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 "libavutil/x86/cpu.h"
22 #include "libavcodec/diracdsp.h"
23 #include "fpel.h"
24
25 void ff_add_rect_clamped_mmx(uint8_t *, const uint16_t *, int, const int16_t *, int, int, int);
26 void ff_add_rect_clamped_sse2(uint8_t *, const uint16_t *, int, const int16_t *, int, int, int);
27
28 void ff_add_dirac_obmc8_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
29 void ff_add_dirac_obmc16_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
30 void ff_add_dirac_obmc32_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
31
32 void ff_add_dirac_obmc16_sse2(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
33 void ff_add_dirac_obmc32_sse2(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
34
35 void ff_put_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
36 void ff_put_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
37 void ff_put_signed_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
38 void ff_put_signed_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
39 void ff_put_signed_rect_clamped_10_sse4(uint8_t *dst, int dst_stride, const uint8_t *src, int src_stride, int width, int height);
40
41 void ff_dequant_subband_32_sse4(uint8_t *src, uint8_t *dst, ptrdiff_t stride, const int qf, const int qs, int tot_v, int tot_h);
42
43 #if HAVE_X86ASM
44
45 #define HPEL_FILTER(MMSIZE, EXT) \
46 void ff_dirac_hpel_filter_v_ ## EXT(uint8_t *, const uint8_t *, int, int); \
47 void ff_dirac_hpel_filter_h_ ## EXT(uint8_t *, const uint8_t *, int); \
48 \
49 static void dirac_hpel_filter_ ## EXT(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, \
50 const uint8_t *src, int stride, int width, int height) \
51 { \
52 while( height-- ) \
53 { \
54 ff_dirac_hpel_filter_v_ ## EXT(dstv-MMSIZE, src-MMSIZE, stride, width+MMSIZE+5); \
55 ff_dirac_hpel_filter_h_ ## EXT(dsth, src, width); \
56 ff_dirac_hpel_filter_h_ ## EXT(dstc, dstv, width); \
57 \
58 dsth += stride; \
59 dstv += stride; \
60 dstc += stride; \
61 src += stride; \
62 } \
63 }
64
65 #define PIXFUNC(PFX, IDX, EXT) \
66 /*MMXDISABLEDc->PFX ## _dirac_pixels_tab[0][IDX] = PFX ## _dirac_pixels8_ ## EXT;*/ \
67 c->PFX ## _dirac_pixels_tab[1][IDX] = PFX ## _dirac_pixels16_ ## EXT; \
68 c->PFX ## _dirac_pixels_tab[2][IDX] = PFX ## _dirac_pixels32_ ## EXT
69
70 #define DIRAC_PIXOP(OPNAME, EXT)\
71 static void OPNAME ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], \
72 int stride, int h) \
73 {\
74 if (h&3)\
75 ff_ ## OPNAME ## _dirac_pixels16_c(dst, src, stride, h);\
76 else\
77 ff_ ## OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
78 }\
79 static void OPNAME ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], \
80 int stride, int h) \
81 {\
82 if (h&3) {\
83 ff_ ## OPNAME ## _dirac_pixels32_c(dst, src, stride, h);\
84 } else {\
85 ff_ ## OPNAME ## _pixels16_ ## EXT(dst , src[0] , stride, h);\
86 ff_ ## OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
87 }\
88 }
89
DIRAC_PIXOP(put,mmx)90 DIRAC_PIXOP(put, mmx)
91 DIRAC_PIXOP(avg, mmx)
92 DIRAC_PIXOP(avg, mmxext)
93
94 DIRAC_PIXOP(put, sse2)
95 DIRAC_PIXOP(avg, sse2)
96
97 #if !ARCH_X86_64
98 HPEL_FILTER(8, mmx)
99 #endif
100 HPEL_FILTER(16, sse2)
101
102 #endif // HAVE_X86ASM
103
104 void ff_diracdsp_init_x86(DiracDSPContext* c)
105 {
106 #if HAVE_X86ASM
107 int mm_flags = av_get_cpu_flags();
108
109 if (EXTERNAL_MMX(mm_flags)) {
110 c->add_dirac_obmc[0] = ff_add_dirac_obmc8_mmx;
111 #if !ARCH_X86_64
112 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_mmx;
113 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
114 c->dirac_hpel_filter = dirac_hpel_filter_mmx;
115 c->add_rect_clamped = ff_add_rect_clamped_mmx;
116 c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_mmx;
117 #endif
118 PIXFUNC(put, 0, mmx);
119 PIXFUNC(avg, 0, mmx);
120 }
121
122 if (EXTERNAL_MMXEXT(mm_flags)) {
123 PIXFUNC(avg, 0, mmxext);
124 }
125
126 if (EXTERNAL_SSE2(mm_flags)) {
127 c->dirac_hpel_filter = dirac_hpel_filter_sse2;
128 c->add_rect_clamped = ff_add_rect_clamped_sse2;
129 c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_sse2;
130
131 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
132 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;
133
134 c->put_dirac_pixels_tab[1][0] = put_dirac_pixels16_sse2;
135 c->avg_dirac_pixels_tab[1][0] = avg_dirac_pixels16_sse2;
136 c->put_dirac_pixels_tab[2][0] = put_dirac_pixels32_sse2;
137 c->avg_dirac_pixels_tab[2][0] = avg_dirac_pixels32_sse2;
138 }
139
140 if (EXTERNAL_SSE4(mm_flags)) {
141 c->dequant_subband[1] = ff_dequant_subband_32_sse4;
142 c->put_signed_rect_clamped[1] = ff_put_signed_rect_clamped_10_sse4;
143 }
144 #endif
145 }
146