hevcdsp: split the epel functions by width
[libav.git] / libavcodec / hevcdsp.c
1 /*
2 * HEVC video decoder
3 *
4 * Copyright (C) 2012 - 2013 Guillaume Martres
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #include "hevcdsp.h"
24
25 static const int8_t transform[32][32] = {
26 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
27 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
28 { 90, 90, 88, 85, 82, 78, 73, 67, 61, 54, 46, 38, 31, 22, 13, 4,
29 -4, -13, -22, -31, -38, -46, -54, -61, -67, -73, -78, -82, -85, -88, -90, -90 },
30 { 90, 87, 80, 70, 57, 43, 25, 9, -9, -25, -43, -57, -70, -80, -87, -90,
31 -90, -87, -80, -70, -57, -43, -25, -9, 9, 25, 43, 57, 70, 80, 87, 90 },
32 { 90, 82, 67, 46, 22, -4, -31, -54, -73, -85, -90, -88, -78, -61, -38, -13,
33 13, 38, 61, 78, 88, 90, 85, 73, 54, 31, 4, -22, -46, -67, -82, -90 },
34 { 89, 75, 50, 18, -18, -50, -75, -89, -89, -75, -50, -18, 18, 50, 75, 89,
35 89, 75, 50, 18, -18, -50, -75, -89, -89, -75, -50, -18, 18, 50, 75, 89 },
36 { 88, 67, 31, -13, -54, -82, -90, -78, -46, -4, 38, 73, 90, 85, 61, 22,
37 -22, -61, -85, -90, -73, -38, 4, 46, 78, 90, 82, 54, 13, -31, -67, -88 },
38 { 87, 57, 9, -43, -80, -90, -70, -25, 25, 70, 90, 80, 43, -9, -57, -87,
39 -87, -57, -9, 43, 80, 90, 70, 25, -25, -70, -90, -80, -43, 9, 57, 87 },
40 { 85, 46, -13, -67, -90, -73, -22, 38, 82, 88, 54, -4, -61, -90, -78, -31,
41 31, 78, 90, 61, 4, -54, -88, -82, -38, 22, 73, 90, 67, 13, -46, -85 },
42 { 83, 36, -36, -83, -83, -36, 36, 83, 83, 36, -36, -83, -83, -36, 36, 83,
43 83, 36, -36, -83, -83, -36, 36, 83, 83, 36, -36, -83, -83, -36, 36, 83 },
44 { 82, 22, -54, -90, -61, 13, 78, 85, 31, -46, -90, -67, 4, 73, 88, 38,
45 -38, -88, -73, -4, 67, 90, 46, -31, -85, -78, -13, 61, 90, 54, -22, -82 },
46 { 80, 9, -70, -87, -25, 57, 90, 43, -43, -90, -57, 25, 87, 70, -9, -80,
47 -80, -9, 70, 87, 25, -57, -90, -43, 43, 90, 57, -25, -87, -70, 9, 80 },
48 { 78, -4, -82, -73, 13, 85, 67, -22, -88, -61, 31, 90, 54, -38, -90, -46,
49 46, 90, 38, -54, -90, -31, 61, 88, 22, -67, -85, -13, 73, 82, 4, -78 },
50 { 75, -18, -89, -50, 50, 89, 18, -75, -75, 18, 89, 50, -50, -89, -18, 75,
51 75, -18, -89, -50, 50, 89, 18, -75, -75, 18, 89, 50, -50, -89, -18, 75 },
52 { 73, -31, -90, -22, 78, 67, -38, -90, -13, 82, 61, -46, -88, -4, 85, 54,
53 -54, -85, 4, 88, 46, -61, -82, 13, 90, 38, -67, -78, 22, 90, 31, -73 },
54 { 70, -43, -87, 9, 90, 25, -80, -57, 57, 80, -25, -90, -9, 87, 43, -70,
55 -70, 43, 87, -9, -90, -25, 80, 57, -57, -80, 25, 90, 9, -87, -43, 70 },
56 { 67, -54, -78, 38, 85, -22, -90, 4, 90, 13, -88, -31, 82, 46, -73, -61,
57 61, 73, -46, -82, 31, 88, -13, -90, -4, 90, 22, -85, -38, 78, 54, -67 },
58 { 64, -64, -64, 64, 64, -64, -64, 64, 64, -64, -64, 64, 64, -64, -64, 64,
59 64, -64, -64, 64, 64, -64, -64, 64, 64, -64, -64, 64, 64, -64, -64, 64 },
60 { 61, -73, -46, 82, 31, -88, -13, 90, -4, -90, 22, 85, -38, -78, 54, 67,
61 -67, -54, 78, 38, -85, -22, 90, 4, -90, 13, 88, -31, -82, 46, 73, -61 },
62 { 57, -80, -25, 90, -9, -87, 43, 70, -70, -43, 87, 9, -90, 25, 80, -57,
63 -57, 80, 25, -90, 9, 87, -43, -70, 70, 43, -87, -9, 90, -25, -80, 57 },
64 { 54, -85, -4, 88, -46, -61, 82, 13, -90, 38, 67, -78, -22, 90, -31, -73,
65 73, 31, -90, 22, 78, -67, -38, 90, -13, -82, 61, 46, -88, 4, 85, -54 },
66 { 50, -89, 18, 75, -75, -18, 89, -50, -50, 89, -18, -75, 75, 18, -89, 50,
67 50, -89, 18, 75, -75, -18, 89, -50, -50, 89, -18, -75, 75, 18, -89, 50 },
68 { 46, -90, 38, 54, -90, 31, 61, -88, 22, 67, -85, 13, 73, -82, 4, 78,
69 -78, -4, 82, -73, -13, 85, -67, -22, 88, -61, -31, 90, -54, -38, 90, -46 },
70 { 43, -90, 57, 25, -87, 70, 9, -80, 80, -9, -70, 87, -25, -57, 90, -43,
71 -43, 90, -57, -25, 87, -70, -9, 80, -80, 9, 70, -87, 25, 57, -90, 43 },
72 { 38, -88, 73, -4, -67, 90, -46, -31, 85, -78, 13, 61, -90, 54, 22, -82,
73 82, -22, -54, 90, -61, -13, 78, -85, 31, 46, -90, 67, 4, -73, 88, -38 },
74 { 36, -83, 83, -36, -36, 83, -83, 36, 36, -83, 83, -36, -36, 83, -83, 36,
75 36, -83, 83, -36, -36, 83, -83, 36, 36, -83, 83, -36, -36, 83, -83, 36 },
76 { 31, -78, 90, -61, 4, 54, -88, 82, -38, -22, 73, -90, 67, -13, -46, 85,
77 -85, 46, 13, -67, 90, -73, 22, 38, -82, 88, -54, -4, 61, -90, 78, -31 },
78 { 25, -70, 90, -80, 43, 9, -57, 87, -87, 57, -9, -43, 80, -90, 70, -25,
79 -25, 70, -90, 80, -43, -9, 57, -87, 87, -57, 9, 43, -80, 90, -70, 25 },
80 { 22, -61, 85, -90, 73, -38, -4, 46, -78, 90, -82, 54, -13, -31, 67, -88,
81 88, -67, 31, 13, -54, 82, -90, 78, -46, 4, 38, -73, 90, -85, 61, -22 },
82 { 18, -50, 75, -89, 89, -75, 50, -18, -18, 50, -75, 89, -89, 75, -50, 18,
83 18, -50, 75, -89, 89, -75, 50, -18, -18, 50, -75, 89, -89, 75, -50, 18 },
84 { 13, -38, 61, -78, 88, -90, 85, -73, 54, -31, 4, 22, -46, 67, -82, 90,
85 -90, 82, -67, 46, -22, -4, 31, -54, 73, -85, 90, -88, 78, -61, 38, -13 },
86 { 9, -25, 43, -57, 70, -80, 87, -90, 90, -87, 80, -70, 57, -43, 25, -9,
87 -9, 25, -43, 57, -70, 80, -87, 90, -90, 87, -80, 70, -57, 43, -25, 9 },
88 { 4, -13, 22, -31, 38, -46, 54, -61, 67, -73, 78, -82, 85, -88, 90, -90,
89 90, -90, 88, -85, 82, -78, 73, -67, 61, -54, 46, -38, 31, -22, 13, -4 },
90 };
91
92 DECLARE_ALIGNED(16, const int8_t, ff_hevc_epel_filters[7][16]) = {
93 { -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2 },
94 { -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2 },
95 { -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4 },
96 { -4, 36, 36, -4, -4, 36, 36, -4, -4, 36, 36, -4, -4, 36, 36, -4 },
97 { -4, 28, 46, -6, -4, 28, 46, -6, -4, 28, 46, -6, -4, 28, 46, -6 },
98 { -2, 16, 54, -4, -2, 16, 54, -4, -2, 16, 54, -4, -2, 16, 54, -4 },
99 { -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2 },
100 };
101
102 #define BIT_DEPTH 8
103 #include "hevcdsp_template.c"
104 #undef BIT_DEPTH
105
106 #define BIT_DEPTH 9
107 #include "hevcdsp_template.c"
108 #undef BIT_DEPTH
109
110 #define BIT_DEPTH 10
111 #include "hevcdsp_template.c"
112 #undef BIT_DEPTH
113
114 void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int bit_depth)
115 {
116 #undef FUNC
117 #define FUNC(a, depth) a ## _ ## depth
118
119 #define QPEL_FUNC(i, width, depth) \
120 hevcdsp->put_hevc_qpel[0][0][i] = FUNC(put_hevc_qpel_pixels_ ## width, depth); \
121 hevcdsp->put_hevc_qpel[0][1][i] = FUNC(put_hevc_qpel_h_ ## width, depth); \
122 hevcdsp->put_hevc_qpel[1][0][i] = FUNC(put_hevc_qpel_v_ ## width, depth); \
123 hevcdsp->put_hevc_qpel[1][1][i] = FUNC(put_hevc_qpel_hv_ ## width, depth); \
124
125 #define EPEL_FUNC(i, width, depth) \
126 hevcdsp->put_hevc_epel[0][0][i] = FUNC(put_hevc_epel_pixels_ ## width, depth); \
127 hevcdsp->put_hevc_epel[0][1][i] = FUNC(put_hevc_epel_h_ ## width, depth); \
128 hevcdsp->put_hevc_epel[1][0][i] = FUNC(put_hevc_epel_v_ ## width, depth); \
129 hevcdsp->put_hevc_epel[1][1][i] = FUNC(put_hevc_epel_hv_ ## width, depth); \
130
131 #define HEVC_DSP(depth) \
132 hevcdsp->put_pcm = FUNC(put_pcm, depth); \
133 hevcdsp->transquant_bypass[0] = FUNC(transquant_bypass4x4, depth); \
134 hevcdsp->transquant_bypass[1] = FUNC(transquant_bypass8x8, depth); \
135 hevcdsp->transquant_bypass[2] = FUNC(transquant_bypass16x16, depth); \
136 hevcdsp->transquant_bypass[3] = FUNC(transquant_bypass32x32, depth); \
137 hevcdsp->transform_skip = FUNC(transform_skip, depth); \
138 hevcdsp->transform_4x4_luma_add = FUNC(transform_4x4_luma_add, depth); \
139 hevcdsp->transform_add[0] = FUNC(transform_4x4_add, depth); \
140 hevcdsp->transform_add[1] = FUNC(transform_8x8_add, depth); \
141 hevcdsp->transform_add[2] = FUNC(transform_16x16_add, depth); \
142 hevcdsp->transform_add[3] = FUNC(transform_32x32_add, depth); \
143 \
144 hevcdsp->sao_band_filter[0] = FUNC(sao_band_filter_0, depth); \
145 hevcdsp->sao_band_filter[1] = FUNC(sao_band_filter_1, depth); \
146 hevcdsp->sao_band_filter[2] = FUNC(sao_band_filter_2, depth); \
147 hevcdsp->sao_band_filter[3] = FUNC(sao_band_filter_3, depth); \
148 \
149 hevcdsp->sao_edge_filter[0] = FUNC(sao_edge_filter_0, depth); \
150 hevcdsp->sao_edge_filter[1] = FUNC(sao_edge_filter_1, depth); \
151 hevcdsp->sao_edge_filter[2] = FUNC(sao_edge_filter_2, depth); \
152 hevcdsp->sao_edge_filter[3] = FUNC(sao_edge_filter_3, depth); \
153 \
154 QPEL_FUNC(0, 4, depth); \
155 QPEL_FUNC(1, 8, depth); \
156 QPEL_FUNC(2, 12, depth); \
157 QPEL_FUNC(3, 16, depth); \
158 QPEL_FUNC(4, 24, depth); \
159 QPEL_FUNC(5, 32, depth); \
160 QPEL_FUNC(6, 48, depth); \
161 QPEL_FUNC(7, 64, depth); \
162 \
163 EPEL_FUNC(0, 2, depth); \
164 EPEL_FUNC(1, 4, depth); \
165 EPEL_FUNC(2, 6, depth); \
166 EPEL_FUNC(3, 8, depth); \
167 EPEL_FUNC(4, 12, depth); \
168 EPEL_FUNC(5, 16, depth); \
169 EPEL_FUNC(6, 24, depth); \
170 EPEL_FUNC(7, 32, depth); \
171 \
172 hevcdsp->put_unweighted_pred = FUNC(put_unweighted_pred, depth); \
173 hevcdsp->put_unweighted_pred_avg = FUNC(put_unweighted_pred_avg, depth); \
174 \
175 hevcdsp->weighted_pred = FUNC(weighted_pred, depth); \
176 hevcdsp->weighted_pred_avg = FUNC(weighted_pred_avg, depth); \
177 \
178 hevcdsp->hevc_h_loop_filter_luma = FUNC(hevc_h_loop_filter_luma, depth); \
179 hevcdsp->hevc_v_loop_filter_luma = FUNC(hevc_v_loop_filter_luma, depth); \
180 hevcdsp->hevc_h_loop_filter_chroma = FUNC(hevc_h_loop_filter_chroma, depth); \
181 hevcdsp->hevc_v_loop_filter_chroma = FUNC(hevc_v_loop_filter_chroma, depth); \
182 hevcdsp->hevc_h_loop_filter_luma_c = FUNC(hevc_h_loop_filter_luma, depth); \
183 hevcdsp->hevc_v_loop_filter_luma_c = FUNC(hevc_v_loop_filter_luma, depth); \
184 hevcdsp->hevc_h_loop_filter_chroma_c = FUNC(hevc_h_loop_filter_chroma, depth); \
185 hevcdsp->hevc_v_loop_filter_chroma_c = FUNC(hevc_v_loop_filter_chroma, depth);
186
187 switch (bit_depth) {
188 case 9:
189 HEVC_DSP(9);
190 break;
191 case 10:
192 HEVC_DSP(10);
193 break;
194 default:
195 HEVC_DSP(8);
196 break;
197 }
198
199 if (ARCH_X86)
200 ff_hevc_dsp_init_x86(hevcdsp, bit_depth);
201 }