This attaches the logic of picking the mode of for the next picture to the output, which simplifies some choices by removing the concept of the picture for which input is not yet available. At the same time, we allow more complex reference structures and track more reference metadata (particularly the contents of the DPB) for use in the codec-specific code. It also adds flags to explicitly track the available features of the different codecs. The new structure also allows open-GOP support, so that is now available for codecs which can do it.
Showing the changes to libavcodec/vaapi_encode.h (1 of 9 files). Show all
Unified · Split
| @@ -38,6 +38,7 @@ struct VAAPIEncodePicture; · 37 unmodified lines | ||
| 38 | 38 | enum { |
| 39 | 39 | MAX_CONFIG_ATTRIBUTES = 4, |
| 40 | 40 | MAX_GLOBAL_PARAMS = 4, |
| 41 | + MAX_DPB_SIZE = 16, | |
| 41 | 42 | MAX_PICTURE_REFERENCES = 2, |
| 42 | 43 | MAX_REORDER_DELAY = 16, |
| 43 | 44 | MAX_PARAM_BUFFER_SIZE = 1024, |
| @@ -66,9 +67,10 @@ typedef struct VAAPIEncodePicture { · 22 unmodified lines | ||
| 66 | 67 | int64_t display_order; |
| 67 | 68 | int64_t encode_order; |
| 68 | 69 | int64_t pts; |
| 70 | + int force_idr; | |
| 69 | 71 | |
| 70 | 72 | int type; |
| 71 | − int input_available; | |
| 73 | + int b_depth; | |
| 72 | 74 | int encode_issued; |
| 73 | 75 | int encode_complete; |
| 74 | 76 | |
| @@ -87,8 +89,26 @@ typedef struct VAAPIEncodePicture { · 12 unmodified lines | ||
| 87 | 89 | void *priv_data; |
| 88 | 90 | void *codec_picture_params; |
| 89 | 91 | |
| 90 | − int nb_refs; | |
| 92 | + // Whether this picture is a reference picture. | |
| 93 | + int is_reference; | |
| 94 | + | |
| 95 | + // The contents of the DPB after this picture has been decoded. | |
| 96 | + // This will contain the picture itself if it is a reference picture, | |
| 97 | + // but not if it isn't. | |
| 98 | + int nb_dpb_pics; | |
| 99 | + struct VAAPIEncodePicture *dpb[MAX_DPB_SIZE]; | |
| 100 | + // The reference pictures used in decoding this picture. If they are | |
| 101 | + // used by later pictures they will also appear in the DPB. | |
| 102 | + int nb_refs; | |
| 91 | 103 | struct VAAPIEncodePicture *refs[MAX_PICTURE_REFERENCES]; |
| 104 | + // The previous reference picture in encode order. Must be in at least | |
| 105 | + // one of the reference list and DPB list. | |
| 106 | + struct VAAPIEncodePicture *prev; | |
| 107 | + // Reference count for other pictures referring to this one through | |
| 108 | + // the above pointers, directly from incomplete pictures and indirectly | |
| 109 | + // through completed pictures. | |
| 110 | + int ref_count[2]; | |
| 111 | + int ref_removed[2]; | |
| 92 | 112 | |
| 93 | 113 | int nb_slices; |
| 94 | 114 | VAAPIEncodeSlice *slices; |
| @@ -120,6 +140,12 @@ typedef struct VAAPIEncodeContext { · 25 unmodified lines | ||
| 120 | 140 | // Use low power encoding mode. |
| 121 | 141 | int low_power; |
| 122 | 142 | |
| 143 | + // Number of I frames between IDR frames. | |
| 144 | + int idr_interval; | |
| 145 | + | |
| 146 | + // Desired B frame reference depth. | |
| 147 | + int desired_b_depth; | |
| 148 | + | |
| 123 | 149 | // Desired packed headers. |
| 124 | 150 | unsigned int desired_packed_headers; |
| 125 | 151 | |
| @@ -207,26 +233,21 @@ typedef struct VAAPIEncodeContext { · 81 unmodified lines | ||
| 207 | 233 | |
| 208 | 234 | // Current encoding window, in display (input) order. |
| 209 | 235 | VAAPIEncodePicture *pic_start, *pic_end; |
| 236 | + // The next picture to use as the previous reference picture in | |
| 237 | + // encoding order. | |
| 238 | + VAAPIEncodePicture *next_prev; | |
| 210 | 239 | |
| 211 | 240 | // Next input order index (display order). |
| 212 | 241 | int64_t input_order; |
| 213 | 242 | // Number of frames that output is behind input. |
| 214 | 243 | int64_t output_delay; |
| 244 | + // Next encode order index. | |
| 245 | + int64_t encode_order; | |
| 215 | 246 | // Number of frames decode output will need to be delayed. |
| 216 | 247 | int64_t decode_delay; |
| 217 | − // Next output order index (encode order). | |
| 248 | + // Next output order index (in encode order). | |
| 218 | 249 | int64_t output_order; |
| 219 | 250 | |
| 220 | − enum { | |
| 221 | − // All encode operations are done independently (synchronise | |
| 222 | − // immediately after every operation). | |
| 223 | − ISSUE_MODE_SERIALISE_EVERYTHING = 0, | |
| 224 | − // Overlap as many operations as possible. | |
| 225 | − ISSUE_MODE_MAXIMISE_THROUGHPUT, | |
| 226 | − // Overlap operations only when satisfying parallel dependencies. | |
| 227 | − ISSUE_MODE_MINIMISE_LATENCY, | |
| 228 | − } issue_mode; | |
| 229 | − | |
| 230 | 251 | // Timestamp handling. |
| 231 | 252 | int64_t first_pts; |
| 232 | 253 | int64_t dts_pts_diff; |
| @@ -240,11 +261,14 @@ typedef struct VAAPIEncodeContext { · 7 unmodified lines | ||
| 240 | 261 | |
| 241 | 262 | // Frame type decision. |
| 242 | 263 | int gop_size; |
| 264 | + int closed_gop; | |
| 265 | + int gop_per_idr; | |
| 243 | 266 | int p_per_i; |
| 267 | + int max_b_depth; | |
| 244 | 268 | int b_per_p; |
| 245 | 269 | int force_idr; |
| 270 | + int idr_counter; | |
| 246 | 271 | int gop_counter; |
| 247 | − int p_counter; | |
| 248 | 272 | int end_of_stream; |
| 249 | 273 | } VAAPIEncodeContext; |
| 250 | 274 | |
| @@ -253,6 +277,15 @@ enum { · 2 unmodified lines | ||
| 253 | 277 | FLAG_SLICE_CONTROL = 1 << 0, |
| 254 | 278 | // Codec only supports constant quality (no rate control). |
| 255 | 279 | FLAG_CONSTANT_QUALITY_ONLY = 1 << 1, |
| 280 | + // Codec is intra-only. | |
| 281 | + FLAG_INTRA_ONLY = 1 << 2, | |
| 282 | + // Codec supports B-pictures. | |
| 283 | + FLAG_B_PICTURES = 1 << 3, | |
| 284 | + // Codec supports referencing B-pictures. | |
| 285 | + FLAG_B_PICTURE_REFERENCES = 1 << 4, | |
| 286 | + // Codec supports non-IDR key pictures (that is, key pictures do | |
| 287 | + // not necessarily empty the DPB). | |
| 288 | + FLAG_NON_IDR_KEY_PICTURES = 1 << 5, | |
| 256 | 289 | }; |
| 257 | 290 | |
| 258 | 291 | typedef struct VAAPIEncodeType { |
| @@ -327,6 +360,9 @@ typedef struct VAAPIEncodeType { · 68 unmodified lines | ||
| 327 | 360 | int ff_vaapi_encode2(AVCodecContext *avctx, AVPacket *pkt, |
| 328 | 361 | const AVFrame *input_image, int *got_packet); |
| 329 | 362 | |
| 363 | +int ff_vaapi_encode_send_frame(AVCodecContext *avctx, const AVFrame *frame); | |
| 364 | +int ff_vaapi_encode_receive_packet(AVCodecContext *avctx, AVPacket *pkt); | |
| 365 | + | |
| 330 | 366 | int ff_vaapi_encode_init(AVCodecContext *avctx); |
| 331 | 367 | int ff_vaapi_encode_close(AVCodecContext *avctx); |
| 332 | 368 | |
| @@ -336,7 +372,15 @@ int ff_vaapi_encode_close(AVCodecContext *avctx); · 3 unmodified lines | ||
| 336 | 372 | "Use low-power encoding mode (only available on some platforms; " \ |
| 337 | 373 | "may not support all encoding features)", \ |
| 338 | 374 | OFFSET(common.low_power), AV_OPT_TYPE_BOOL, \ |
| 339 | − { .i64 = 0 }, 0, 1, FLAGS } | |
| 375 | + { .i64 = 0 }, 0, 1, FLAGS }, \ | |
| 376 | + { "idr_interval", \ | |
| 377 | + "Distance (in I-frames) between IDR frames", \ | |
| 378 | + OFFSET(common.idr_interval), AV_OPT_TYPE_INT, \ | |
| 379 | + { .i64 = 0 }, 0, INT_MAX, FLAGS }, \ | |
| 380 | + { "b_depth", \ | |
| 381 | + "Maximum B-frame reference depth", \ | |
| 382 | + OFFSET(common.desired_b_depth), AV_OPT_TYPE_INT, \ | |
| 383 | + { .i64 = 1 }, 1, INT_MAX, FLAGS } | |
| 340 | 384 | |
| 341 | 385 | |
| 342 | 386 | #endif /* AVCODEC_VAAPI_ENCODE_H */ |