avcodec/h264: mmx2, sse2, avx 10-bit h chroma deblock/loop filter
[ffmpeg.git] / libavcodec / x86 / vc1dsp_init.c
1 /*
2  * VC-1 and WMV3 - DSP functions MMX-optimized
3  * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>
4  *
5  * Permission is hereby granted, free of charge, to any person
6  * obtaining a copy of this software and associated documentation
7  * files (the "Software"), to deal in the Software without
8  * restriction, including without limitation the rights to use,
9  * copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following
12  * conditions:
13  *
14  * The above copyright notice and this permission notice shall be
15  * included in all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
19  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
20  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
21  * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
22  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
24  * OTHER DEALINGS IN THE SOFTWARE.
25  */
26
27 #include "libavutil/attributes.h"
28 #include "libavutil/cpu.h"
29 #include "libavutil/x86/cpu.h"
30 #include "libavutil/x86/asm.h"
31 #include "libavcodec/vc1dsp.h"
32 #include "fpel.h"
33 #include "vc1dsp.h"
34 #include "config.h"
35
36 #define LOOP_FILTER(EXT) \
37 void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
38 void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
39 void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
40 void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
41 \
42 static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
43 { \
44     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
45     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
46 } \
47 \
48 static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
49 { \
50     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
51     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
52 }
53
54 #if HAVE_YASM
55 LOOP_FILTER(mmxext)
56 LOOP_FILTER(sse2)
57 LOOP_FILTER(ssse3)
58
59 void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
60
61 static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
62 {
63     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
64     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
65 }
66
67 #define DECLARE_FUNCTION(OP, DEPTH, INSN)                       \
68     static void OP##vc1_mspel_mc00_##DEPTH##INSN(uint8_t *dst,          \
69                              const uint8_t *src, ptrdiff_t stride, int rnd) \
70     {                                                                       \
71         ff_ ## OP ## pixels ## DEPTH ## INSN(dst, src, stride, DEPTH);     \
72     }
73
74 DECLARE_FUNCTION(put_,  8, _mmx)
75 DECLARE_FUNCTION(put_, 16, _mmx)
76 DECLARE_FUNCTION(avg_,  8, _mmx)
77 DECLARE_FUNCTION(avg_, 16, _mmx)
78 DECLARE_FUNCTION(avg_,  8, _mmxext)
79 DECLARE_FUNCTION(avg_, 16, _mmxext)
80 DECLARE_FUNCTION(put_, 16, _sse2)
81 DECLARE_FUNCTION(avg_, 16, _sse2)
82
83 #endif /* HAVE_YASM */
84
85 void ff_put_vc1_chroma_mc8_nornd_mmx  (uint8_t *dst, uint8_t *src,
86                                        int stride, int h, int x, int y);
87 void ff_avg_vc1_chroma_mc8_nornd_mmxext(uint8_t *dst, uint8_t *src,
88                                         int stride, int h, int x, int y);
89 void ff_avg_vc1_chroma_mc8_nornd_3dnow(uint8_t *dst, uint8_t *src,
90                                        int stride, int h, int x, int y);
91 void ff_put_vc1_chroma_mc8_nornd_ssse3(uint8_t *dst, uint8_t *src,
92                                        int stride, int h, int x, int y);
93 void ff_avg_vc1_chroma_mc8_nornd_ssse3(uint8_t *dst, uint8_t *src,
94                                        int stride, int h, int x, int y);
95 void ff_vc1_inv_trans_4x4_dc_mmxext(uint8_t *dest, int linesize,
96                                     int16_t *block);
97 void ff_vc1_inv_trans_4x8_dc_mmxext(uint8_t *dest, int linesize,
98                                     int16_t *block);
99 void ff_vc1_inv_trans_8x4_dc_mmxext(uint8_t *dest, int linesize,
100                                     int16_t *block);
101 void ff_vc1_inv_trans_8x8_dc_mmxext(uint8_t *dest, int linesize,
102                                     int16_t *block);
103
104
105 av_cold void ff_vc1dsp_init_x86(VC1DSPContext *dsp)
106 {
107     int cpu_flags = av_get_cpu_flags();
108
109     if (HAVE_6REGS && INLINE_MMX(cpu_flags) && EXTERNAL_MMX(cpu_flags))
110         ff_vc1dsp_init_mmx(dsp);
111
112     if (HAVE_6REGS && INLINE_MMXEXT(cpu_flags) && EXTERNAL_MMXEXT(cpu_flags))
113         ff_vc1dsp_init_mmxext(dsp);
114
115 #define ASSIGN_LF(EXT) \
116         dsp->vc1_v_loop_filter4  = ff_vc1_v_loop_filter4_ ## EXT; \
117         dsp->vc1_h_loop_filter4  = ff_vc1_h_loop_filter4_ ## EXT; \
118         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_ ## EXT; \
119         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_ ## EXT; \
120         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_ ## EXT; \
121         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_ ## EXT
122
123 #if HAVE_YASM
124     if (EXTERNAL_MMX(cpu_flags)) {
125         dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_vc1_chroma_mc8_nornd_mmx;
126
127         dsp->put_vc1_mspel_pixels_tab[1][0]      = put_vc1_mspel_mc00_8_mmx;
128         dsp->put_vc1_mspel_pixels_tab[0][0]      = put_vc1_mspel_mc00_16_mmx;
129         dsp->avg_vc1_mspel_pixels_tab[1][0]      = avg_vc1_mspel_mc00_8_mmx;
130         dsp->avg_vc1_mspel_pixels_tab[0][0]      = avg_vc1_mspel_mc00_16_mmx;
131     }
132     if (EXTERNAL_AMD3DNOW(cpu_flags)) {
133         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_nornd_3dnow;
134     }
135     if (EXTERNAL_MMXEXT(cpu_flags)) {
136         ASSIGN_LF(mmxext);
137         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_nornd_mmxext;
138
139         dsp->avg_vc1_mspel_pixels_tab[1][0]      = avg_vc1_mspel_mc00_8_mmxext;
140         dsp->avg_vc1_mspel_pixels_tab[0][0]      = avg_vc1_mspel_mc00_16_mmxext;
141
142         dsp->vc1_inv_trans_8x8_dc                = ff_vc1_inv_trans_8x8_dc_mmxext;
143         dsp->vc1_inv_trans_4x8_dc                = ff_vc1_inv_trans_4x8_dc_mmxext;
144         dsp->vc1_inv_trans_8x4_dc                = ff_vc1_inv_trans_8x4_dc_mmxext;
145         dsp->vc1_inv_trans_4x4_dc                = ff_vc1_inv_trans_4x4_dc_mmxext;
146     }
147     if (EXTERNAL_SSE2(cpu_flags)) {
148         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_sse2;
149         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse2;
150         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_sse2;
151         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse2;
152
153         dsp->put_vc1_mspel_pixels_tab[0][0]      = put_vc1_mspel_mc00_16_sse2;
154         dsp->avg_vc1_mspel_pixels_tab[0][0]      = avg_vc1_mspel_mc00_16_sse2;
155     }
156     if (EXTERNAL_SSSE3(cpu_flags)) {
157         ASSIGN_LF(ssse3);
158         dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_vc1_chroma_mc8_nornd_ssse3;
159         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_nornd_ssse3;
160     }
161     if (EXTERNAL_SSE4(cpu_flags)) {
162         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse4;
163         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse4;
164     }
165 #endif /* HAVE_YASM */
166 }