Statistics
| Branch: | Revision:

ffmpeg / libavfilter / x86 / yadif.c @ d370e3e9

History | View | Annotate | Download (9.33 KB)

1
/*
2
 * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or modify
7
 * it under the terms of the GNU General Public License as published by
8
 * the Free Software Foundation; either version 2 of the License, or
9
 * (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
14
 * GNU General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU General Public License along
17
 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
18
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19
 */
20

    
21
#include "libavutil/cpu.h"
22
#include "libavutil/x86_cpu.h"
23
#include "libavfilter/yadif.h"
24

    
25
#define LOAD4(mem,dst) \
26
            "movd      "mem", "#dst" \n\t"\
27
            "punpcklbw %%mm7, "#dst" \n\t"
28

    
29
#define PABS(tmp,dst) \
30
            "pxor     "#tmp", "#tmp" \n\t"\
31
            "psubw    "#dst", "#tmp" \n\t"\
32
            "pmaxsw   "#tmp", "#dst" \n\t"
33

    
34
#define CHECK(pj,mj) \
35
            "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
36
            "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
37
            "movq      %%mm2, %%mm4 \n\t"\
38
            "movq      %%mm2, %%mm5 \n\t"\
39
            "pxor      %%mm3, %%mm4 \n\t"\
40
            "pavgb     %%mm3, %%mm5 \n\t"\
41
            "pand     "MANGLE(pb_1)", %%mm4 \n\t"\
42
            "psubusb   %%mm4, %%mm5 \n\t"\
43
            "psrlq     $8,    %%mm5 \n\t"\
44
            "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
45
            "movq      %%mm2, %%mm4 \n\t"\
46
            "psubusb   %%mm3, %%mm2 \n\t"\
47
            "psubusb   %%mm4, %%mm3 \n\t"\
48
            "pmaxub    %%mm3, %%mm2 \n\t"\
49
            "movq      %%mm2, %%mm3 \n\t"\
50
            "movq      %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
51
            "psrlq      $8,   %%mm3 \n\t" /* ABS(cur[x-refs  +j] - cur[x+refs  -j]) */\
52
            "psrlq     $16,   %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
53
            "punpcklbw %%mm7, %%mm2 \n\t"\
54
            "punpcklbw %%mm7, %%mm3 \n\t"\
55
            "punpcklbw %%mm7, %%mm4 \n\t"\
56
            "paddw     %%mm3, %%mm2 \n\t"\
57
            "paddw     %%mm4, %%mm2 \n\t" /* score */
58

    
59
#define CHECK1 \
60
            "movq      %%mm0, %%mm3 \n\t"\
61
            "pcmpgtw   %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
62
            "pminsw    %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
63
            "movq      %%mm3, %%mm6 \n\t"\
64
            "pand      %%mm3, %%mm5 \n\t"\
65
            "pandn     %%mm1, %%mm3 \n\t"\
66
            "por       %%mm5, %%mm3 \n\t"\
67
            "movq      %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
68

    
69
#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
70
                  hurts both quality and speed, but matches the C version. */\
71
            "paddw    "MANGLE(pw_1)", %%mm6 \n\t"\
72
            "psllw     $14,   %%mm6 \n\t"\
73
            "paddsw    %%mm6, %%mm2 \n\t"\
74
            "movq      %%mm0, %%mm3 \n\t"\
75
            "pcmpgtw   %%mm2, %%mm3 \n\t"\
76
            "pminsw    %%mm2, %%mm0 \n\t"\
77
            "pand      %%mm3, %%mm5 \n\t"\
78
            "pandn     %%mm1, %%mm3 \n\t"\
79
            "por       %%mm5, %%mm3 \n\t"\
80
            "movq      %%mm3, %%mm1 \n\t"
81

    
82
DECLARE_ASM_CONST(16, uint64_t, pw_1) = 0x0001000100010001ULL;
83
DECLARE_ASM_CONST(16, uint64_t, pb_1) = 0x0101010101010101ULL;
84

    
85
void ff_yadif_filter_line_mmx(uint8_t *dst,
86
                              uint8_t *prev, uint8_t *cur, uint8_t *next,
87
                              int w, int refs, int parity, int mode)
88
{
89
    uint64_t tmp0, tmp1, tmp2, tmp3;
90
    int x;
91

    
92
#define FILTER\
93
    for(x=0; x<w; x+=4){\
94
        __asm__ volatile(\
95
            "pxor      %%mm7, %%mm7 \n\t"\
96
            LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
97
            LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
98
            LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
99
            LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
100
            "movq      %%mm3, %%mm4 \n\t"\
101
            "paddw     %%mm2, %%mm3 \n\t"\
102
            "psraw     $1,    %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
103
            "movq      %%mm0, %[tmp0] \n\t" /* c */\
104
            "movq      %%mm3, %[tmp1] \n\t" /* d */\
105
            "movq      %%mm1, %[tmp2] \n\t" /* e */\
106
            "psubw     %%mm4, %%mm2 \n\t"\
107
            PABS(      %%mm4, %%mm2) /* temporal_diff0 */\
108
            LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
109
            LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
110
            "psubw     %%mm0, %%mm3 \n\t"\
111
            "psubw     %%mm1, %%mm4 \n\t"\
112
            PABS(      %%mm5, %%mm3)\
113
            PABS(      %%mm5, %%mm4)\
114
            "paddw     %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
115
            "psrlw     $1,    %%mm2 \n\t"\
116
            "psrlw     $1,    %%mm3 \n\t"\
117
            "pmaxsw    %%mm3, %%mm2 \n\t"\
118
            LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
119
            LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
120
            "psubw     %%mm0, %%mm3 \n\t"\
121
            "psubw     %%mm1, %%mm4 \n\t"\
122
            PABS(      %%mm5, %%mm3)\
123
            PABS(      %%mm5, %%mm4)\
124
            "paddw     %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
125
            "psrlw     $1,    %%mm3 \n\t"\
126
            "pmaxsw    %%mm3, %%mm2 \n\t"\
127
            "movq      %%mm2, %[tmp3] \n\t" /* diff */\
128
\
129
            "paddw     %%mm0, %%mm1 \n\t"\
130
            "paddw     %%mm0, %%mm0 \n\t"\
131
            "psubw     %%mm1, %%mm0 \n\t"\
132
            "psrlw     $1,    %%mm1 \n\t" /* spatial_pred */\
133
            PABS(      %%mm2, %%mm0)      /* ABS(c-e) */\
134
\
135
            "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
136
            "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
137
            "movq      %%mm2, %%mm4 \n\t"\
138
            "psubusb   %%mm3, %%mm2 \n\t"\
139
            "psubusb   %%mm4, %%mm3 \n\t"\
140
            "pmaxub    %%mm3, %%mm2 \n\t"\
141
            "pshufw $9,%%mm2, %%mm3 \n\t"\
142
            "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
143
            "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
144
            "paddw     %%mm2, %%mm0 \n\t"\
145
            "paddw     %%mm3, %%mm0 \n\t"\
146
            "psubw    "MANGLE(pw_1)", %%mm0 \n\t" /* spatial_score */\
147
\
148
            CHECK(-2,0)\
149
            CHECK1\
150
            CHECK(-3,1)\
151
            CHECK2\
152
            CHECK(0,-2)\
153
            CHECK1\
154
            CHECK(1,-3)\
155
            CHECK2\
156
\
157
            /* if(p->mode<2) ... */\
158
            "movq    %[tmp3], %%mm6 \n\t" /* diff */\
159
            "cmpl      $2, %[mode] \n\t"\
160
            "jge       1f \n\t"\
161
            LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
162
            LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
163
            LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
164
            LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
165
            "paddw     %%mm4, %%mm2 \n\t"\
166
            "paddw     %%mm5, %%mm3 \n\t"\
167
            "psrlw     $1,    %%mm2 \n\t" /* b */\
168
            "psrlw     $1,    %%mm3 \n\t" /* f */\
169
            "movq    %[tmp0], %%mm4 \n\t" /* c */\
170
            "movq    %[tmp1], %%mm5 \n\t" /* d */\
171
            "movq    %[tmp2], %%mm7 \n\t" /* e */\
172
            "psubw     %%mm4, %%mm2 \n\t" /* b-c */\
173
            "psubw     %%mm7, %%mm3 \n\t" /* f-e */\
174
            "movq      %%mm5, %%mm0 \n\t"\
175
            "psubw     %%mm4, %%mm5 \n\t" /* d-c */\
176
            "psubw     %%mm7, %%mm0 \n\t" /* d-e */\
177
            "movq      %%mm2, %%mm4 \n\t"\
178
            "pminsw    %%mm3, %%mm2 \n\t"\
179
            "pmaxsw    %%mm4, %%mm3 \n\t"\
180
            "pmaxsw    %%mm5, %%mm2 \n\t"\
181
            "pminsw    %%mm5, %%mm3 \n\t"\
182
            "pmaxsw    %%mm0, %%mm2 \n\t" /* max */\
183
            "pminsw    %%mm0, %%mm3 \n\t" /* min */\
184
            "pxor      %%mm4, %%mm4 \n\t"\
185
            "pmaxsw    %%mm3, %%mm6 \n\t"\
186
            "psubw     %%mm2, %%mm4 \n\t" /* -max */\
187
            "pmaxsw    %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
188
            "1: \n\t"\
189
\
190
            "movq    %[tmp1], %%mm2 \n\t" /* d */\
191
            "movq      %%mm2, %%mm3 \n\t"\
192
            "psubw     %%mm6, %%mm2 \n\t" /* d-diff */\
193
            "paddw     %%mm6, %%mm3 \n\t" /* d+diff */\
194
            "pmaxsw    %%mm2, %%mm1 \n\t"\
195
            "pminsw    %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
196
            "packuswb  %%mm1, %%mm1 \n\t"\
197
\
198
            :[tmp0]"=m"(tmp0),\
199
             [tmp1]"=m"(tmp1),\
200
             [tmp2]"=m"(tmp2),\
201
             [tmp3]"=m"(tmp3)\
202
            :[prev] "r"(prev),\
203
             [cur]  "r"(cur),\
204
             [next] "r"(next),\
205
             [prefs]"r"((x86_reg)refs),\
206
             [mrefs]"r"((x86_reg)-refs),\
207
             [mode] "g"(mode)\
208
        );\
209
        __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
210
        dst += 4;\
211
        prev+= 4;\
212
        cur += 4;\
213
        next+= 4;\
214
    }
215

    
216
    if (parity) {
217
#define prev2 "prev"
218
#define next2 "cur"
219
        FILTER
220
#undef prev2
221
#undef next2
222
    } else {
223
#define prev2 "cur"
224
#define next2 "next"
225
        FILTER
226
#undef prev2
227
#undef next2
228
    }
229
}
230
#undef LOAD4
231
#undef PABS
232
#undef CHECK
233
#undef CHECK1
234
#undef CHECK2
235
#undef FILTER