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
24 #ifndef AVCODEC_VP9DSP_H
25 #define AVCODEC_VP9DSP_H
30 #include "libavcodec/vp9.h"
31 #include "libavutil/attributes_internal.h"
33 typedef void (*vp9_mc_func
)(uint8_t *dst
, ptrdiff_t dst_stride
,
34 const uint8_t *ref
, ptrdiff_t ref_stride
,
35 int h
, int mx
, int my
);
36 typedef void (*vp9_scaled_mc_func
)(uint8_t *dst
, ptrdiff_t dst_stride
,
37 const uint8_t *ref
, ptrdiff_t ref_stride
,
38 int h
, int mx
, int my
, int dx
, int dy
);
40 typedef struct VP9DSPContext
{
42 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32
43 * dimension 2: intra prediction modes
45 * dst/left/top is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
46 * stride is aligned by 16 pixels
47 * top[-1] is top/left; top[4,7] is top-right for 4x4
49 // FIXME(rbultje) maybe replace left/top pointers with HAVE_TOP/
50 // HAVE_LEFT/HAVE_TOPRIGHT flags instead, and then handle it in-place?
51 // also needs to fit in with what H.264/VP8/etc do
52 void (*intra_pred
[N_TXFM_SIZES
][N_INTRA_PRED_MODES
])(uint8_t *dst
,
58 * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32, 4=lossless (3-4=dct only)
59 * dimension 2: 0=dct/dct, 1=dct/adst, 2=adst/dct, 3=adst/adst
61 * dst is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
62 * stride is aligned by 16 pixels
63 * block is 16-byte aligned
64 * eob indicates the position (+1) of the last non-zero coefficient,
65 * in scan-order. This can be used to write faster versions, e.g. a
66 * dc-only 4x4/8x8/16x16/32x32, or a 4x4-only (eob<10) 8x8/16x16/32x32,
69 // FIXME also write idct_add_block() versions for whole (inter) pred
70 // blocks, so we can do 2 4x4s at once
71 void (*itxfm_add
[N_TXFM_SIZES
+ 1][N_TXFM_TYPES
])(uint8_t *dst
,
73 int16_t *block
, int eob
);
76 * dimension 1: width of filter (0=4, 1=8, 2=16)
77 * dimension 2: 0=col-edge filter (h), 1=row-edge filter (v)
79 * dst/stride are aligned by 8
81 void (*loop_filter_8
[3][2])(uint8_t *dst
, ptrdiff_t stride
,
82 int mb_lim
, int lim
, int hev_thr
);
85 * dimension 1: 0=col-edge filter (h), 1=row-edge filter (v)
87 * The width of filter is assumed to be 16; dst/stride are aligned by 16
89 void (*loop_filter_16
[2])(uint8_t *dst
, ptrdiff_t stride
,
90 int mb_lim
, int lim
, int hev_thr
);
93 * dimension 1/2: width of filter (0=4, 1=8) for each filter half
94 * dimension 3: 0=col-edge filter (h), 1=row-edge filter (v)
96 * dst/stride are aligned by operation size
97 * this basically calls loop_filter[d1][d3][0](), followed by
98 * loop_filter[d2][d3][0]() on the next 8 pixels
99 * mb_lim/lim/hev_thr contain two values in the lowest two bytes of the
102 // FIXME perhaps a mix4 that operates on 32px (for AVX2)
103 void (*loop_filter_mix2
[2][2][2])(uint8_t *dst
, ptrdiff_t stride
,
104 int mb_lim
, int lim
, int hev_thr
);
107 * dimension 1: hsize (0: 64, 1: 32, 2: 16, 3: 8, 4: 4)
108 * dimension 2: filter type (0: smooth, 1: regular, 2: sharp, 3: bilin)
109 * dimension 3: averaging type (0: put, 1: avg)
110 * dimension 4: x subpel interpolation (0: none, 1: 8tap/bilin)
111 * dimension 5: y subpel interpolation (0: none, 1: 8tap/bilin)
113 * dst/stride are aligned by hsize
115 vp9_mc_func mc
[5][N_FILTERS
][2][2][2];
118 * for scalable MC, first 3 dimensions identical to above, the other two
119 * don't exist since it changes per stepsize.
121 vp9_scaled_mc_func smc
[5][N_FILTERS
][2];
124 extern const int16_t attribute_visibility_hidden ff_vp9_subpel_filters
[3][16][8];
126 void ff_vp9dsp_init(VP9DSPContext
*dsp
, int bpp
, int bitexact
);
128 void ff_vp9dsp_init_8(VP9DSPContext
*dsp
);
129 void ff_vp9dsp_init_10(VP9DSPContext
*dsp
);
130 void ff_vp9dsp_init_12(VP9DSPContext
*dsp
);
132 void ff_vp9dsp_init_aarch64(VP9DSPContext
*dsp
, int bpp
);
133 void ff_vp9dsp_init_arm(VP9DSPContext
*dsp
, int bpp
);
134 void ff_vp9dsp_init_x86(VP9DSPContext
*dsp
, int bpp
, int bitexact
);
135 void ff_vp9dsp_init_mips(VP9DSPContext
*dsp
, int bpp
);
136 void ff_vp9dsp_init_loongarch(VP9DSPContext
*dsp
, int bpp
);
138 #endif /* AVCODEC_VP9DSP_H */