]> git.sesse.net Git - ffmpeg/blob - libavfilter/x86/colorspacedsp_init.c
avformat/mpegtsenc: move is_dvb_subtitle/is_dvb_teletext initialization upwards
[ffmpeg] / libavfilter / x86 / colorspacedsp_init.c
1 /*
2  * Copyright (c) 2016 Ronald S. Bultje <rsbultje@gmail.com>
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg 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  * FFmpeg 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 FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include "libavutil/x86/cpu.h"
22
23 #include "libavfilter/colorspacedsp.h"
24
25 #define decl_yuv2yuv_fn(t) \
26 void ff_yuv2yuv_##t##_sse2(uint8_t *yuv_out[3], const ptrdiff_t yuv_out_stride[3], \
27                            uint8_t *yuv_in[3], const ptrdiff_t yuv_in_stride[3], \
28                            int w, int h, const int16_t yuv2yuv_coeffs[3][3][8], \
29                            const int16_t yuv_offset[2][8])
30
31 #define decl_yuv2yuv_fns(ss) \
32 decl_yuv2yuv_fn(ss##p8to8); \
33 decl_yuv2yuv_fn(ss##p10to8); \
34 decl_yuv2yuv_fn(ss##p12to8); \
35 decl_yuv2yuv_fn(ss##p8to10); \
36 decl_yuv2yuv_fn(ss##p10to10); \
37 decl_yuv2yuv_fn(ss##p12to10); \
38 decl_yuv2yuv_fn(ss##p8to12); \
39 decl_yuv2yuv_fn(ss##p10to12); \
40 decl_yuv2yuv_fn(ss##p12to12)
41
42 decl_yuv2yuv_fns(420);
43 decl_yuv2yuv_fns(422);
44 decl_yuv2yuv_fns(444);
45
46 #define decl_yuv2rgb_fn(t) \
47 void ff_yuv2rgb_##t##_sse2(int16_t *rgb_out[3], ptrdiff_t rgb_stride, \
48                            uint8_t *yuv_in[3], const ptrdiff_t yuv_stride[3], \
49                            int w, int h, const int16_t coeff[3][3][8], \
50                            const int16_t yuv_offset[8])
51
52 #define decl_yuv2rgb_fns(ss) \
53 decl_yuv2rgb_fn(ss##p8); \
54 decl_yuv2rgb_fn(ss##p10); \
55 decl_yuv2rgb_fn(ss##p12)
56
57 decl_yuv2rgb_fns(420);
58 decl_yuv2rgb_fns(422);
59 decl_yuv2rgb_fns(444);
60
61 #define decl_rgb2yuv_fn(t) \
62 void ff_rgb2yuv_##t##_sse2(uint8_t *yuv_out[3], const ptrdiff_t yuv_stride[3], \
63                            int16_t *rgb_in[3], ptrdiff_t rgb_stride, \
64                            int w, int h, const int16_t coeff[3][3][8], \
65                            const int16_t yuv_offset[8])
66
67 #define decl_rgb2yuv_fns(ss) \
68 decl_rgb2yuv_fn(ss##p8); \
69 decl_rgb2yuv_fn(ss##p10); \
70 decl_rgb2yuv_fn(ss##p12)
71
72 decl_rgb2yuv_fns(420);
73 decl_rgb2yuv_fns(422);
74 decl_rgb2yuv_fns(444);
75
76 void ff_multiply3x3_sse2(int16_t *data[3], ptrdiff_t stride, int w, int h,
77                          const int16_t coeff[3][3][8]);
78
79 void ff_colorspacedsp_x86_init(ColorSpaceDSPContext *dsp)
80 {
81     int cpu_flags = av_get_cpu_flags();
82
83     if (ARCH_X86_64 && EXTERNAL_SSE2(cpu_flags)) {
84 #define assign_yuv2yuv_fns(ss) \
85         dsp->yuv2yuv[BPP_8 ][BPP_8 ][SS_##ss] = ff_yuv2yuv_##ss##p8to8_sse2; \
86         dsp->yuv2yuv[BPP_8 ][BPP_10][SS_##ss] = ff_yuv2yuv_##ss##p8to10_sse2; \
87         dsp->yuv2yuv[BPP_8 ][BPP_12][SS_##ss] = ff_yuv2yuv_##ss##p8to12_sse2; \
88         dsp->yuv2yuv[BPP_10][BPP_8 ][SS_##ss] = ff_yuv2yuv_##ss##p10to8_sse2; \
89         dsp->yuv2yuv[BPP_10][BPP_10][SS_##ss] = ff_yuv2yuv_##ss##p10to10_sse2; \
90         dsp->yuv2yuv[BPP_10][BPP_12][SS_##ss] = ff_yuv2yuv_##ss##p10to12_sse2; \
91         dsp->yuv2yuv[BPP_12][BPP_8 ][SS_##ss] = ff_yuv2yuv_##ss##p12to8_sse2; \
92         dsp->yuv2yuv[BPP_12][BPP_10][SS_##ss] = ff_yuv2yuv_##ss##p12to10_sse2; \
93         dsp->yuv2yuv[BPP_12][BPP_12][SS_##ss] = ff_yuv2yuv_##ss##p12to12_sse2
94
95         assign_yuv2yuv_fns(420);
96         assign_yuv2yuv_fns(422);
97         assign_yuv2yuv_fns(444);
98
99 #define assign_yuv2rgb_fns(ss) \
100         dsp->yuv2rgb[BPP_8 ][SS_##ss] = ff_yuv2rgb_##ss##p8_sse2; \
101         dsp->yuv2rgb[BPP_10][SS_##ss] = ff_yuv2rgb_##ss##p10_sse2; \
102         dsp->yuv2rgb[BPP_12][SS_##ss] = ff_yuv2rgb_##ss##p12_sse2
103
104         assign_yuv2rgb_fns(420);
105         assign_yuv2rgb_fns(422);
106         assign_yuv2rgb_fns(444);
107
108 #define assign_rgb2yuv_fns(ss) \
109         dsp->rgb2yuv[BPP_8 ][SS_##ss] = ff_rgb2yuv_##ss##p8_sse2; \
110         dsp->rgb2yuv[BPP_10][SS_##ss] = ff_rgb2yuv_##ss##p10_sse2; \
111         dsp->rgb2yuv[BPP_12][SS_##ss] = ff_rgb2yuv_##ss##p12_sse2
112
113         assign_rgb2yuv_fns(420);
114         assign_rgb2yuv_fns(422);
115         assign_rgb2yuv_fns(444);
116
117         dsp->multiply3x3 = ff_multiply3x3_sse2;
118     }
119 }