1 /*
2 * ARM NEON optimised DSP functions
3 * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
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 <stdint.h>
23
24 #include "config.h"
25 #include "libavutil/attributes.h"
26 #include "libavutil/arm/cpu.h"
27 #include "libavcodec/h264qpel.h"
28
29 void ff_put_h264_qpel16_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
30 void ff_put_h264_qpel16_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
31 void ff_put_h264_qpel16_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
32 void ff_put_h264_qpel16_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
33 void ff_put_h264_qpel16_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
34 void ff_put_h264_qpel16_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
35 void ff_put_h264_qpel16_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
36 void ff_put_h264_qpel16_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
37 void ff_put_h264_qpel16_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
38 void ff_put_h264_qpel16_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
39 void ff_put_h264_qpel16_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
40 void ff_put_h264_qpel16_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
41 void ff_put_h264_qpel16_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
42 void ff_put_h264_qpel16_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
43 void ff_put_h264_qpel16_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
44 void ff_put_h264_qpel16_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
45
46 void ff_put_h264_qpel8_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
47 void ff_put_h264_qpel8_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
48 void ff_put_h264_qpel8_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
49 void ff_put_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
50 void ff_put_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
51 void ff_put_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
52 void ff_put_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
53 void ff_put_h264_qpel8_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
54 void ff_put_h264_qpel8_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
55 void ff_put_h264_qpel8_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
56 void ff_put_h264_qpel8_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
57 void ff_put_h264_qpel8_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
58 void ff_put_h264_qpel8_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
59 void ff_put_h264_qpel8_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
60 void ff_put_h264_qpel8_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
61 void ff_put_h264_qpel8_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
62
63 void ff_avg_h264_qpel16_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
64 void ff_avg_h264_qpel16_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
65 void ff_avg_h264_qpel16_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
66 void ff_avg_h264_qpel16_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
67 void ff_avg_h264_qpel16_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
68 void ff_avg_h264_qpel16_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
69 void ff_avg_h264_qpel16_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
70 void ff_avg_h264_qpel16_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
71 void ff_avg_h264_qpel16_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
72 void ff_avg_h264_qpel16_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
73 void ff_avg_h264_qpel16_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
74 void ff_avg_h264_qpel16_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
75 void ff_avg_h264_qpel16_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
76 void ff_avg_h264_qpel16_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
77 void ff_avg_h264_qpel16_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
78 void ff_avg_h264_qpel16_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
79
80 void ff_avg_h264_qpel8_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
81 void ff_avg_h264_qpel8_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
82 void ff_avg_h264_qpel8_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
83 void ff_avg_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
84 void ff_avg_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
85 void ff_avg_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
86 void ff_avg_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
87 void ff_avg_h264_qpel8_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
88 void ff_avg_h264_qpel8_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
89 void ff_avg_h264_qpel8_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
90 void ff_avg_h264_qpel8_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
91 void ff_avg_h264_qpel8_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
92 void ff_avg_h264_qpel8_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
93 void ff_avg_h264_qpel8_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
94 void ff_avg_h264_qpel8_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
95 void ff_avg_h264_qpel8_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
96
ff_h264qpel_init_arm(H264QpelContext * c,int bit_depth)97 av_cold void ff_h264qpel_init_arm(H264QpelContext *c, int bit_depth)
98 {
99 const int high_bit_depth = bit_depth > 8;
100 int cpu_flags = av_get_cpu_flags();
101
102 if (have_neon(cpu_flags) && !high_bit_depth) {
103 c->put_h264_qpel_pixels_tab[0][ 0] = ff_put_h264_qpel16_mc00_neon;
104 c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon;
105 c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon;
106 c->put_h264_qpel_pixels_tab[0][ 3] = ff_put_h264_qpel16_mc30_neon;
107 c->put_h264_qpel_pixels_tab[0][ 4] = ff_put_h264_qpel16_mc01_neon;
108 c->put_h264_qpel_pixels_tab[0][ 5] = ff_put_h264_qpel16_mc11_neon;
109 c->put_h264_qpel_pixels_tab[0][ 6] = ff_put_h264_qpel16_mc21_neon;
110 c->put_h264_qpel_pixels_tab[0][ 7] = ff_put_h264_qpel16_mc31_neon;
111 c->put_h264_qpel_pixels_tab[0][ 8] = ff_put_h264_qpel16_mc02_neon;
112 c->put_h264_qpel_pixels_tab[0][ 9] = ff_put_h264_qpel16_mc12_neon;
113 c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_neon;
114 c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_neon;
115 c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_neon;
116 c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_neon;
117 c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_neon;
118 c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
119
120 c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
121 c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
122 c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_neon;
123 c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
124 c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
125 c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
126 c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
127 c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
128 c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
129 c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
130 c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
131 c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
132 c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
133 c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
134 c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
135 c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
136
137 c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;
138 c->avg_h264_qpel_pixels_tab[0][ 1] = ff_avg_h264_qpel16_mc10_neon;
139 c->avg_h264_qpel_pixels_tab[0][ 2] = ff_avg_h264_qpel16_mc20_neon;
140 c->avg_h264_qpel_pixels_tab[0][ 3] = ff_avg_h264_qpel16_mc30_neon;
141 c->avg_h264_qpel_pixels_tab[0][ 4] = ff_avg_h264_qpel16_mc01_neon;
142 c->avg_h264_qpel_pixels_tab[0][ 5] = ff_avg_h264_qpel16_mc11_neon;
143 c->avg_h264_qpel_pixels_tab[0][ 6] = ff_avg_h264_qpel16_mc21_neon;
144 c->avg_h264_qpel_pixels_tab[0][ 7] = ff_avg_h264_qpel16_mc31_neon;
145 c->avg_h264_qpel_pixels_tab[0][ 8] = ff_avg_h264_qpel16_mc02_neon;
146 c->avg_h264_qpel_pixels_tab[0][ 9] = ff_avg_h264_qpel16_mc12_neon;
147 c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_neon;
148 c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_neon;
149 c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_neon;
150 c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_neon;
151 c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_neon;
152 c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_neon;
153
154 c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon;
155 c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_neon;
156 c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_neon;
157 c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_neon;
158 c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_neon;
159 c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_neon;
160 c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_neon;
161 c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_neon;
162 c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_neon;
163 c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_neon;
164 c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_neon;
165 c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_neon;
166 c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_neon;
167 c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_neon;
168 c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_neon;
169 c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
170 }
171 }
172