Merge commit 'a7a07cc98ac548297b5b0628cb81280e11952e3f'
[ffmpeg.git] / libavcodec / x86 / diracdsp_mmx.c
1 /*
2  * Copyright (C) 2010 David Conrad
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 "dsputil_x86.h"
22 #include "diracdsp_mmx.h"
23
24 void ff_put_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
25 void ff_put_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
26 void ff_put_signed_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
27 void ff_put_signed_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
28
29 #define HPEL_FILTER(MMSIZE, EXT)                                                             \
30     void ff_dirac_hpel_filter_v_ ## EXT(uint8_t *, const uint8_t *, int, int);               \
31     void ff_dirac_hpel_filter_h_ ## EXT(uint8_t *, const uint8_t *, int);                    \
32                                                                                              \
33     static void dirac_hpel_filter_ ## EXT(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,       \
34                                           const uint8_t *src, int stride, int width, int height)   \
35     {                                                                                        \
36         while( height-- )                                                                    \
37         {                                                                                    \
38             ff_dirac_hpel_filter_v_ ## EXT(dstv-MMSIZE, src-MMSIZE, stride, width+MMSIZE+5); \
39             ff_dirac_hpel_filter_h_ ## EXT(dsth, src, width);                                \
40             ff_dirac_hpel_filter_h_ ## EXT(dstc, dstv, width);                               \
41                                                                                              \
42             dsth += stride;                                                                  \
43             dstv += stride;                                                                  \
44             dstc += stride;                                                                  \
45             src  += stride;                                                                  \
46         }                                                                                    \
47     }
48
49 #if !ARCH_X86_64
50 HPEL_FILTER(8, mmx)
51 #endif
52 HPEL_FILTER(16, sse2)
53
54 #define PIXFUNC(PFX, IDX, EXT)                                                   \
55     /*MMXDISABLEDc->PFX ## _dirac_pixels_tab[0][IDX] = ff_ ## PFX ## _dirac_pixels8_ ## EXT;*/  \
56     c->PFX ## _dirac_pixels_tab[1][IDX] = ff_ ## PFX ## _dirac_pixels16_ ## EXT; \
57     c->PFX ## _dirac_pixels_tab[2][IDX] = ff_ ## PFX ## _dirac_pixels32_ ## EXT
58
59 void ff_diracdsp_init_mmx(DiracDSPContext* c)
60 {
61     int mm_flags = av_get_cpu_flags();
62
63     if (!(mm_flags & AV_CPU_FLAG_MMX))
64         return;
65
66 #if HAVE_YASM
67     c->add_dirac_obmc[0] = ff_add_dirac_obmc8_mmx;
68 #if !ARCH_X86_64
69     c->add_dirac_obmc[1] = ff_add_dirac_obmc16_mmx;
70     c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
71     c->dirac_hpel_filter = dirac_hpel_filter_mmx;
72     c->add_rect_clamped = ff_add_rect_clamped_mmx;
73     c->put_signed_rect_clamped = ff_put_signed_rect_clamped_mmx;
74 #endif
75 #endif
76
77 #if HAVE_MMX_INLINE
78     PIXFUNC(put, 0, mmx);
79     PIXFUNC(avg, 0, mmx);
80 #endif
81
82 #if HAVE_MMXEXT_INLINE
83     if (mm_flags & AV_CPU_FLAG_MMX2) {
84         PIXFUNC(avg, 0, mmxext);
85     }
86 #endif
87
88     if (mm_flags & AV_CPU_FLAG_SSE2) {
89 #if HAVE_YASM
90         c->dirac_hpel_filter = dirac_hpel_filter_sse2;
91         c->add_rect_clamped = ff_add_rect_clamped_sse2;
92         c->put_signed_rect_clamped = ff_put_signed_rect_clamped_sse2;
93
94         c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
95         c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;
96 #endif
97 #if HAVE_SSE2_INLINE
98         c->put_dirac_pixels_tab[1][0] = ff_put_dirac_pixels16_sse2;
99         c->avg_dirac_pixels_tab[1][0] = ff_avg_dirac_pixels16_sse2;
100         c->put_dirac_pixels_tab[2][0] = ff_put_dirac_pixels32_sse2;
101         c->avg_dirac_pixels_tab[2][0] = ff_avg_dirac_pixels32_sse2;
102 #endif
103     }
104 }