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 Libav.
9 * Libav 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 * Libav 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 Libav; 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);
134 typedef struct VP9DSPContext {
136 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32
137 * dimension 2: intra prediction modes
139 * dst/left/top is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
140 * stride is aligned by 16 pixels
141 * top[-1] is top/left; top[4,7] is top-right for 4x4
143 // FIXME(rbultje) maybe replace left/top pointers with HAVE_TOP/
144 // HAVE_LEFT/HAVE_TOPRIGHT flags instead, and then handle it in-place?
145 // also needs to fit in with what H.264/VP8/etc do
146 void (*intra_pred[N_TXFM_SIZES][N_INTRA_PRED_MODES])(uint8_t *dst,
152 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32, 4=lossless (3-4=dct only)
153 * dimension 2: 0=dct/dct, 1=dct/adst, 2=adst/dct, 3=adst/adst
155 * dst is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
156 * stride is aligned by 16 pixels
157 * block is 16-byte aligned
158 * eob indicates the position (+1) of the last non-zero coefficient,
159 * in scan-order. This can be used to write faster versions, e.g. a
160 * dc-only 4x4/8x8/16x16/32x32, or a 4x4-only (eob<10) 8x8/16x16/32x32,
163 // FIXME also write idct_add_block() versions for whole (inter) pred
164 // blocks, so we can do 2 4x4s at once
165 void (*itxfm_add[N_TXFM_SIZES + 1][N_TXFM_TYPES])(uint8_t *dst,
167 int16_t *block, int eob);
170 * dimension 1: width of filter (0=4, 1=8, 2=16)
171 * dimension 2: 0=col-edge filter (h), 1=row-edge filter (v)
173 * dst/stride are aligned by 8
175 void (*loop_filter_8[3][2])(uint8_t *dst, ptrdiff_t stride,
176 int mb_lim, int lim, int hev_thr);
179 * dimension 1: 0=col-edge filter (h), 1=row-edge filter (v)
181 * The width of filter is assumed to be 16; dst/stride are aligned by 16
183 void (*loop_filter_16[2])(uint8_t *dst, ptrdiff_t stride,
184 int mb_lim, int lim, int hev_thr);
187 * dimension 1/2: width of filter (0=4, 1=8) for each filter half
188 * dimension 3: 0=col-edge filter (h), 1=row-edge filter (v)
190 * dst/stride are aligned by operation size
191 * this basically calls loop_filter[d1][d3][0](), followed by
192 * loop_filter[d2][d3][0]() on the next 8 pixels
193 * mb_lim/lim/hev_thr contain two values in the lowest two bytes of the
196 // FIXME perhaps a mix4 that operates on 32px (for AVX2)
197 void (*loop_filter_mix2[2][2][2])(uint8_t *dst, ptrdiff_t stride,
198 int mb_lim, int lim, int hev_thr);
201 * dimension 1: hsize (0: 64, 1: 32, 2: 16, 3: 8, 4: 4)
202 * dimension 2: filter type (0: smooth, 1: regular, 2: sharp, 3: bilin)
203 * dimension 3: averaging type (0: put, 1: avg)
204 * dimension 4: x subpel interpolation (0: none, 1: 8tap/bilin)
205 * dimension 5: y subpel interpolation (1: none, 1: 8tap/bilin)
207 * dst/stride are aligned by hsize
209 vp9_mc_func mc[5][4][2][2][2];
218 typedef struct VP9MVRefPair {
223 typedef struct VP9Filter {
224 uint8_t level[8 * 8];
225 uint8_t /* bit=col */ mask[2 /* 0=y, 1=uv */][2 /* 0=col, 1=row */]
226 [8 /* rows */][4 /* 0=16, 1=8, 2=4, 3=inner4 */];
229 typedef struct VP9Frame {
232 uint8_t *segmentation_map;
235 AVBufferRef *segmentation_map_buf;
263 typedef struct VP9Block {
264 uint8_t seg_id, intra, comp, ref[2], mode[4], uvmode, skip;
265 enum FilterMode filter;
266 VP56mv mv[4 /* b_idx */][2 /* ref */];
268 enum TxfmMode tx, uvtx;
270 int row, row7, col, col7;
272 ptrdiff_t y_stride, uv_stride;
275 enum BlockPartition bp;
278 typedef struct VP9Context {
280 VideoDSPContext vdsp;
298 uint8_t keyframe, last_keyframe;
300 uint8_t use_last_frame_mvs;
308 uint8_t refreshrefmask;
309 uint8_t highprecisionmvs;
310 enum FilterMode filtermode;
311 uint8_t allowcompinter;
314 uint8_t parallelmode;
318 uint8_t varcompref[2];
330 uint8_t mblim_lut[64];
338 int8_t ydc_qdelta, uvdc_qdelta, uvac_qdelta;
343 uint8_t absolute_vals;
345 #define MAX_SEGMENT 8
350 uint8_t skip_enabled;
359 unsigned log2_tile_cols, log2_tile_rows;
360 unsigned tile_cols, tile_rows;
361 unsigned tile_row_start, tile_row_end, tile_col_start, tile_col_end;
363 unsigned sb_cols, sb_rows, rows, cols;
366 uint8_t coef[4][2][2][6][6][3];
370 uint8_t coef[4][2][2][6][6][11];
375 unsigned y_mode[4][10];
376 unsigned uv_mode[10][10];
377 unsigned filter[4][3];
378 unsigned mv_mode[7][4];
379 unsigned intra[4][2];
381 unsigned single_ref[5][2][2];
382 unsigned comp_ref[5][2];
383 unsigned tx32p[2][4];
384 unsigned tx16p[2][3];
387 unsigned mv_joint[4];
390 unsigned classes[11];
392 unsigned bits[10][2];
393 unsigned class0_fp[2][4];
395 unsigned class0_hp[2];
398 unsigned partition[4][4][4];
399 unsigned coef[4][2][2][6][6][3];
400 unsigned eob[4][2][2][6][6][2];
402 enum TxfmMode txfmmode;
403 enum CompPredMode comppredmode;
405 // contextual (left/above) cache
406 uint8_t left_partition_ctx[8], *above_partition_ctx;
407 uint8_t left_mode_ctx[16], *above_mode_ctx;
408 // FIXME maybe merge some of the below in a flags field?
409 uint8_t left_y_nnz_ctx[16], *above_y_nnz_ctx;
410 uint8_t left_uv_nnz_ctx[2][8], *above_uv_nnz_ctx[2];
411 uint8_t left_skip_ctx[8], *above_skip_ctx; // 1bit
412 uint8_t left_txfm_ctx[8], *above_txfm_ctx; // 2bit
413 uint8_t left_segpred_ctx[8], *above_segpred_ctx; // 1bit
414 uint8_t left_intra_ctx[8], *above_intra_ctx; // 1bit
415 uint8_t left_comp_ctx[8], *above_comp_ctx; // 1bit
416 uint8_t left_ref_ctx[8], *above_ref_ctx; // 2bit
417 uint8_t left_filter_ctx[8], *above_filter_ctx;
418 VP56mv left_mv_ctx[16][2], (*above_mv_ctx)[2];
421 uint8_t *intra_pred_data[3];
423 // This requires 64 + 8 rows, with 80 bytes stride
424 DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[72 * 80];
426 // block reconstruction intermediates
427 int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
428 uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];
429 struct { int x, y; } min_mv, max_mv;
430 DECLARE_ALIGNED(32, uint8_t, tmp_y)[64 * 64];
431 DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][32 * 32];
434 extern const int8_t ff_vp9_subpel_filters[3][15][8];
436 void ff_vp9dsp_init(VP9DSPContext *dsp);
438 void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp);
439 void ff_vp9dsp_init_arm(VP9DSPContext *dsp);
440 void ff_vp9dsp_init_x86(VP9DSPContext *dsp);
442 void ff_vp9_fill_mv(VP9Context *s, VP56mv *mv, int mode, int sb);
444 void ff_vp9_adapt_probs(VP9Context *s);
446 int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
447 VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
448 enum BlockLevel bl, enum BlockPartition bp);
450 #endif /* AVCODEC_VP9_H */