rv40: NEON optimised rv40 qpel motion compensation
[ffmpeg.git] / libavcodec / arm / rv40dsp_init_neon.c
1 /*
2  * Copyright (c) 2011 Janne Grunau <janne-libav@jannau.net>
3  *
4  * This file is part of Libav.
5  *
6  * Libav is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * Libav is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with Libav; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include <stdint.h>
22
23 #include "libavcodec/avcodec.h"
24 #include "libavcodec/rv34dsp.h"
25
26 #define DECL_QPEL3(type, w, pos) \
27     void ff_##type##_rv40_qpel##w##_mc##pos##_neon(uint8_t *dst, uint8_t *src,\
28                                                    int stride)
29 #define DECL_QPEL2(w, pos)                      \
30     DECL_QPEL3(put, w, pos);                    \
31     DECL_QPEL3(avg, w, pos)
32
33 #define DECL_QPEL_XY(x, y)                      \
34     DECL_QPEL2(16, x ## y);                     \
35     DECL_QPEL2(8,  x ## y)
36
37 #define DECL_QPEL_Y(y)                          \
38     DECL_QPEL_XY(0, y);                         \
39     DECL_QPEL_XY(1, y);                         \
40     DECL_QPEL_XY(2, y);                         \
41     DECL_QPEL_XY(3, y);                         \
42
43 DECL_QPEL_Y(0);
44 DECL_QPEL_Y(1);
45 DECL_QPEL_Y(2);
46 DECL_QPEL_Y(3);
47
48 void ff_put_rv40_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
49 void ff_put_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
50
51 void ff_avg_rv40_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
52 void ff_avg_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
53
54 void ff_rv40_weight_func_16_neon(uint8_t *, uint8_t *, uint8_t *, int, int, int);
55 void ff_rv40_weight_func_8_neon(uint8_t *, uint8_t *, uint8_t *, int, int, int);
56
57 void ff_rv40dsp_init_neon(RV34DSPContext *c, DSPContext* dsp)
58 {
59     c->put_pixels_tab[0][ 1] = ff_put_rv40_qpel16_mc10_neon;
60     c->put_pixels_tab[0][ 3] = ff_put_rv40_qpel16_mc30_neon;
61     c->put_pixels_tab[0][ 4] = ff_put_rv40_qpel16_mc01_neon;
62     c->put_pixels_tab[0][ 5] = ff_put_rv40_qpel16_mc11_neon;
63     c->put_pixels_tab[0][ 6] = ff_put_rv40_qpel16_mc21_neon;
64     c->put_pixels_tab[0][ 7] = ff_put_rv40_qpel16_mc31_neon;
65     c->put_pixels_tab[0][ 9] = ff_put_rv40_qpel16_mc12_neon;
66     c->put_pixels_tab[0][10] = ff_put_rv40_qpel16_mc22_neon;
67     c->put_pixels_tab[0][11] = ff_put_rv40_qpel16_mc32_neon;
68     c->put_pixels_tab[0][12] = ff_put_rv40_qpel16_mc03_neon;
69     c->put_pixels_tab[0][13] = ff_put_rv40_qpel16_mc13_neon;
70     c->put_pixels_tab[0][14] = ff_put_rv40_qpel16_mc23_neon;
71     c->put_pixels_tab[0][15] = ff_put_rv40_qpel16_mc33_neon;
72     c->avg_pixels_tab[0][ 1] = ff_avg_rv40_qpel16_mc10_neon;
73     c->avg_pixels_tab[0][ 3] = ff_avg_rv40_qpel16_mc30_neon;
74     c->avg_pixels_tab[0][ 4] = ff_avg_rv40_qpel16_mc01_neon;
75     c->avg_pixels_tab[0][ 5] = ff_avg_rv40_qpel16_mc11_neon;
76     c->avg_pixels_tab[0][ 6] = ff_avg_rv40_qpel16_mc21_neon;
77     c->avg_pixels_tab[0][ 7] = ff_avg_rv40_qpel16_mc31_neon;
78     c->avg_pixels_tab[0][ 9] = ff_avg_rv40_qpel16_mc12_neon;
79     c->avg_pixels_tab[0][10] = ff_avg_rv40_qpel16_mc22_neon;
80     c->avg_pixels_tab[0][11] = ff_avg_rv40_qpel16_mc32_neon;
81     c->avg_pixels_tab[0][12] = ff_avg_rv40_qpel16_mc03_neon;
82     c->avg_pixels_tab[0][13] = ff_avg_rv40_qpel16_mc13_neon;
83     c->avg_pixels_tab[0][14] = ff_avg_rv40_qpel16_mc23_neon;
84     c->avg_pixels_tab[0][15] = ff_avg_rv40_qpel16_mc33_neon;
85     c->put_pixels_tab[1][ 1] = ff_put_rv40_qpel8_mc10_neon;
86     c->put_pixels_tab[1][ 3] = ff_put_rv40_qpel8_mc30_neon;
87     c->put_pixels_tab[1][ 4] = ff_put_rv40_qpel8_mc01_neon;
88     c->put_pixels_tab[1][ 5] = ff_put_rv40_qpel8_mc11_neon;
89     c->put_pixels_tab[1][ 6] = ff_put_rv40_qpel8_mc21_neon;
90     c->put_pixels_tab[1][ 7] = ff_put_rv40_qpel8_mc31_neon;
91     c->put_pixels_tab[1][ 9] = ff_put_rv40_qpel8_mc12_neon;
92     c->put_pixels_tab[1][10] = ff_put_rv40_qpel8_mc22_neon;
93     c->put_pixels_tab[1][11] = ff_put_rv40_qpel8_mc32_neon;
94     c->put_pixels_tab[1][12] = ff_put_rv40_qpel8_mc03_neon;
95     c->put_pixels_tab[1][13] = ff_put_rv40_qpel8_mc13_neon;
96     c->put_pixels_tab[1][14] = ff_put_rv40_qpel8_mc23_neon;
97     c->put_pixels_tab[1][15] = ff_put_rv40_qpel8_mc33_neon;
98     c->avg_pixels_tab[1][ 1] = ff_avg_rv40_qpel8_mc10_neon;
99     c->avg_pixels_tab[1][ 3] = ff_avg_rv40_qpel8_mc30_neon;
100     c->avg_pixels_tab[1][ 4] = ff_avg_rv40_qpel8_mc01_neon;
101     c->avg_pixels_tab[1][ 5] = ff_avg_rv40_qpel8_mc11_neon;
102     c->avg_pixels_tab[1][ 6] = ff_avg_rv40_qpel8_mc21_neon;
103     c->avg_pixels_tab[1][ 7] = ff_avg_rv40_qpel8_mc31_neon;
104     c->avg_pixels_tab[1][ 9] = ff_avg_rv40_qpel8_mc12_neon;
105     c->avg_pixels_tab[1][10] = ff_avg_rv40_qpel8_mc22_neon;
106     c->avg_pixels_tab[1][11] = ff_avg_rv40_qpel8_mc32_neon;
107     c->avg_pixels_tab[1][12] = ff_avg_rv40_qpel8_mc03_neon;
108     c->avg_pixels_tab[1][13] = ff_avg_rv40_qpel8_mc13_neon;
109     c->avg_pixels_tab[1][14] = ff_avg_rv40_qpel8_mc23_neon;
110     c->avg_pixels_tab[1][15] = ff_avg_rv40_qpel8_mc33_neon;
111
112     c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_neon;
113     c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_neon;
114     c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_neon;
115     c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_neon;
116
117     c->rv40_weight_pixels_tab[0] = ff_rv40_weight_func_16_neon;
118     c->rv40_weight_pixels_tab[1] = ff_rv40_weight_func_8_neon;
119 }