2 * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of MPlayer.
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.
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.
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.
28 #include "cpudetect.h"
32 #include "img_format.h"
35 #include "libvo/fastmemcpy.h"
36 #include "libavutil/common.h"
38 //===========================================================================//
46 double buffered_pts_delta
;
47 mp_image_t
*buffered_mpi
;
53 static void (*filter_line
)(struct vf_priv_s
*p
, uint8_t *dst
, uint8_t *prev
, uint8_t *cur
, uint8_t *next
, int w
, int refs
, int parity
);
55 static void store_ref(struct vf_priv_s
*p
, uint8_t *src
[3], int src_stride
[3], int width
, int height
){
58 memcpy (p
->ref
[3], p
->ref
[0], sizeof(uint8_t *)*3);
59 memmove(p
->ref
[0], p
->ref
[1], sizeof(uint8_t *)*3*3);
64 memcpy_pic(p
->ref
[2][i
], src
[i
], width
>>is_chroma
, height
>>is_chroma
, p
->stride
[i
], src_stride
[i
]);
70 #define LOAD4(mem,dst) \
71 "movd "mem", "#dst" \n\t"\
72 "punpcklbw %%mm7, "#dst" \n\t"
74 #define PABS(tmp,dst) \
75 "pxor "#tmp", "#tmp" \n\t"\
76 "psubw "#dst", "#tmp" \n\t"\
77 "pmaxsw "#tmp", "#dst" \n\t"
79 #define CHECK(pj,mj) \
80 "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
81 "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
82 "movq %%mm2, %%mm4 \n\t"\
83 "movq %%mm2, %%mm5 \n\t"\
84 "pxor %%mm3, %%mm4 \n\t"\
85 "pavgb %%mm3, %%mm5 \n\t"\
86 "pand %[pb1], %%mm4 \n\t"\
87 "psubusb %%mm4, %%mm5 \n\t"\
88 "psrlq $8, %%mm5 \n\t"\
89 "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
90 "movq %%mm2, %%mm4 \n\t"\
91 "psubusb %%mm3, %%mm2 \n\t"\
92 "psubusb %%mm4, %%mm3 \n\t"\
93 "pmaxub %%mm3, %%mm2 \n\t"\
94 "movq %%mm2, %%mm3 \n\t"\
95 "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
96 "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
97 "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
98 "punpcklbw %%mm7, %%mm2 \n\t"\
99 "punpcklbw %%mm7, %%mm3 \n\t"\
100 "punpcklbw %%mm7, %%mm4 \n\t"\
101 "paddw %%mm3, %%mm2 \n\t"\
102 "paddw %%mm4, %%mm2 \n\t" /* score */
105 "movq %%mm0, %%mm3 \n\t"\
106 "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
107 "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
108 "movq %%mm3, %%mm6 \n\t"\
109 "pand %%mm3, %%mm5 \n\t"\
110 "pandn %%mm1, %%mm3 \n\t"\
111 "por %%mm5, %%mm3 \n\t"\
112 "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
114 #define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
115 hurts both quality and speed, but matches the C version. */\
116 "paddw %[pw1], %%mm6 \n\t"\
117 "psllw $14, %%mm6 \n\t"\
118 "paddsw %%mm6, %%mm2 \n\t"\
119 "movq %%mm0, %%mm3 \n\t"\
120 "pcmpgtw %%mm2, %%mm3 \n\t"\
121 "pminsw %%mm2, %%mm0 \n\t"\
122 "pand %%mm3, %%mm5 \n\t"\
123 "pandn %%mm1, %%mm3 \n\t"\
124 "por %%mm5, %%mm3 \n\t"\
125 "movq %%mm3, %%mm1 \n\t"
127 static void 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
){
128 static const uint64_t pw_1
= 0x0001000100010001ULL
;
129 static const uint64_t pb_1
= 0x0101010101010101ULL
;
130 const int mode
= p
->mode
;
131 uint64_t tmp0
, tmp1
, tmp2
, tmp3
;
135 for(x=0; x<w; x+=4){\
137 "pxor %%mm7, %%mm7 \n\t"\
138 LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
139 LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
140 LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
141 LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
142 "movq %%mm3, %%mm4 \n\t"\
143 "paddw %%mm2, %%mm3 \n\t"\
144 "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
145 "movq %%mm0, %[tmp0] \n\t" /* c */\
146 "movq %%mm3, %[tmp1] \n\t" /* d */\
147 "movq %%mm1, %[tmp2] \n\t" /* e */\
148 "psubw %%mm4, %%mm2 \n\t"\
149 PABS( %%mm4, %%mm2) /* temporal_diff0 */\
150 LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
151 LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
152 "psubw %%mm0, %%mm3 \n\t"\
153 "psubw %%mm1, %%mm4 \n\t"\
156 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
157 "psrlw $1, %%mm2 \n\t"\
158 "psrlw $1, %%mm3 \n\t"\
159 "pmaxsw %%mm3, %%mm2 \n\t"\
160 LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
161 LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
162 "psubw %%mm0, %%mm3 \n\t"\
163 "psubw %%mm1, %%mm4 \n\t"\
166 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
167 "psrlw $1, %%mm3 \n\t"\
168 "pmaxsw %%mm3, %%mm2 \n\t"\
169 "movq %%mm2, %[tmp3] \n\t" /* diff */\
171 "paddw %%mm0, %%mm1 \n\t"\
172 "paddw %%mm0, %%mm0 \n\t"\
173 "psubw %%mm1, %%mm0 \n\t"\
174 "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
175 PABS( %%mm2, %%mm0) /* ABS(c-e) */\
177 "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
178 "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
179 "movq %%mm2, %%mm4 \n\t"\
180 "psubusb %%mm3, %%mm2 \n\t"\
181 "psubusb %%mm4, %%mm3 \n\t"\
182 "pmaxub %%mm3, %%mm2 \n\t"\
183 "pshufw $9,%%mm2, %%mm3 \n\t"\
184 "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
185 "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
186 "paddw %%mm2, %%mm0 \n\t"\
187 "paddw %%mm3, %%mm0 \n\t"\
188 "psubw %[pw1], %%mm0 \n\t" /* spatial_score */\
199 /* if(p->mode<2) ... */\
200 "movq %[tmp3], %%mm6 \n\t" /* diff */\
201 "cmpl $2, %[mode] \n\t"\
203 LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
204 LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
205 LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
206 LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
207 "paddw %%mm4, %%mm2 \n\t"\
208 "paddw %%mm5, %%mm3 \n\t"\
209 "psrlw $1, %%mm2 \n\t" /* b */\
210 "psrlw $1, %%mm3 \n\t" /* f */\
211 "movq %[tmp0], %%mm4 \n\t" /* c */\
212 "movq %[tmp1], %%mm5 \n\t" /* d */\
213 "movq %[tmp2], %%mm7 \n\t" /* e */\
214 "psubw %%mm4, %%mm2 \n\t" /* b-c */\
215 "psubw %%mm7, %%mm3 \n\t" /* f-e */\
216 "movq %%mm5, %%mm0 \n\t"\
217 "psubw %%mm4, %%mm5 \n\t" /* d-c */\
218 "psubw %%mm7, %%mm0 \n\t" /* d-e */\
219 "movq %%mm2, %%mm4 \n\t"\
220 "pminsw %%mm3, %%mm2 \n\t"\
221 "pmaxsw %%mm4, %%mm3 \n\t"\
222 "pmaxsw %%mm5, %%mm2 \n\t"\
223 "pminsw %%mm5, %%mm3 \n\t"\
224 "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
225 "pminsw %%mm0, %%mm3 \n\t" /* min */\
226 "pxor %%mm4, %%mm4 \n\t"\
227 "pmaxsw %%mm3, %%mm6 \n\t"\
228 "psubw %%mm2, %%mm4 \n\t" /* -max */\
229 "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
232 "movq %[tmp1], %%mm2 \n\t" /* d */\
233 "movq %%mm2, %%mm3 \n\t"\
234 "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
235 "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
236 "pmaxsw %%mm2, %%mm1 \n\t"\
237 "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
238 "packuswb %%mm1, %%mm1 \n\t"\
247 [prefs]"r"((x86_reg)refs),\
248 [mrefs]"r"((x86_reg)-refs),\
253 __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
281 #endif /* HAVE_MMX */
283 static void 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
){
285 uint8_t *prev2
= parity
? prev
: cur
;
286 uint8_t *next2
= parity
? cur
: next
;
289 int d
= (prev2
[0] + next2
[0])>>1;
291 int temporal_diff0
= FFABS(prev2
[0] - next2
[0]);
292 int temporal_diff1
=( FFABS(prev
[-refs
] - c
) + FFABS(prev
[+refs
] - e
) )>>1;
293 int temporal_diff2
=( FFABS(next
[-refs
] - c
) + FFABS(next
[+refs
] - e
) )>>1;
294 int diff
= FFMAX3(temporal_diff0
>>1, temporal_diff1
, temporal_diff2
);
295 int spatial_pred
= (c
+e
)>>1;
296 int spatial_score
= FFABS(cur
[-refs
-1] - cur
[+refs
-1]) + FFABS(c
-e
)
297 + FFABS(cur
[-refs
+1] - cur
[+refs
+1]) - 1;
300 { int score= FFABS(cur[-refs-1+j] - cur[+refs-1-j])\
301 + FFABS(cur[-refs +j] - cur[+refs -j])\
302 + FFABS(cur[-refs+1+j] - cur[+refs+1-j]);\
303 if(score < spatial_score){\
304 spatial_score= score;\
305 spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
307 CHECK(-1) CHECK(-2) }} }}
308 CHECK( 1) CHECK( 2) }} }}
311 int b
= (prev2
[-2*refs
] + next2
[-2*refs
])>>1;
312 int f
= (prev2
[+2*refs
] + next2
[+2*refs
])>>1;
316 int max
= FFMAX3(d
-e
, d
-c
, FFMIN3(FFMAX(b
-c
,f
-e
),FFMAX(b
-c
,b
-a
),FFMAX(f
-g
,f
-e
)) );
317 int min
= FFMIN3(d
-e
, d
-c
, FFMAX3(FFMIN(b
-c
,f
-e
),FFMIN(b
-c
,b
-a
),FFMIN(f
-g
,f
-e
)) );
319 int max
= FFMAX3(d
-e
, d
-c
, FFMIN(b
-c
, f
-e
));
320 int min
= FFMIN3(d
-e
, d
-c
, FFMAX(b
-c
, f
-e
));
323 diff
= FFMAX3(diff
, min
, -max
);
326 if(spatial_pred
> d
+ diff
)
327 spatial_pred
= d
+ diff
;
328 else if(spatial_pred
< d
- diff
)
329 spatial_pred
= d
- diff
;
331 dst
[0] = spatial_pred
;
342 static void filter(struct vf_priv_s
*p
, uint8_t *dst
[3], int dst_stride
[3], int width
, int height
, int parity
, int tff
){
347 int w
= width
>>is_chroma
;
348 int h
= height
>>is_chroma
;
349 int refs
= p
->stride
[i
];
352 if((y
^ parity
) & 1){
353 uint8_t *prev
= &p
->ref
[0][i
][y
*refs
];
354 uint8_t *cur
= &p
->ref
[1][i
][y
*refs
];
355 uint8_t *next
= &p
->ref
[2][i
][y
*refs
];
356 uint8_t *dst2
= &dst
[i
][y
*dst_stride
[i
]];
357 filter_line(p
, dst2
, prev
, cur
, next
, w
, refs
, parity
^ tff
);
359 fast_memcpy(&dst
[i
][y
*dst_stride
[i
]], &p
->ref
[1][i
][y
*refs
], w
);
364 if(gCpuCaps
.hasMMX2
) __asm__
volatile("emms \n\t" : : : "memory");
368 static int config(struct vf_instance
*vf
,
369 int width
, int height
, int d_width
, int d_height
,
370 unsigned int flags
, unsigned int outfmt
){
375 int w
= ((width
+ 31) & (~31))>>is_chroma
;
376 int h
= ((height
+6+ 31) & (~31))>>is_chroma
;
378 vf
->priv
->stride
[i
]= w
;
380 vf
->priv
->ref
[j
][i
]= (char *)malloc(w
*h
*sizeof(uint8_t))+3*w
;
383 return vf_next_config(vf
,width
,height
,d_width
,d_height
,flags
,outfmt
);
386 static int continue_buffered_image(struct vf_instance
*vf
);
388 static int put_image(struct vf_instance
*vf
, mp_image_t
*mpi
, double pts
){
391 if(vf
->priv
->parity
< 0) {
392 if (mpi
->fields
& MP_IMGFIELD_ORDERED
)
393 tff
= !!(mpi
->fields
& MP_IMGFIELD_TOP_FIRST
);
397 else tff
= (vf
->priv
->parity
&1)^1;
399 store_ref(vf
->priv
, mpi
->planes
, mpi
->stride
, mpi
->w
, mpi
->h
);
403 if (vf
->priv
->buffered_pts
== MP_NOPTS_VALUE
)
404 delta
= 1001.0/60000.0; // delta = field time distance
406 delta
= (pts
- vf
->priv
->buffered_pts
) / 2;
407 if (delta
<= 0.0 || delta
>= 0.5)
409 vf
->priv
->buffered_pts_delta
= delta
;
412 vf
->priv
->buffered_mpi
= mpi
;
413 vf
->priv
->buffered_tff
= tff
;
414 vf
->priv
->buffered_i
= 0;
415 vf
->priv
->buffered_pts
= pts
;
417 if(vf
->priv
->do_deinterlace
== 0)
418 return vf_next_put_image(vf
, mpi
, pts
);
419 else if(vf
->priv
->do_deinterlace
== 1){
420 vf
->priv
->do_deinterlace
= 2;
423 return continue_buffered_image(vf
);
426 static int continue_buffered_image(struct vf_instance
*vf
)
428 mp_image_t
*mpi
= vf
->priv
->buffered_mpi
;
429 int tff
= vf
->priv
->buffered_tff
;
430 double pts
= vf
->priv
->buffered_pts
;
435 pts
+= (vf
->priv
->buffered_i
- 0.5 * (vf
->priv
->mode
&1)) * vf
->priv
->buffered_pts_delta
;
437 for(i
= vf
->priv
->buffered_i
; i
<=(vf
->priv
->mode
&1); i
++){
438 dmpi
=vf_get_image(vf
->next
,mpi
->imgfmt
,
440 MP_IMGFLAG_ACCEPT_STRIDE
|MP_IMGFLAG_PREFER_ALIGNED_STRIDE
,
441 mpi
->width
,mpi
->height
);
442 vf_clone_mpi_attributes(dmpi
, mpi
);
443 filter(vf
->priv
, dmpi
->planes
, dmpi
->stride
, mpi
->w
, mpi
->h
, i
^ tff
^ 1, tff
);
444 if (i
< (vf
->priv
->mode
& 1))
445 vf_queue_frame(vf
, continue_buffered_image
);
446 ret
|= vf_next_put_image(vf
, dmpi
, pts
);
449 vf
->priv
->buffered_i
= 1;
453 static void uninit(struct vf_instance
*vf
){
455 if(!vf
->priv
) return;
457 for(i
=0; i
<3*3; i
++){
458 uint8_t **p
= &vf
->priv
->ref
[i
%3][i
/3];
459 if(*p
) free(*p
- 3*vf
->priv
->stride
[i
/3]);
466 //===========================================================================//
467 static int query_format(struct vf_instance
*vf
, unsigned int fmt
){
474 return vf_next_query_format(vf
,fmt
);
479 static int control(struct vf_instance
*vf
, int request
, void* data
){
481 case VFCTRL_GET_DEINTERLACE
:
482 *(int*)data
= vf
->priv
->do_deinterlace
;
484 case VFCTRL_SET_DEINTERLACE
:
485 vf
->priv
->do_deinterlace
= 2*!!*(int*)data
;
488 return vf_next_control (vf
, request
, data
);
491 static int vf_open(vf_instance_t
*vf
, char *args
){
494 vf
->put_image
=put_image
;
495 vf
->query_format
=query_format
;
497 vf
->priv
=malloc(sizeof(struct vf_priv_s
));
499 memset(vf
->priv
, 0, sizeof(struct vf_priv_s
));
502 vf
->priv
->parity
= -1;
503 vf
->priv
->do_deinterlace
=1;
505 if (args
) sscanf(args
, "%d:%d", &vf
->priv
->mode
, &vf
->priv
->parity
);
507 filter_line
= filter_line_c
;
509 if(gCpuCaps
.hasMMX2
) filter_line
= filter_line_mmx2
;
515 const vf_info_t vf_info_yadif
= {
516 "Yet Another DeInterlacing Filter",
518 "Michael Niedermayer",