• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #ifndef AVCODEC_VAAPI_ENCODE_H
20 #define AVCODEC_VAAPI_ENCODE_H
21 
22 #include <stdint.h>
23 
24 #include <va/va.h>
25 
26 #if VA_CHECK_VERSION(1, 0, 0)
27 #include <va/va_str.h>
28 #endif
29 
30 #include "libavutil/hwcontext.h"
31 #include "libavutil/hwcontext_vaapi.h"
32 
33 #include "avcodec.h"
34 #include "hwconfig.h"
35 
36 struct VAAPIEncodeType;
37 struct VAAPIEncodePicture;
38 
39 enum {
40     MAX_CONFIG_ATTRIBUTES  = 4,
41     MAX_GLOBAL_PARAMS      = 4,
42     MAX_DPB_SIZE           = 16,
43     MAX_PICTURE_REFERENCES = 2,
44     MAX_REORDER_DELAY      = 16,
45     MAX_PARAM_BUFFER_SIZE  = 1024,
46     // A.4.1: table A.6 allows at most 22 tile rows for any level.
47     MAX_TILE_ROWS          = 22,
48     // A.4.1: table A.6 allows at most 20 tile columns for any level.
49     MAX_TILE_COLS          = 20,
50 };
51 
52 extern const AVCodecHWConfigInternal *const ff_vaapi_encode_hw_configs[];
53 
54 enum {
55     PICTURE_TYPE_IDR = 0,
56     PICTURE_TYPE_I   = 1,
57     PICTURE_TYPE_P   = 2,
58     PICTURE_TYPE_B   = 3,
59 };
60 
61 typedef struct VAAPIEncodeSlice {
62     int             index;
63     int             row_start;
64     int             row_size;
65     int             block_start;
66     int             block_size;
67     void           *codec_slice_params;
68 } VAAPIEncodeSlice;
69 
70 typedef struct VAAPIEncodePicture {
71     struct VAAPIEncodePicture *next;
72 
73     int64_t         display_order;
74     int64_t         encode_order;
75     int64_t         pts;
76     int             force_idr;
77 
78 #if VA_CHECK_VERSION(1, 0, 0)
79     // ROI regions.
80     VAEncROI       *roi;
81 #else
82     void           *roi;
83 #endif
84 
85     int             type;
86     int             b_depth;
87     int             encode_issued;
88     int             encode_complete;
89 
90     AVFrame        *input_image;
91     VASurfaceID     input_surface;
92 
93     AVFrame        *recon_image;
94     VASurfaceID     recon_surface;
95 
96     int          nb_param_buffers;
97     VABufferID     *param_buffers;
98 
99     AVBufferRef    *output_buffer_ref;
100     VABufferID      output_buffer;
101 
102     void           *priv_data;
103     void           *codec_picture_params;
104 
105     // Whether this picture is a reference picture.
106     int             is_reference;
107 
108     // The contents of the DPB after this picture has been decoded.
109     // This will contain the picture itself if it is a reference picture,
110     // but not if it isn't.
111     int                     nb_dpb_pics;
112     struct VAAPIEncodePicture *dpb[MAX_DPB_SIZE];
113     // The reference pictures used in decoding this picture.  If they are
114     // used by later pictures they will also appear in the DPB.
115     int                     nb_refs;
116     struct VAAPIEncodePicture *refs[MAX_PICTURE_REFERENCES];
117     // The previous reference picture in encode order.  Must be in at least
118     // one of the reference list and DPB list.
119     struct VAAPIEncodePicture *prev;
120     // Reference count for other pictures referring to this one through
121     // the above pointers, directly from incomplete pictures and indirectly
122     // through completed pictures.
123     int             ref_count[2];
124     int             ref_removed[2];
125 
126     int          nb_slices;
127     VAAPIEncodeSlice *slices;
128 } VAAPIEncodePicture;
129 
130 typedef struct VAAPIEncodeProfile {
131     // lavc profile value (FF_PROFILE_*).
132     int       av_profile;
133     // Supported bit depth.
134     int       depth;
135     // Number of components.
136     int       nb_components;
137     // Chroma subsampling in width dimension.
138     int       log2_chroma_w;
139     // Chroma subsampling in height dimension.
140     int       log2_chroma_h;
141     // VAAPI profile value.
142     VAProfile va_profile;
143 } VAAPIEncodeProfile;
144 
145 enum {
146     RC_MODE_AUTO,
147     RC_MODE_CQP,
148     RC_MODE_CBR,
149     RC_MODE_VBR,
150     RC_MODE_ICQ,
151     RC_MODE_QVBR,
152     RC_MODE_AVBR,
153     RC_MODE_MAX = RC_MODE_AVBR,
154 };
155 
156 typedef struct VAAPIEncodeRCMode {
157     // Mode from above enum (RC_MODE_*).
158     int mode;
159     // Name.
160     const char *name;
161     // Supported in the compile-time VAAPI version.
162     int supported;
163     // VA mode value (VA_RC_*).
164     uint32_t va_mode;
165     // Uses bitrate parameters.
166     int bitrate;
167     // Supports maxrate distinct from bitrate.
168     int maxrate;
169     // Uses quality value.
170     int quality;
171     // Supports HRD/VBV parameters.
172     int hrd;
173 } VAAPIEncodeRCMode;
174 
175 typedef struct VAAPIEncodeContext {
176     const AVClass *class;
177 
178     // Codec-specific hooks.
179     const struct VAAPIEncodeType *codec;
180 
181     // Global options.
182 
183     // Use low power encoding mode.
184     int             low_power;
185 
186     // Number of I frames between IDR frames.
187     int             idr_interval;
188 
189     // Desired B frame reference depth.
190     int             desired_b_depth;
191 
192     // Explicitly set RC mode (otherwise attempt to pick from
193     // available modes).
194     int             explicit_rc_mode;
195 
196     // Explicitly-set QP, for use with the "qp" options.
197     // (Forces CQP mode when set, overriding everything else.)
198     int             explicit_qp;
199 
200     // Desired packed headers.
201     unsigned int    desired_packed_headers;
202 
203     // The required size of surfaces.  This is probably the input
204     // size (AVCodecContext.width|height) aligned up to whatever
205     // block size is required by the codec.
206     int             surface_width;
207     int             surface_height;
208 
209     // The block size for slice calculations.
210     int             slice_block_width;
211     int             slice_block_height;
212 
213     // Everything above this point must be set before calling
214     // ff_vaapi_encode_init().
215 
216     // Chosen encoding profile details.
217     const VAAPIEncodeProfile *profile;
218 
219     // Chosen rate control mode details.
220     const VAAPIEncodeRCMode *rc_mode;
221     // RC quality level - meaning depends on codec and RC mode.
222     // In CQP mode this sets the fixed quantiser value.
223     int             rc_quality;
224 
225     // Encoding profile (VAProfile*).
226     VAProfile       va_profile;
227     // Encoding entrypoint (VAEntryoint*).
228     VAEntrypoint    va_entrypoint;
229     // Rate control mode.
230     unsigned int    va_rc_mode;
231     // Bitrate for codec-specific encoder parameters.
232     unsigned int    va_bit_rate;
233     // Packed headers which will actually be sent.
234     unsigned int    va_packed_headers;
235 
236     // Configuration attributes to use when creating va_config.
237     VAConfigAttrib  config_attributes[MAX_CONFIG_ATTRIBUTES];
238     int          nb_config_attributes;
239 
240     VAConfigID      va_config;
241     VAContextID     va_context;
242 
243     AVBufferRef    *device_ref;
244     AVHWDeviceContext *device;
245     AVVAAPIDeviceContext *hwctx;
246 
247     // The hardware frame context containing the input frames.
248     AVBufferRef    *input_frames_ref;
249     AVHWFramesContext *input_frames;
250 
251     // The hardware frame context containing the reconstructed frames.
252     AVBufferRef    *recon_frames_ref;
253     AVHWFramesContext *recon_frames;
254 
255     // Pool of (reusable) bitstream output buffers.
256     AVBufferPool   *output_buffer_pool;
257 
258     // Global parameters which will be applied at the start of the
259     // sequence (includes rate control parameters below).
260     int             global_params_type[MAX_GLOBAL_PARAMS];
261     const void     *global_params     [MAX_GLOBAL_PARAMS];
262     size_t          global_params_size[MAX_GLOBAL_PARAMS];
263     int          nb_global_params;
264 
265     // Rate control parameters.
266     VAEncMiscParameterRateControl rc_params;
267     VAEncMiscParameterHRD        hrd_params;
268     VAEncMiscParameterFrameRate   fr_params;
269 #if VA_CHECK_VERSION(0, 36, 0)
270     VAEncMiscParameterBufferQualityLevel quality_params;
271 #endif
272 
273     // Per-sequence parameter structure (VAEncSequenceParameterBuffer*).
274     void           *codec_sequence_params;
275 
276     // Per-sequence parameters found in the per-picture parameter
277     // structure (VAEncPictureParameterBuffer*).
278     void           *codec_picture_params;
279 
280     // Current encoding window, in display (input) order.
281     VAAPIEncodePicture *pic_start, *pic_end;
282     // The next picture to use as the previous reference picture in
283     // encoding order.
284     VAAPIEncodePicture *next_prev;
285 
286     // Next input order index (display order).
287     int64_t         input_order;
288     // Number of frames that output is behind input.
289     int64_t         output_delay;
290     // Next encode order index.
291     int64_t         encode_order;
292     // Number of frames decode output will need to be delayed.
293     int64_t         decode_delay;
294     // Next output order index (in encode order).
295     int64_t         output_order;
296 
297     // Timestamp handling.
298     int64_t         first_pts;
299     int64_t         dts_pts_diff;
300     int64_t         ts_ring[MAX_REORDER_DELAY * 3];
301 
302     // Slice structure.
303     int slice_block_rows;
304     int slice_block_cols;
305     int nb_slices;
306     int slice_size;
307 
308     // Tile encoding.
309     int tile_cols;
310     int tile_rows;
311     // Tile width of the i-th column.
312     int col_width[MAX_TILE_COLS];
313     // Tile height of i-th row.
314     int row_height[MAX_TILE_ROWS];
315     // Location of the i-th tile column boundary.
316     int col_bd[MAX_TILE_COLS + 1];
317     // Location of the i-th tile row boundary.
318     int row_bd[MAX_TILE_ROWS + 1];
319 
320     // Frame type decision.
321     int gop_size;
322     int closed_gop;
323     int gop_per_idr;
324     int p_per_i;
325     int max_b_depth;
326     int b_per_p;
327     int force_idr;
328     int idr_counter;
329     int gop_counter;
330     int end_of_stream;
331 
332     // Whether the driver supports ROI at all.
333     int             roi_allowed;
334     // Maximum number of regions supported by the driver.
335     int             roi_max_regions;
336     // Quantisation range for offset calculations.  Set by codec-specific
337     // code, as it may change based on parameters.
338     int             roi_quant_range;
339 
340     // The encoder does not support cropping information, so warn about
341     // it the first time we encounter any nonzero crop fields.
342     int             crop_warned;
343     // If the driver does not support ROI then warn the first time we
344     // encounter a frame with ROI side data.
345     int             roi_warned;
346 
347     AVFrame         *frame;
348 } VAAPIEncodeContext;
349 
350 enum {
351     // Codec supports controlling the subdivision of pictures into slices.
352     FLAG_SLICE_CONTROL         = 1 << 0,
353     // Codec only supports constant quality (no rate control).
354     FLAG_CONSTANT_QUALITY_ONLY = 1 << 1,
355     // Codec is intra-only.
356     FLAG_INTRA_ONLY            = 1 << 2,
357     // Codec supports B-pictures.
358     FLAG_B_PICTURES            = 1 << 3,
359     // Codec supports referencing B-pictures.
360     FLAG_B_PICTURE_REFERENCES  = 1 << 4,
361     // Codec supports non-IDR key pictures (that is, key pictures do
362     // not necessarily empty the DPB).
363     FLAG_NON_IDR_KEY_PICTURES  = 1 << 5,
364 };
365 
366 typedef struct VAAPIEncodeType {
367     // List of supported profiles and corresponding VAAPI profiles.
368     // (Must end with FF_PROFILE_UNKNOWN.)
369     const VAAPIEncodeProfile *profiles;
370 
371     // Codec feature flags.
372     int flags;
373 
374     // Default quality for this codec - used as quantiser or RC quality
375     // factor depending on RC mode.
376     int default_quality;
377 
378     // Perform any extra codec-specific configuration after the
379     // codec context is initialised (set up the private data and
380     // add any necessary global parameters).
381     int (*configure)(AVCodecContext *avctx);
382 
383     // The size of any private data structure associated with each
384     // picture (can be zero if not required).
385     size_t picture_priv_data_size;
386 
387     // The size of the parameter structures:
388     // sizeof(VAEnc{type}ParameterBuffer{codec}).
389     size_t sequence_params_size;
390     size_t picture_params_size;
391     size_t slice_params_size;
392 
393     // Fill the parameter structures.
394     int  (*init_sequence_params)(AVCodecContext *avctx);
395     int   (*init_picture_params)(AVCodecContext *avctx,
396                                  VAAPIEncodePicture *pic);
397     int     (*init_slice_params)(AVCodecContext *avctx,
398                                  VAAPIEncodePicture *pic,
399                                  VAAPIEncodeSlice *slice);
400 
401     // The type used by the packed header: this should look like
402     // VAEncPackedHeader{something}.
403     int sequence_header_type;
404     int picture_header_type;
405     int slice_header_type;
406 
407     // Write the packed header data to the provided buffer.
408     // The sequence header is also used to fill the codec extradata
409     // when the encoder is starting.
410     int (*write_sequence_header)(AVCodecContext *avctx,
411                                  char *data, size_t *data_len);
412     int  (*write_picture_header)(AVCodecContext *avctx,
413                                  VAAPIEncodePicture *pic,
414                                  char *data, size_t *data_len);
415     int    (*write_slice_header)(AVCodecContext *avctx,
416                                  VAAPIEncodePicture *pic,
417                                  VAAPIEncodeSlice *slice,
418                                  char *data, size_t *data_len);
419 
420     // Fill an extra parameter structure, which will then be
421     // passed to vaRenderPicture().  Will be called repeatedly
422     // with increasing index argument until AVERROR_EOF is
423     // returned.
424     int    (*write_extra_buffer)(AVCodecContext *avctx,
425                                  VAAPIEncodePicture *pic,
426                                  int index, int *type,
427                                  char *data, size_t *data_len);
428 
429     // Write an extra packed header.  Will be called repeatedly
430     // with increasing index argument until AVERROR_EOF is
431     // returned.
432     int    (*write_extra_header)(AVCodecContext *avctx,
433                                  VAAPIEncodePicture *pic,
434                                  int index, int *type,
435                                  char *data, size_t *data_len);
436 } VAAPIEncodeType;
437 
438 
439 int ff_vaapi_encode_receive_packet(AVCodecContext *avctx, AVPacket *pkt);
440 
441 int ff_vaapi_encode_init(AVCodecContext *avctx);
442 int ff_vaapi_encode_close(AVCodecContext *avctx);
443 
444 
445 #define VAAPI_ENCODE_COMMON_OPTIONS \
446     { "low_power", \
447       "Use low-power encoding mode (only available on some platforms; " \
448       "may not support all encoding features)", \
449       OFFSET(common.low_power), AV_OPT_TYPE_BOOL, \
450       { .i64 = 0 }, 0, 1, FLAGS }, \
451     { "idr_interval", \
452       "Distance (in I-frames) between IDR frames", \
453       OFFSET(common.idr_interval), AV_OPT_TYPE_INT, \
454       { .i64 = 0 }, 0, INT_MAX, FLAGS }, \
455     { "b_depth", \
456       "Maximum B-frame reference depth", \
457       OFFSET(common.desired_b_depth), AV_OPT_TYPE_INT, \
458       { .i64 = 1 }, 1, INT_MAX, FLAGS }
459 
460 #define VAAPI_ENCODE_RC_MODE(name, desc) \
461     { #name, desc, 0, AV_OPT_TYPE_CONST, { .i64 = RC_MODE_ ## name }, \
462       0, 0, FLAGS, "rc_mode" }
463 #define VAAPI_ENCODE_RC_OPTIONS \
464     { "rc_mode",\
465       "Set rate control mode", \
466       OFFSET(common.explicit_rc_mode), AV_OPT_TYPE_INT, \
467       { .i64 = RC_MODE_AUTO }, RC_MODE_AUTO, RC_MODE_MAX, FLAGS, "rc_mode" }, \
468     { "auto", "Choose mode automatically based on other parameters", \
469       0, AV_OPT_TYPE_CONST, { .i64 = RC_MODE_AUTO }, 0, 0, FLAGS, "rc_mode" }, \
470     VAAPI_ENCODE_RC_MODE(CQP,  "Constant-quality"), \
471     VAAPI_ENCODE_RC_MODE(CBR,  "Constant-bitrate"), \
472     VAAPI_ENCODE_RC_MODE(VBR,  "Variable-bitrate"), \
473     VAAPI_ENCODE_RC_MODE(ICQ,  "Intelligent constant-quality"), \
474     VAAPI_ENCODE_RC_MODE(QVBR, "Quality-defined variable-bitrate"), \
475     VAAPI_ENCODE_RC_MODE(AVBR, "Average variable-bitrate")
476 
477 
478 #endif /* AVCODEC_VAAPI_ENCODE_H */
479