skip_enabled;
127 
128  ppc->segmentFeatureData[i][0] = h->h.segmentation.feat[i].q_val;
129  ppc->segmentFeatureData[i][1] = h->h.segmentation.feat[i].lf_val;
130  ppc->segmentFeatureData[i][2] = h->h.segmentation.feat[i].ref_val;
131  ppc->segmentFeatureData[i][3] = 0;
132  }
133 
134  switch (avctx->colorspace) {
135  default:
137  ppc->colorSpace = 0;
138  break;
139  case AVCOL_SPC_BT470BG:
140  ppc->colorSpace = 1;
141  break;
142  case AVCOL_SPC_BT709:
143  ppc->colorSpace = 2;
144  break;
145  case AVCOL_SPC_SMPTE170M:
146  ppc->colorSpace = 3;
147  break;
148  case AVCOL_SPC_SMPTE240M:
149  ppc->colorSpace = 4;
150  break;
152  ppc->colorSpace = 5;
153  break;
154  case AVCOL_SPC_RESERVED:
155  ppc->colorSpace = 6;
156  break;
157  case AVCOL_SPC_RGB:
158  ppc->colorSpace = 7;
159  break;
160  }
161 
162  return 0;
163 }
164 
166  AVBufferRef *hw_frames_ctx)
167 {
168  // VP9 uses a fixed size pool of 8 possible reference frames
169  return ff_nvdec_frame_params(avctx, hw_frames_ctx, 8, 0);
170 }
171 
173  .name = "vp9_nvdec",
174  .type = AVMEDIA_TYPE_VIDEO,
175  .id = AV_CODEC_ID_VP9,
176  .pix_fmt = AV_PIX_FMT_CUDA,
177  .start_frame = nvdec_vp9_start_frame,
178  .end_frame = ff_nvdec_simple_end_frame,
179  .decode_slice = ff_nvdec_simple_decode_slice,
180  .frame_params = nvdec_vp9_frame_params,
181  .init = ff_nvdec_decode_init,
182  .uninit = ff_nvdec_decode_uninit,
183  .priv_data_size = sizeof(NVDECContext),
184 };
AV_PIX_FMT_CUDA
@ AV_PIX_FMT_CUDA
HW acceleration through CUDA.
Definition: pixfmt.h:235
AVCodecContext::colorspace
enum AVColorSpace colorspace
YUV colorspace type.
Definition: avcodec.h:1164
nvdec_vp9_start_frame
static int nvdec_vp9_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
Definition: nvdec_vp9.c:31
av_pix_fmt_desc_get
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
Definition: pixdesc.c:2573
AVBufferRef::data
uint8_t * data
The data buffer.
Definition: buffer.h:92
ff_nvdec_get_ref_idx
int ff_nvdec_get_ref_idx(AVFrame *frame)
Definition: nvdec.c:747
FrameDecodeData
This struct stores per-frame lavc-internal data and is attached to it via private_ref.
Definition: decode.h:34
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:318
pixdesc.h
AVFrame::width
int width
Definition: frame.h:376
internal.h
AVComponentDescriptor::depth
int depth
Number of bits in the component.
Definition: pixdesc.h:58
AVCOL_SPC_RGB
@ AVCOL_SPC_RGB
order of coefficients is actually GBR, also IEC 61966-2-1 (sRGB)
Definition: pixfmt.h:513
AVCOL_SPC_BT470BG
@ AVCOL_SPC_BT470BG
also ITU-R BT601-6 625 / ITU-R BT1358 625 / ITU-R BT1700 625 PAL & SECAM / IEC 61966-2-4 xvYCC601
Definition: pixfmt.h:518
AVCOL_SPC_RESERVED
@ AVCOL_SPC_RESERVED
Definition: pixfmt.h:516
AVHWAccel
Definition: avcodec.h:2438
ff_nvdec_start_frame
int ff_nvdec_start_frame(AVCodecContext *avctx, AVFrame *frame)
Definition: nvdec.c:559
vp9shared.h
VP9SharedContext
Definition: vp9shared.h:159
AVCOL_SPC_SMPTE170M
@ AVCOL_SPC_SMPTE170M
also ITU-R BT601-6 525 / ITU-R BT1358 525 / ITU-R BT1700 NTSC
Definition: pixfmt.h:519
AV_CODEC_ID_VP9
@ AV_CODEC_ID_VP9
Definition: codec_id.h:217
NVDECFrame
Definition: nvdec.h:44
ctx
AVFormatContext * ctx
Definition: movenc.c:48
decode.h
AVPixFmtDescriptor::log2_chroma_w
uint8_t log2_chroma_w
Amount to shift the luma width right to find the chroma width.
Definition: pixdesc.h:92
ff_vp9_nvdec_hwaccel
const AVHWAccel ff_vp9_nvdec_hwaccel
Definition: nvdec_vp9.c:172
ff_nvdec_simple_end_frame
int ff_nvdec_simple_end_frame(AVCodecContext *avctx)
Definition: nvdec.c:665
ff_nvdec_decode_init
int ff_nvdec_decode_init(AVCodecContext *avctx)
Definition: nvdec.c:330
AVCodecContext::internal
struct AVCodecInternal * internal
Private context used for internal data.
Definition: avcodec.h:571
nvdec_vp9_frame_params
static int nvdec_vp9_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx)
Definition: nvdec_vp9.c:165
for
for(j=16;j >0;--j)
Definition: h264pred_template.c:469
AVCodecInternal::hwaccel_priv_data
void * hwaccel_priv_data
hwaccel-specific private data
Definition: internal.h:180
NVDECFrame::idx
unsigned int idx
Definition: nvdec.h:45
size
int size
Definition: twinvq_data.h:10344
nvdec.h
ff_nvdec_decode_uninit
int ff_nvdec_decode_uninit(AVCodecContext *avctx)
Definition: nvdec.c:262
AVFrame::private_ref
AVBufferRef * private_ref
AVBufferRef for internal use by a single libav* library.
Definition: frame.h:697
AVHWAccel::name
const char * name
Name of the hardware accelerated codec.
Definition: avcodec.h:2444
AVCOL_SPC_SMPTE240M
@ AVCOL_SPC_SMPTE240M
functionally identical to above
Definition: pixfmt.h:520
i
int i
Definition: input.c:407
AVCOL_SPC_BT2020_NCL
@ AVCOL_SPC_BT2020_NCL
ITU-R BT2020 non-constant luminance system.
Definition: pixfmt.h:523
uint8_t
uint8_t
Definition: audio_convert.c:194
AVCOL_SPC_UNSPECIFIED
@ AVCOL_SPC_UNSPECIFIED
Definition: pixfmt.h:515
avcodec.h
ret
ret
Definition: filter_design.txt:187
AVCodecContext
main external API structure.
Definition: avcodec.h:536
AVFrame::height
int height
Definition: frame.h:376
ff_nvdec_simple_decode_slice
int ff_nvdec_simple_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
Definition: nvdec.c:673
buffer
the frame and frame reference mechanism is intended to as much as expensive copies of that data while still allowing the filters to produce correct results The data is stored in buffers represented by AVFrame structures Several references can point to the same frame buffer
Definition: filter_design.txt:49
ff_nvdec_frame_params
int ff_nvdec_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx, int dpb_size, int supports_444)
Definition: nvdec.c:695
AVPixFmtDescriptor::comp
AVComponentDescriptor comp[4]
Parameters that describe how pixels are packed.
Definition: pixdesc.h:117
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
AVBufferRef
A reference to a data buffer.
Definition: buffer.h:84
CUR_FRAME
#define CUR_FRAME
Definition: vp9shared.h:163
AVPixFmtDescriptor
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
Definition: pixdesc.h:81
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:563
FrameDecodeData::hwaccel_priv
void * hwaccel_priv
Per-frame private data for hwaccels.
Definition: decode.h:52
h
h
Definition: vp9dsp_template.c:2038
AVCodecContext::sw_pix_fmt
enum AVPixelFormat sw_pix_fmt
Nominal unaccelerated pixel format, see AV_PIX_FMT_xxx.
Definition: avcodec.h:2078
NVDECContext
Definition: nvdec.h:52
AVCOL_SPC_BT709
@ AVCOL_SPC_BT709
also ITU-R BT1361 / IEC 61966-2-4 xvYCC709 / SMPTE RP177 Annex B
Definition: pixfmt.h:514
AVPixFmtDescriptor::log2_chroma_h
uint8_t log2_chroma_h
Amount to shift the luma height right to find the chroma height.
Definition: pixdesc.h:101