2
* This file is part of Libav.
4
* Libav 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.
9
* Libav 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.
14
* You should have received a copy of the GNU Lesser General Public
15
* License along with Libav; if not, write to the Free Software
16
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
#include "libavcodec/vp8dsp.h"
23
void ff_vp8_luma_dc_wht_neon(DCTELEM block[4][4][16], DCTELEM dc[16]);
25
void ff_vp8_idct_add_neon(uint8_t *dst, DCTELEM block[16], ptrdiff_t stride);
26
void ff_vp8_idct_dc_add_neon(uint8_t *dst, DCTELEM block[16], ptrdiff_t stride);
27
void ff_vp8_idct_dc_add4y_neon(uint8_t *dst, DCTELEM block[4][16], ptrdiff_t stride);
28
void ff_vp8_idct_dc_add4uv_neon(uint8_t *dst, DCTELEM block[4][16], ptrdiff_t stride);
40
av_cold void ff_vp8dsp_init_neon(VP8DSPContext *dsp)
42
dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_neon;
44
dsp->vp8_idct_add = ff_vp8_idct_add_neon;
45
dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_neon;
46
dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_neon;
47
dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_neon;
49
dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_neon;
50
dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_neon;
51
dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_neon;
52
dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_neon;
54
dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_neon;
55
dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_neon;
56
dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_neon;
57
dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_neon;
59
dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_neon;
60
dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_neon;
62
dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
63
dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_neon;
64
dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_neon;
65
dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_neon;
67
dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
68
dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_neon;
69
dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_neon;
70
dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_neon;
71
dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_neon;
72
dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_neon;
73
dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_neon;
74
dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_neon;
75
dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_neon;
77
dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_neon;
78
dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_neon;
79
dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_neon;
80
dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_neon;
81
dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_neon;
82
dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_neon;
83
dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_neon;
84
dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_neon;
86
dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
87
dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_neon;
88
dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_neon;
89
dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_neon;
90
dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilin16_hv_neon;
91
dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilin16_hv_neon;
92
dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_neon;
93
dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilin16_hv_neon;
94
dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_neon;
96
dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
97
dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_neon;
98
dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_neon;
99
dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_neon;
100
dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_neon;
101
dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_neon;
102
dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_neon;
103
dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_neon;
104
dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_neon;
106
dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_neon;
107
dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_neon;
108
dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_neon;
109
dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_neon;
110
dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_neon;
111
dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_neon;
112
dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_neon;
113
dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_neon;