avcodec/v210dec: Fix alignment check for AVX2
[ffmpeg.git] / libavfilter / x86 / vf_gradfun_init.c
1 /*
2  * Copyright (C) 2009 Loren Merritt <lorenm@u.washington.edu>
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 "config.h"
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/mem.h"
25 #include "libavutil/x86/cpu.h"
26 #include "libavfilter/gradfun.h"
27
28 void ff_gradfun_filter_line_mmxext(intptr_t x, uint8_t *dst, const uint8_t *src,
29                                    const uint16_t *dc, int thresh,
30                                    const uint16_t *dithers);
31 void ff_gradfun_filter_line_ssse3(intptr_t x, uint8_t *dst, const uint8_t *src,
32                                   const uint16_t *dc, int thresh,
33                                   const uint16_t *dithers);
34
35 void ff_gradfun_blur_line_movdqa_sse2(intptr_t x, uint16_t *buf,
36                                       const uint16_t *buf1, uint16_t *dc,
37                                       const uint8_t *src1, const uint8_t *src2);
38 void ff_gradfun_blur_line_movdqu_sse2(intptr_t x, uint16_t *buf,
39                                       const uint16_t *buf1, uint16_t *dc,
40                                       const uint8_t *src1, const uint8_t *src2);
41
42 #if HAVE_X86ASM
43 static void gradfun_filter_line_mmxext(uint8_t *dst, const uint8_t *src,
44                                        const uint16_t *dc,
45                                        int width, int thresh,
46                                        const uint16_t *dithers)
47 {
48     intptr_t x;
49     if (width & 3) {
50         x = width & ~3;
51         ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2,
52                                  width - x, thresh, dithers);
53         width = x;
54     }
55     x = -width;
56     ff_gradfun_filter_line_mmxext(x, dst + width, src + width, dc + width / 2,
57                                   thresh, dithers);
58 }
59
60 static void gradfun_filter_line_ssse3(uint8_t *dst, const uint8_t *src, const uint16_t *dc,
61                                       int width, int thresh,
62                                       const uint16_t *dithers)
63 {
64     intptr_t x;
65     if (width & 7) {
66         // could be 10% faster if I somehow eliminated this
67         x = width & ~7;
68         ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2,
69                                  width - x, thresh, dithers);
70         width = x;
71     }
72     x = -width;
73     ff_gradfun_filter_line_ssse3(x, dst + width, src + width, dc + width / 2,
74                                  thresh, dithers);
75 }
76
77 static void gradfun_blur_line_sse2(uint16_t *dc, uint16_t *buf, const uint16_t *buf1,
78                                    const uint8_t *src, int src_linesize, int width)
79 {
80     intptr_t x = -2 * width;
81     if (((intptr_t) src | src_linesize) & 15)
82         ff_gradfun_blur_line_movdqu_sse2(x, buf + width, buf1 + width,
83                                          dc + width, src + width * 2,
84                                          src + width * 2 + src_linesize);
85     else
86         ff_gradfun_blur_line_movdqa_sse2(x, buf + width, buf1 + width,
87                                          dc + width, src + width * 2,
88                                          src + width * 2 + src_linesize);
89 }
90 #endif /* HAVE_X86ASM */
91
92 av_cold void ff_gradfun_init_x86(GradFunContext *gf)
93 {
94 #if HAVE_X86ASM
95     int cpu_flags = av_get_cpu_flags();
96
97     if (EXTERNAL_MMXEXT(cpu_flags))
98         gf->filter_line = gradfun_filter_line_mmxext;
99     if (EXTERNAL_SSSE3(cpu_flags))
100         gf->filter_line = gradfun_filter_line_ssse3;
101
102     if (EXTERNAL_SSE2(cpu_flags))
103         gf->blur_line = gradfun_blur_line_sse2;
104 #endif /* HAVE_X86ASM */
105 }