yading@10
|
1 /*
|
yading@10
|
2 * Copyright (C) 2010 David Conrad
|
yading@10
|
3 *
|
yading@10
|
4 * This file is part of FFmpeg.
|
yading@10
|
5 *
|
yading@10
|
6 * FFmpeg is free software; you can redistribute it and/or
|
yading@10
|
7 * modify it under the terms of the GNU Lesser General Public
|
yading@10
|
8 * License as published by the Free Software Foundation; either
|
yading@10
|
9 * version 2.1 of the License, or (at your option) any later version.
|
yading@10
|
10 *
|
yading@10
|
11 * FFmpeg is distributed in the hope that it will be useful,
|
yading@10
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
yading@10
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
yading@10
|
14 * Lesser General Public License for more details.
|
yading@10
|
15 *
|
yading@10
|
16 * You should have received a copy of the GNU Lesser General Public
|
yading@10
|
17 * License along with FFmpeg; if not, write to the Free Software
|
yading@10
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
yading@10
|
19 */
|
yading@10
|
20
|
yading@10
|
21 #include "dsputil_mmx.h"
|
yading@10
|
22 #include "diracdsp_mmx.h"
|
yading@10
|
23
|
yading@10
|
24 void ff_put_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
|
yading@10
|
25 void ff_put_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
|
yading@10
|
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);
|
yading@10
|
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);
|
yading@10
|
28
|
yading@10
|
29 #define HPEL_FILTER(MMSIZE, EXT) \
|
yading@10
|
30 void ff_dirac_hpel_filter_v_ ## EXT(uint8_t *, const uint8_t *, int, int); \
|
yading@10
|
31 void ff_dirac_hpel_filter_h_ ## EXT(uint8_t *, const uint8_t *, int); \
|
yading@10
|
32 \
|
yading@10
|
33 static void dirac_hpel_filter_ ## EXT(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, \
|
yading@10
|
34 const uint8_t *src, int stride, int width, int height) \
|
yading@10
|
35 { \
|
yading@10
|
36 while( height-- ) \
|
yading@10
|
37 { \
|
yading@10
|
38 ff_dirac_hpel_filter_v_ ## EXT(dstv-MMSIZE, src-MMSIZE, stride, width+MMSIZE+5); \
|
yading@10
|
39 ff_dirac_hpel_filter_h_ ## EXT(dsth, src, width); \
|
yading@10
|
40 ff_dirac_hpel_filter_h_ ## EXT(dstc, dstv, width); \
|
yading@10
|
41 \
|
yading@10
|
42 dsth += stride; \
|
yading@10
|
43 dstv += stride; \
|
yading@10
|
44 dstc += stride; \
|
yading@10
|
45 src += stride; \
|
yading@10
|
46 } \
|
yading@10
|
47 }
|
yading@10
|
48
|
yading@10
|
49 #if !ARCH_X86_64
|
yading@10
|
50 HPEL_FILTER(8, mmx)
|
yading@10
|
51 #endif
|
yading@10
|
52 HPEL_FILTER(16, sse2)
|
yading@10
|
53
|
yading@10
|
54 #define PIXFUNC(PFX, IDX, EXT) \
|
yading@10
|
55 /*MMXDISABLEDc->PFX ## _dirac_pixels_tab[0][IDX] = ff_ ## PFX ## _dirac_pixels8_ ## EXT;*/ \
|
yading@10
|
56 c->PFX ## _dirac_pixels_tab[1][IDX] = ff_ ## PFX ## _dirac_pixels16_ ## EXT; \
|
yading@10
|
57 c->PFX ## _dirac_pixels_tab[2][IDX] = ff_ ## PFX ## _dirac_pixels32_ ## EXT
|
yading@10
|
58
|
yading@10
|
59 void ff_diracdsp_init_mmx(DiracDSPContext* c)
|
yading@10
|
60 {
|
yading@10
|
61 int mm_flags = av_get_cpu_flags();
|
yading@10
|
62
|
yading@10
|
63 if (!(mm_flags & AV_CPU_FLAG_MMX))
|
yading@10
|
64 return;
|
yading@10
|
65
|
yading@10
|
66 #if HAVE_YASM
|
yading@10
|
67 c->add_dirac_obmc[0] = ff_add_dirac_obmc8_mmx;
|
yading@10
|
68 #if !ARCH_X86_64
|
yading@10
|
69 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_mmx;
|
yading@10
|
70 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
|
yading@10
|
71 c->dirac_hpel_filter = dirac_hpel_filter_mmx;
|
yading@10
|
72 c->add_rect_clamped = ff_add_rect_clamped_mmx;
|
yading@10
|
73 c->put_signed_rect_clamped = ff_put_signed_rect_clamped_mmx;
|
yading@10
|
74 #endif
|
yading@10
|
75 #endif
|
yading@10
|
76
|
yading@10
|
77 #if HAVE_MMX_INLINE
|
yading@10
|
78 PIXFUNC(put, 0, mmx);
|
yading@10
|
79 PIXFUNC(avg, 0, mmx);
|
yading@10
|
80 #endif
|
yading@10
|
81
|
yading@10
|
82 #if HAVE_MMXEXT_INLINE
|
yading@10
|
83 if (mm_flags & AV_CPU_FLAG_MMX2) {
|
yading@10
|
84 PIXFUNC(avg, 0, mmxext);
|
yading@10
|
85 }
|
yading@10
|
86 #endif
|
yading@10
|
87
|
yading@10
|
88 if (mm_flags & AV_CPU_FLAG_SSE2) {
|
yading@10
|
89 #if HAVE_YASM
|
yading@10
|
90 c->dirac_hpel_filter = dirac_hpel_filter_sse2;
|
yading@10
|
91 c->add_rect_clamped = ff_add_rect_clamped_sse2;
|
yading@10
|
92 c->put_signed_rect_clamped = ff_put_signed_rect_clamped_sse2;
|
yading@10
|
93
|
yading@10
|
94 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
|
yading@10
|
95 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;
|
yading@10
|
96 #endif
|
yading@10
|
97 #if HAVE_SSE2_INLINE
|
yading@10
|
98 c->put_dirac_pixels_tab[1][0] = ff_put_dirac_pixels16_sse2;
|
yading@10
|
99 c->avg_dirac_pixels_tab[1][0] = ff_avg_dirac_pixels16_sse2;
|
yading@10
|
100 c->put_dirac_pixels_tab[2][0] = ff_put_dirac_pixels32_sse2;
|
yading@10
|
101 c->avg_dirac_pixels_tab[2][0] = ff_avg_dirac_pixels32_sse2;
|
yading@10
|
102 #endif
|
yading@10
|
103 }
|
yading@10
|
104 }
|