1 /*
2 * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@imgtec.com)
3 * 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 #include "libavutil/mips/cpu.h"
23 #include "h264dsp_mips.h"
24
ff_h264qpel_init_mips(H264QpelContext * c,int bit_depth)25 av_cold void ff_h264qpel_init_mips(H264QpelContext *c, int bit_depth)
26 {
27 int cpu_flags = av_get_cpu_flags();
28
29 if (have_mmi(cpu_flags)) {
30 if (bit_depth == 8) {
31 c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_mmi;
32 c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_mmi;
33 c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_mmi;
34 c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_mmi;
35 c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_mmi;
36 c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_mmi;
37 c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_mmi;
38 c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_mmi;
39 c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_mmi;
40 c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_mmi;
41 c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_mmi;
42 c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_mmi;
43 c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_mmi;
44 c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_mmi;
45 c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_mmi;
46 c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_mmi;
47
48 c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_mmi;
49 c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_mmi;
50 c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_mmi;
51 c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_mmi;
52 c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_mmi;
53 c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_mmi;
54 c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_mmi;
55 c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_mmi;
56 c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_mmi;
57 c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_mmi;
58 c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_mmi;
59 c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_mmi;
60 c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_mmi;
61 c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_mmi;
62 c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_mmi;
63 c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_mmi;
64
65 c->put_h264_qpel_pixels_tab[2][0] = ff_put_h264_qpel4_mc00_mmi;
66 c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_mmi;
67 c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_mmi;
68 c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_mmi;
69 c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_mmi;
70 c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_mmi;
71 c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_mmi;
72 c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_mmi;
73 c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_mmi;
74 c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_mmi;
75 c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_mmi;
76 c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_mmi;
77 c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_mmi;
78 c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_mmi;
79 c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_mmi;
80 c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_mmi;
81
82 c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_mmi;
83 c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_mmi;
84 c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_mmi;
85 c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_mmi;
86 c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_mmi;
87 c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_mmi;
88 c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_mmi;
89 c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_mmi;
90 c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_mmi;
91 c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_mmi;
92 c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_mmi;
93 c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_mmi;
94 c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_mmi;
95 c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_mmi;
96 c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_mmi;
97 c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_mmi;
98
99 c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_mmi;
100 c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_mmi;
101 c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_mmi;
102 c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_mmi;
103 c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_mmi;
104 c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_mmi;
105 c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_mmi;
106 c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_mmi;
107 c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_mmi;
108 c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_mmi;
109 c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_mmi;
110 c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_mmi;
111 c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_mmi;
112 c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_mmi;
113 c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_mmi;
114 c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_mmi;
115
116 c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_mmi;
117 c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_mmi;
118 c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_mmi;
119 c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_mmi;
120 c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_mmi;
121 c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_mmi;
122 c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_mmi;
123 c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_mmi;
124 c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_mmi;
125 c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_mmi;
126 c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_mmi;
127 c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_mmi;
128 c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_mmi;
129 c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_mmi;
130 c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_mmi;
131 c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_mmi;
132 }
133 }
134
135 if (have_msa(cpu_flags)) {
136 if (bit_depth == 8) {
137 c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_msa;
138 c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_msa;
139 c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_msa;
140 c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_msa;
141 c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_msa;
142 c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_msa;
143 c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_msa;
144 c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_msa;
145 c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_msa;
146 c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_msa;
147 c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_msa;
148 c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_msa;
149 c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_msa;
150 c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_msa;
151 c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_msa;
152 c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_msa;
153
154 c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_msa;
155 c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_msa;
156 c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_msa;
157 c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_msa;
158 c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_msa;
159 c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_msa;
160 c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_msa;
161 c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_msa;
162 c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_msa;
163 c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_msa;
164 c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_msa;
165 c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_msa;
166 c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_msa;
167 c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_msa;
168 c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_msa;
169 c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_msa;
170
171 c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_msa;
172 c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_msa;
173 c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_msa;
174 c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_msa;
175 c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_msa;
176 c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_msa;
177 c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_msa;
178 c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_msa;
179 c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_msa;
180 c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_msa;
181 c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_msa;
182 c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_msa;
183 c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_msa;
184 c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_msa;
185 c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_msa;
186
187 c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_msa;
188 c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_msa;
189 c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_msa;
190 c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_msa;
191 c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_msa;
192 c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_msa;
193 c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_msa;
194 c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_msa;
195 c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_msa;
196 c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_msa;
197 c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_msa;
198 c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_msa;
199 c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_msa;
200 c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_msa;
201 c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_msa;
202 c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_msa;
203
204 c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_msa;
205 c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_msa;
206 c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_msa;
207 c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_msa;
208 c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_msa;
209 c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_msa;
210 c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_msa;
211 c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_msa;
212 c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_msa;
213 c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_msa;
214 c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_msa;
215 c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_msa;
216 c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_msa;
217 c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_msa;
218 c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_msa;
219 c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_msa;
220
221 c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_msa;
222 c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_msa;
223 c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_msa;
224 c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_msa;
225 c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_msa;
226 c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_msa;
227 c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_msa;
228 c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_msa;
229 c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_msa;
230 c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_msa;
231 c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_msa;
232 c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_msa;
233 c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_msa;
234 c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_msa;
235 c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_msa;
236 c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_msa;
237 }
238 }
239 }
240