2 * VP9 compatible video decoder
4 * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
5 * Copyright (C) 2013 Clément Bœsch <u pkh me>
7 * This file is part of FFmpeg.
9 * FFmpeg 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.
14 * FFmpeg 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.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 #include "libavutil/buffer.h"
31 #include "libavutil/internal.h"
43 TX_SWITCHABLE = N_TXFM_SIZES,
83 PARTITION_NONE, // [ ] <-.
86 PARTITION_SPLIT, // [+] --'
103 typedef struct ProbContext {
104 uint8_t y_mode[4][9];
105 uint8_t uv_mode[10][9];
106 uint8_t filter[4][2];
107 uint8_t mv_mode[7][3];
110 uint8_t single_ref[5][2];
122 uint8_t class0_fp[2][3];
127 uint8_t partition[4][4][3];
130 typedef void (*vp9_mc_func)(uint8_t *dst, ptrdiff_t dst_stride,
131 const uint8_t *ref, ptrdiff_t ref_stride,
132 int h, int mx, int my);
133 typedef void (*vp9_scaled_mc_func)(uint8_t *dst, ptrdiff_t dst_stride,
134 const uint8_t *ref, ptrdiff_t ref_stride,
135 int h, int mx, int my, int dx, int dy);
137 typedef struct VP9DSPContext {
139 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32
140 * dimension 2: intra prediction modes
142 * dst/left/top is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
143 * stride is aligned by 16 pixels
144 * top[-1] is top/left; top[4,7] is top-right for 4x4
146 // FIXME(rbultje) maybe replace left/top pointers with HAVE_TOP/
147 // HAVE_LEFT/HAVE_TOPRIGHT flags instead, and then handle it in-place?
148 // also needs to fit in with what H.264/VP8/etc do
149 void (*intra_pred[N_TXFM_SIZES][N_INTRA_PRED_MODES])(uint8_t *dst,
155 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32, 4=lossless (3-4=dct only)
156 * dimension 2: 0=dct/dct, 1=dct/adst, 2=adst/dct, 3=adst/adst
158 * dst is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
159 * stride is aligned by 16 pixels
160 * block is 16-byte aligned
161 * eob indicates the position (+1) of the last non-zero coefficient,
162 * in scan-order. This can be used to write faster versions, e.g. a
163 * dc-only 4x4/8x8/16x16/32x32, or a 4x4-only (eob<10) 8x8/16x16/32x32,
166 // FIXME also write idct_add_block() versions for whole (inter) pred
167 // blocks, so we can do 2 4x4s at once
168 void (*itxfm_add[N_TXFM_SIZES + 1][N_TXFM_TYPES])(uint8_t *dst,
170 int16_t *block, int eob);
173 * dimension 1: width of filter (0=4, 1=8, 2=16)
174 * dimension 2: 0=col-edge filter (h), 1=row-edge filter (v)
176 * dst/stride are aligned by 8
178 void (*loop_filter_8[3][2])(uint8_t *dst, ptrdiff_t stride,
179 int mb_lim, int lim, int hev_thr);
182 * dimension 1: 0=col-edge filter (h), 1=row-edge filter (v)
184 * The width of filter is assumed to be 16; dst/stride are aligned by 16
186 void (*loop_filter_16[2])(uint8_t *dst, ptrdiff_t stride,
187 int mb_lim, int lim, int hev_thr);
190 * dimension 1/2: width of filter (0=4, 1=8) for each filter half
191 * dimension 3: 0=col-edge filter (h), 1=row-edge filter (v)
193 * dst/stride are aligned by operation size
194 * this basically calls loop_filter[d1][d3][0](), followed by
195 * loop_filter[d2][d3][0]() on the next 8 pixels
196 * mb_lim/lim/hev_thr contain two values in the lowest two bytes of the
199 // FIXME perhaps a mix4 that operates on 32px (for AVX2)
200 void (*loop_filter_mix2[2][2][2])(uint8_t *dst, ptrdiff_t stride,
201 int mb_lim, int lim, int hev_thr);
204 * dimension 1: hsize (0: 64, 1: 32, 2: 16, 3: 8, 4: 4)
205 * dimension 2: filter type (0: smooth, 1: regular, 2: sharp, 3: bilin)
206 * dimension 3: averaging type (0: put, 1: avg)
207 * dimension 4: x subpel interpolation (0: none, 1: 8tap/bilin)
208 * dimension 5: y subpel interpolation (0: none, 1: 8tap/bilin)
210 * dst/stride are aligned by hsize
212 vp9_mc_func mc[5][4][2][2][2];
215 * for scalable MC, first 3 dimensions identical to above, the other two
216 * don't exist since it changes per stepsize.
218 vp9_scaled_mc_func smc[5][4][2];
227 typedef struct VP9mvrefPair {
232 typedef struct VP9Filter {
233 uint8_t level[8 * 8];
234 uint8_t /* bit=col */ mask[2 /* 0=y, 1=uv */][2 /* 0=col, 1=row */]
235 [8 /* rows */][4 /* 0=16, 1=8, 2=4, 3=inner4 */];
238 typedef struct VP9Frame {
240 AVBufferRef *extradata;
241 uint8_t *segmentation_map;
245 AVBufferRef *hwaccel_priv_buf;
246 void *hwaccel_picture_private;
273 typedef struct VP9Block {
274 uint8_t seg_id, intra, comp, ref[2], mode[4], uvmode, skip;
275 enum FilterMode filter;
276 VP56mv mv[4 /* b_idx */][2 /* ref */];
278 enum TxfmMode tx, uvtx;
280 enum BlockPartition bp;
283 typedef struct VP9BitstreamHeader {
292 uint8_t refreshrefmask;
293 uint8_t highprecisionmvs;
294 enum FilterMode filtermode;
295 uint8_t allowcompinter;
297 uint8_t parallelmode;
299 uint8_t use_last_frame_mvs;
303 uint8_t varcompref[2];
315 int8_t ydc_qdelta, uvdc_qdelta, uvac_qdelta;
317 #define MAX_SEGMENT 8
321 uint8_t absolute_vals;
324 uint8_t pred_prob[3];
329 uint8_t skip_enabled;
337 enum TxfmMode txfmmode;
338 enum CompPredMode comppredmode;
340 unsigned log2_tile_cols, log2_tile_rows;
341 unsigned tile_cols, tile_rows;
344 int uncompressed_header_size;
345 int compressed_header_size;
346 } VP9BitstreamHeader;
348 typedef struct VP9SharedContext {
349 VP9BitstreamHeader h;
353 #define REF_FRAME_MVPAIR 1
354 #define REF_FRAME_SEGMAP 2
358 typedef struct VP9Context {
362 VideoDSPContext vdsp;
367 VP9Block *b_base, *b;
369 int row, row7, col, col7;
371 ptrdiff_t y_stride, uv_stride;
374 uint8_t last_bpp, bpp_index, bytesperpixel;
375 uint8_t last_keyframe;
376 // sb_cols/rows, rows/cols and last_fmt are used for allocating all internal
377 // arrays, and are thus per-thread. w/h and gf_fmt are synced between threads
378 // and are therefore per-stream. pix_fmt represents the value in the header
379 // of the currently processed frame.
381 enum AVPixelFormat pix_fmt, last_fmt, gf_fmt;
382 unsigned sb_cols, sb_rows, rows, cols;
383 ThreadFrame next_refs[8];
387 uint8_t mblim_lut[64];
389 unsigned tile_row_start, tile_row_end, tile_col_start, tile_col_end;
392 uint8_t coef[4][2][2][6][6][3];
396 uint8_t coef[4][2][2][6][6][11];
399 unsigned y_mode[4][10];
400 unsigned uv_mode[10][10];
401 unsigned filter[4][3];
402 unsigned mv_mode[7][4];
403 unsigned intra[4][2];
405 unsigned single_ref[5][2][2];
406 unsigned comp_ref[5][2];
407 unsigned tx32p[2][4];
408 unsigned tx16p[2][3];
411 unsigned mv_joint[4];
414 unsigned classes[11];
416 unsigned bits[10][2];
417 unsigned class0_fp[2][4];
419 unsigned class0_hp[2];
422 unsigned partition[4][4][4];
423 unsigned coef[4][2][2][6][6][3];
424 unsigned eob[4][2][2][6][6][2];
427 // contextual (left/above) cache
428 DECLARE_ALIGNED(16, uint8_t, left_y_nnz_ctx)[16];
429 DECLARE_ALIGNED(16, uint8_t, left_mode_ctx)[16];
430 DECLARE_ALIGNED(16, VP56mv, left_mv_ctx)[16][2];
431 DECLARE_ALIGNED(16, uint8_t, left_uv_nnz_ctx)[2][16];
432 DECLARE_ALIGNED(8, uint8_t, left_partition_ctx)[8];
433 DECLARE_ALIGNED(8, uint8_t, left_skip_ctx)[8];
434 DECLARE_ALIGNED(8, uint8_t, left_txfm_ctx)[8];
435 DECLARE_ALIGNED(8, uint8_t, left_segpred_ctx)[8];
436 DECLARE_ALIGNED(8, uint8_t, left_intra_ctx)[8];
437 DECLARE_ALIGNED(8, uint8_t, left_comp_ctx)[8];
438 DECLARE_ALIGNED(8, uint8_t, left_ref_ctx)[8];
439 DECLARE_ALIGNED(8, uint8_t, left_filter_ctx)[8];
440 uint8_t *above_partition_ctx;
441 uint8_t *above_mode_ctx;
442 // FIXME maybe merge some of the below in a flags field?
443 uint8_t *above_y_nnz_ctx;
444 uint8_t *above_uv_nnz_ctx[2];
445 uint8_t *above_skip_ctx; // 1bit
446 uint8_t *above_txfm_ctx; // 2bit
447 uint8_t *above_segpred_ctx; // 1bit
448 uint8_t *above_intra_ctx; // 1bit
449 uint8_t *above_comp_ctx; // 1bit
450 uint8_t *above_ref_ctx; // 2bit
451 uint8_t *above_filter_ctx;
452 VP56mv (*above_mv_ctx)[2];
455 uint8_t *intra_pred_data[3];
457 DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[135 * 144 * 2];
459 // block reconstruction intermediates
460 int block_alloc_using_2pass;
461 int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
462 uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];
463 struct { int x, y; } min_mv, max_mv;
464 DECLARE_ALIGNED(32, uint8_t, tmp_y)[64 * 64 * 2];
465 DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][64 * 64 * 2];
466 uint16_t mvscale[3][2];
467 uint8_t mvstep[3][2];
470 extern const int16_t ff_vp9_subpel_filters[3][16][8];
472 void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp, int bitexact);
474 void ff_vp9dsp_init_8(VP9DSPContext *dsp);
475 void ff_vp9dsp_init_10(VP9DSPContext *dsp);
476 void ff_vp9dsp_init_12(VP9DSPContext *dsp);
478 void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp, int bpp);
479 void ff_vp9dsp_init_arm(VP9DSPContext *dsp, int bpp);
480 void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact);
481 void ff_vp9dsp_init_mips(VP9DSPContext *dsp, int bpp);
483 void ff_vp9_fill_mv(VP9Context *s, VP56mv *mv, int mode, int sb);
485 void ff_vp9_adapt_probs(VP9Context *s);
487 void ff_vp9_decode_block(AVCodecContext *ctx, int row, int col,
488 VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
489 enum BlockLevel bl, enum BlockPartition bp);
491 #endif /* AVCODEC_VP9_H */