1 /*
2 * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
3 * Copyright (c) 2016 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * VP8 compatible video decoder
25 */
26
27 #include "libavutil/mips/cpu.h"
28 #include "config.h"
29 #include "libavutil/attributes.h"
30 #include "libavcodec/vp8dsp.h"
31 #include "vp8dsp_mips.h"
32
33 #define VP8_MC_MIPS_FUNC(IDX, SIZE) \
34 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = \
35 ff_put_vp8_epel##SIZE##_h4_msa; \
36 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = \
37 ff_put_vp8_epel##SIZE##_h6_msa; \
38 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = \
39 ff_put_vp8_epel##SIZE##_v4_msa; \
40 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = \
41 ff_put_vp8_epel##SIZE##_h4v4_msa; \
42 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = \
43 ff_put_vp8_epel##SIZE##_h6v4_msa; \
44 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = \
45 ff_put_vp8_epel##SIZE##_v6_msa; \
46 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = \
47 ff_put_vp8_epel##SIZE##_h4v6_msa; \
48 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = \
49 ff_put_vp8_epel##SIZE##_h6v6_msa
50
51 #define VP8_BILINEAR_MC_MIPS_FUNC(IDX, SIZE) \
52 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = \
53 ff_put_vp8_bilinear##SIZE##_h_msa; \
54 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = \
55 ff_put_vp8_bilinear##SIZE##_h_msa; \
56 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = \
57 ff_put_vp8_bilinear##SIZE##_v_msa; \
58 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = \
59 ff_put_vp8_bilinear##SIZE##_hv_msa; \
60 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = \
61 ff_put_vp8_bilinear##SIZE##_hv_msa; \
62 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = \
63 ff_put_vp8_bilinear##SIZE##_v_msa; \
64 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = \
65 ff_put_vp8_bilinear##SIZE##_hv_msa; \
66 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = \
67 ff_put_vp8_bilinear##SIZE##_hv_msa
68
69 #define VP8_MC_MIPS_COPY(IDX, SIZE) \
70 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = \
71 ff_put_vp8_pixels##SIZE##_msa; \
72 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = \
73 ff_put_vp8_pixels##SIZE##_msa;
74
75
76
ff_vp8dsp_init_mips(VP8DSPContext * dsp)77 av_cold void ff_vp8dsp_init_mips(VP8DSPContext *dsp)
78 {
79 int cpu_flags = av_get_cpu_flags();
80
81 if (have_mmi(cpu_flags)) {
82 dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmi;
83 dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_mmi;
84 dsp->vp8_idct_add = ff_vp8_idct_add_mmi;
85 dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmi;
86 dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_mmi;
87 dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmi;
88
89 dsp->put_vp8_epel_pixels_tab[0][0][1] = ff_put_vp8_epel16_h4_mmi;
90 dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_mmi;
91 dsp->put_vp8_epel_pixels_tab[0][1][0] = ff_put_vp8_epel16_v4_mmi;
92 dsp->put_vp8_epel_pixels_tab[0][1][1] = ff_put_vp8_epel16_h4v4_mmi;
93 dsp->put_vp8_epel_pixels_tab[0][1][2] = ff_put_vp8_epel16_h6v4_mmi;
94 dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_mmi;
95 dsp->put_vp8_epel_pixels_tab[0][2][1] = ff_put_vp8_epel16_h4v6_mmi;
96 dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_mmi;
97
98 dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_mmi;
99 dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_mmi;
100 dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_mmi;
101 dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_mmi;
102 dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_mmi;
103 dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_mmi;
104 dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_mmi;
105 dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_mmi;
106
107 dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_mmi;
108 dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_mmi;
109 dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_mmi;
110 dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_mmi;
111 dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_mmi;
112 dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_mmi;
113 dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_mmi;
114 dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_mmi;
115
116 dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilinear16_h_mmi;
117 dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilinear16_h_mmi;
118 dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilinear16_v_mmi;
119 dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilinear16_hv_mmi;
120 dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilinear16_hv_mmi;
121 dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilinear16_v_mmi;
122 dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilinear16_hv_mmi;
123 dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilinear16_hv_mmi;
124
125 dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilinear8_h_mmi;
126 dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilinear8_h_mmi;
127 dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilinear8_v_mmi;
128 dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilinear8_hv_mmi;
129 dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilinear8_hv_mmi;
130 dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilinear8_v_mmi;
131 dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilinear8_hv_mmi;
132 dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilinear8_hv_mmi;
133
134 dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilinear4_h_mmi;
135 dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilinear4_h_mmi;
136 dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilinear4_v_mmi;
137 dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilinear4_hv_mmi;
138 dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilinear4_hv_mmi;
139 dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilinear4_v_mmi;
140 dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilinear4_hv_mmi;
141 dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilinear4_hv_mmi;
142
143 dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi;
144 dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi;
145
146 dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi;
147 dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi;
148
149 dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_mmi;
150 dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_mmi;
151 dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mmi;
152 dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mmi;
153
154 dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_mmi;
155 dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_mmi;
156 dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmi;
157 dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmi;
158
159 dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmi;
160 dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmi;
161 }
162
163 if (have_msa(cpu_flags)) {
164 dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_msa;
165 dsp->vp8_idct_add = ff_vp8_idct_add_msa;
166 dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_msa;
167 dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_msa;
168 dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_msa;
169
170 VP8_MC_MIPS_FUNC(0, 16);
171 VP8_MC_MIPS_FUNC(1, 8);
172 VP8_MC_MIPS_FUNC(2, 4);
173
174 VP8_BILINEAR_MC_MIPS_FUNC(0, 16);
175 VP8_BILINEAR_MC_MIPS_FUNC(1, 8);
176 VP8_BILINEAR_MC_MIPS_FUNC(2, 4);
177
178 VP8_MC_MIPS_COPY(0, 16);
179 VP8_MC_MIPS_COPY(1, 8);
180
181 dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_msa;
182 dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_msa;
183 dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_msa;
184 dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_msa;
185
186 dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_msa;
187 dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_msa;
188 dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_msa;
189 dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_msa;
190
191 dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_msa;
192 dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_msa;
193 }
194 }
195