?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Clarify live555 version error
[vlc/asuraparaju-public.git] / modules / video_filter / yadif.h
blob2d2a56f071587ff103c5908d043c68981e0ce3f0
1 /*
2 * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of MPlayer.
5 *
6 * MPlayer 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 * MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 /* */
22 #if defined(CAN_COMPILE_SSE2) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ > 0))
23
24 #define HAVE_YADIF_SSE2
25
26 #define LOAD4(mem,dst) \
27 "movd "mem", "#dst" \n\t"\
28 "punpcklbw %%mm7, "#dst" \n\t"
29
30 #define PABS(tmp,dst) \
31 "pxor "#tmp", "#tmp" \n\t"\
32 "psubw "#dst", "#tmp" \n\t"\
33 "pmaxsw "#tmp", "#dst" \n\t"
34
35 #define CHECK(pj,mj) \
36 "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
37 "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
38 "movq %%mm2, %%mm4 \n\t"\
39 "movq %%mm2, %%mm5 \n\t"\
40 "pxor %%mm3, %%mm4 \n\t"\
41 "pavgb %%mm3, %%mm5 \n\t"\
42 "pand %[pb1], %%mm4 \n\t"\
43 "psubusb %%mm4, %%mm5 \n\t"\
44 "psrlq $8, %%mm5 \n\t"\
45 "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
46 "movq %%mm2, %%mm4 \n\t"\
47 "psubusb %%mm3, %%mm2 \n\t"\
48 "psubusb %%mm4, %%mm3 \n\t"\
49 "pmaxub %%mm3, %%mm2 \n\t"\
50 "movq %%mm2, %%mm3 \n\t"\
51 "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
52 "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
53 "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
54 "punpcklbw %%mm7, %%mm2 \n\t"\
55 "punpcklbw %%mm7, %%mm3 \n\t"\
56 "punpcklbw %%mm7, %%mm4 \n\t"\
57 "paddw %%mm3, %%mm2 \n\t"\
58 "paddw %%mm4, %%mm2 \n\t" /* score */
59
60 #define CHECK1 \
61 "movq %%mm0, %%mm3 \n\t"\
62 "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
63 "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
64 "movq %%mm3, %%mm6 \n\t"\
65 "pand %%mm3, %%mm5 \n\t"\
66 "pandn %%mm1, %%mm3 \n\t"\
67 "por %%mm5, %%mm3 \n\t"\
68 "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
69
70 #define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
71 hurts both quality and speed, but matches the C version. */\
72 "paddw %[pw1], %%mm6 \n\t"\
73 "psllw $14, %%mm6 \n\t"\
74 "paddsw %%mm6, %%mm2 \n\t"\
75 "movq %%mm0, %%mm3 \n\t"\
76 "pcmpgtw %%mm2, %%mm3 \n\t"\
77 "pminsw %%mm2, %%mm0 \n\t"\
78 "pand %%mm3, %%mm5 \n\t"\
79 "pandn %%mm1, %%mm3 \n\t"\
80 "por %%mm5, %%mm3 \n\t"\
81 "movq %%mm3, %%mm1 \n\t"
82
83 static void yadif_filter_line_mmx2(struct vf_priv_s *p, uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int refs, int parity){
84 static const uint64_t pw_1 = 0x0001000100010001ULL;
85 static const uint64_t pb_1 = 0x0101010101010101ULL;
86 const int mode = p->mode;
87 uint64_t tmp0, tmp1, tmp2, tmp3;
88 int x;
89
90 #define FILTER\
91 for(x=0; x<w; x+=4){\
92 __asm__ volatile(\
93 "pxor %%mm7, %%mm7 \n\t"\
94 LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
95 LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
96 LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
97 LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
98 "movq %%mm3, %%mm4 \n\t"\
99 "paddw %%mm2, %%mm3 \n\t"\
100 "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
101 "movq %%mm0, %[tmp0] \n\t" /* c */\
102 "movq %%mm3, %[tmp1] \n\t" /* d */\
103 "movq %%mm1, %[tmp2] \n\t" /* e */\
104 "psubw %%mm4, %%mm2 \n\t"\
105 PABS( %%mm4, %%mm2) /* temporal_diff0 */\
106 LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
107 LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
108 "psubw %%mm0, %%mm3 \n\t"\
109 "psubw %%mm1, %%mm4 \n\t"\
110 PABS( %%mm5, %%mm3)\
111 PABS( %%mm5, %%mm4)\
112 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
113 "psrlw $1, %%mm2 \n\t"\
114 "psrlw $1, %%mm3 \n\t"\
115 "pmaxsw %%mm3, %%mm2 \n\t"\
116 LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
117 LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
118 "psubw %%mm0, %%mm3 \n\t"\
119 "psubw %%mm1, %%mm4 \n\t"\
120 PABS( %%mm5, %%mm3)\
121 PABS( %%mm5, %%mm4)\
122 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
123 "psrlw $1, %%mm3 \n\t"\
124 "pmaxsw %%mm3, %%mm2 \n\t"\
125 "movq %%mm2, %[tmp3] \n\t" /* diff */\
126 \
127 "paddw %%mm0, %%mm1 \n\t"\
128 "paddw %%mm0, %%mm0 \n\t"\
129 "psubw %%mm1, %%mm0 \n\t"\
130 "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
131 PABS( %%mm2, %%mm0) /* ABS(c-e) */\
132 \
133 "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
134 "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
135 "movq %%mm2, %%mm4 \n\t"\
136 "psubusb %%mm3, %%mm2 \n\t"\
137 "psubusb %%mm4, %%mm3 \n\t"\
138 "pmaxub %%mm3, %%mm2 \n\t"\
139 "pshufw $9,%%mm2, %%mm3 \n\t"\
140 "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
141 "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
142 "paddw %%mm2, %%mm0 \n\t"\
143 "paddw %%mm3, %%mm0 \n\t"\
144 "psubw %[pw1], %%mm0 \n\t" /* spatial_score */\
145 \
146 CHECK(-2,0)\
147 CHECK1\
148 CHECK(-3,1)\
149 CHECK2\
150 CHECK(0,-2)\
151 CHECK1\
152 CHECK(1,-3)\
153 CHECK2\
154 \
155 /* if(p->mode<2) ... */\
156 "movq %[tmp3], %%mm6 \n\t" /* diff */\
157 "cmp $2, %[mode] \n\t"\
158 "jge 1f \n\t"\
159 LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
160 LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
161 LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
162 LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
163 "paddw %%mm4, %%mm2 \n\t"\
164 "paddw %%mm5, %%mm3 \n\t"\
165 "psrlw $1, %%mm2 \n\t" /* b */\
166 "psrlw $1, %%mm3 \n\t" /* f */\
167 "movq %[tmp0], %%mm4 \n\t" /* c */\
168 "movq %[tmp1], %%mm5 \n\t" /* d */\
169 "movq %[tmp2], %%mm7 \n\t" /* e */\
170 "psubw %%mm4, %%mm2 \n\t" /* b-c */\
171 "psubw %%mm7, %%mm3 \n\t" /* f-e */\
172 "movq %%mm5, %%mm0 \n\t"\
173 "psubw %%mm4, %%mm5 \n\t" /* d-c */\
174 "psubw %%mm7, %%mm0 \n\t" /* d-e */\
175 "movq %%mm2, %%mm4 \n\t"\
176 "pminsw %%mm3, %%mm2 \n\t"\
177 "pmaxsw %%mm4, %%mm3 \n\t"\
178 "pmaxsw %%mm5, %%mm2 \n\t"\
179 "pminsw %%mm5, %%mm3 \n\t"\
180 "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
181 "pminsw %%mm0, %%mm3 \n\t" /* min */\
182 "pxor %%mm4, %%mm4 \n\t"\
183 "pmaxsw %%mm3, %%mm6 \n\t"\
184 "psubw %%mm2, %%mm4 \n\t" /* -max */\
185 "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
186 "1: \n\t"\
187 \
188 "movq %[tmp1], %%mm2 \n\t" /* d */\
189 "movq %%mm2, %%mm3 \n\t"\
190 "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
191 "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
192 "pmaxsw %%mm2, %%mm1 \n\t"\
193 "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
194 "packuswb %%mm1, %%mm1 \n\t"\
195 \
196 :[tmp0]"=m"(tmp0),\
197 [tmp1]"=m"(tmp1),\
198 [tmp2]"=m"(tmp2),\
199 [tmp3]"=m"(tmp3)\
200 :[prev] "r"(prev),\
201 [cur] "r"(cur),\
202 [next] "r"(next),\
203 [prefs]"r"((x86_reg)refs),\
204 [mrefs]"r"((x86_reg)-refs),\
205 [pw1] "m"(pw_1),\
206 [pb1] "m"(pb_1),\
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
236
237 #endif
238
239 static void yadif_filter_line_c(struct vf_priv_s *p, uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int refs, int parity){
240 int x;
241 uint8_t *prev2= parity ? prev : cur ;
242 uint8_t *next2= parity ? cur : next;
243 for(x=0; x<w; x++){
244 int c= cur[-refs];
245 int d= (prev2[0] + next2[0])>>1;
246 int e= cur[+refs];
247 int temporal_diff0= FFABS(prev2[0] - next2[0]);
248 int temporal_diff1=( FFABS(prev[-refs] - c) + FFABS(prev[+refs] - e) )>>1;
249 int temporal_diff2=( FFABS(next[-refs] - c) + FFABS(next[+refs] - e) )>>1;
250 int diff= FFMAX3(temporal_diff0>>1, temporal_diff1, temporal_diff2);
251 int spatial_pred= (c+e)>>1;
252 int spatial_score= FFABS(cur[-refs-1] - cur[+refs-1]) + FFABS(c-e)
253 + FFABS(cur[-refs+1] - cur[+refs+1]) - 1;
254
255 #define CHECK(j)\
256 { int score= FFABS(cur[-refs-1+j] - cur[+refs-1-j])\
257 + FFABS(cur[-refs +j] - cur[+refs -j])\
258 + FFABS(cur[-refs+1+j] - cur[+refs+1-j]);\
259 if(score < spatial_score){\
260 spatial_score= score;\
261 spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
262
263 CHECK(-1) CHECK(-2) }} }}
264 CHECK( 1) CHECK( 2) }} }}
265
266 if(p->mode<2){
267 int b= (prev2[-2*refs] + next2[-2*refs])>>1;
268 int f= (prev2[+2*refs] + next2[+2*refs])>>1;
269 #if 0
270 int a= cur[-3*refs];
271 int g= cur[+3*refs];
272 int max= FFMAX3(d-e, d-c, FFMIN3(FFMAX(b-c,f-e),FFMAX(b-c,b-a),FFMAX(f-g,f-e)) );
273 int min= FFMIN3(d-e, d-c, FFMAX3(FFMIN(b-c,f-e),FFMIN(b-c,b-a),FFMIN(f-g,f-e)) );
274 #else
275 int max= FFMAX3(d-e, d-c, FFMIN(b-c, f-e));
276 int min= FFMIN3(d-e, d-c, FFMAX(b-c, f-e));
277 #endif
278
279 diff= FFMAX3(diff, min, -max);
280 }
281
282 if(spatial_pred > d + diff)
283 spatial_pred = d + diff;
284 else if(spatial_pred < d - diff)
285 spatial_pred = d - diff;
286
287 dst[0] = spatial_pred;
288
289 dst++;
290 cur++;
291 prev++;
292 next++;
293 prev2++;
294 next2++;
295 }
296 }
297