2 Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "cpudetect.h"
34 #include "img_format.h"
37 #include "libvo/fastmemcpy.h"
39 #define MIN(a,b) ((a) > (b) ? (b) : (a))
40 #define MAX(a,b) ((a) < (b) ? (b) : (a))
41 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
43 #define MIN3(a,b,c) MIN(MIN(a,b),c)
44 #define MAX3(a,b,c) MAX(MAX(a,b),c)
46 //===========================================================================//
54 mp_image_t
*buffered_mpi
;
60 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
);
62 static void store_ref(struct vf_priv_s
*p
, uint8_t *src
[3], int src_stride
[3], int width
, int height
){
65 memcpy (p
->ref
[3], p
->ref
[0], sizeof(uint8_t *)*3);
66 memmove(p
->ref
[0], p
->ref
[1], sizeof(uint8_t *)*3*3);
71 memcpy_pic(p
->ref
[2][i
], src
[i
], width
>>is_chroma
, height
>>is_chroma
, p
->stride
[i
], src_stride
[i
]);
75 #if defined(HAVE_MMX) && defined(NAMED_ASM_ARGS)
77 #define LOAD4(mem,dst) \
78 "movd "mem", "#dst" \n\t"\
79 "punpcklbw %%mm7, "#dst" \n\t"
81 #define PABS(tmp,dst) \
82 "pxor "#tmp", "#tmp" \n\t"\
83 "psubw "#dst", "#tmp" \n\t"\
84 "pmaxsw "#tmp", "#dst" \n\t"
86 #define CHECK(pj,mj) \
87 "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
88 "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
89 "movq %%mm2, %%mm4 \n\t"\
90 "movq %%mm2, %%mm5 \n\t"\
91 "pxor %%mm3, %%mm4 \n\t"\
92 "pavgb %%mm3, %%mm5 \n\t"\
93 "pand %[pb1], %%mm4 \n\t"\
94 "psubusb %%mm4, %%mm5 \n\t"\
95 "psrlq $8, %%mm5 \n\t"\
96 "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
97 "movq %%mm2, %%mm4 \n\t"\
98 "psubusb %%mm3, %%mm2 \n\t"\
99 "psubusb %%mm4, %%mm3 \n\t"\
100 "pmaxub %%mm3, %%mm2 \n\t"\
101 "movq %%mm2, %%mm3 \n\t"\
102 "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
103 "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
104 "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
105 "punpcklbw %%mm7, %%mm2 \n\t"\
106 "punpcklbw %%mm7, %%mm3 \n\t"\
107 "punpcklbw %%mm7, %%mm4 \n\t"\
108 "paddw %%mm3, %%mm2 \n\t"\
109 "paddw %%mm4, %%mm2 \n\t" /* score */
112 "movq %%mm0, %%mm3 \n\t"\
113 "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
114 "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
115 "movq %%mm3, %%mm6 \n\t"\
116 "pand %%mm3, %%mm5 \n\t"\
117 "pandn %%mm1, %%mm3 \n\t"\
118 "por %%mm5, %%mm3 \n\t"\
119 "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
121 #define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
122 hurts both quality and speed, but matches the C version. */\
123 "paddw %[pw1], %%mm6 \n\t"\
124 "psllw $14, %%mm6 \n\t"\
125 "paddsw %%mm6, %%mm2 \n\t"\
126 "movq %%mm0, %%mm3 \n\t"\
127 "pcmpgtw %%mm2, %%mm3 \n\t"\
128 "pminsw %%mm2, %%mm0 \n\t"\
129 "pand %%mm3, %%mm5 \n\t"\
130 "pandn %%mm1, %%mm3 \n\t"\
131 "por %%mm5, %%mm3 \n\t"\
132 "movq %%mm3, %%mm1 \n\t"
134 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
){
135 static const uint64_t pw_1
= 0x0001000100010001ULL
;
136 static const uint64_t pb_1
= 0x0101010101010101ULL
;
137 const int mode
= p
->mode
;
138 uint64_t tmp0
, tmp1
, tmp2
, tmp3
;
142 for(x=0; x<w; x+=4){\
144 "pxor %%mm7, %%mm7 \n\t"\
145 LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
146 LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
147 LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
148 LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
149 "movq %%mm3, %%mm4 \n\t"\
150 "paddw %%mm2, %%mm3 \n\t"\
151 "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
152 "movq %%mm0, %[tmp0] \n\t" /* c */\
153 "movq %%mm3, %[tmp1] \n\t" /* d */\
154 "movq %%mm1, %[tmp2] \n\t" /* e */\
155 "psubw %%mm4, %%mm2 \n\t"\
156 PABS( %%mm4, %%mm2) /* temporal_diff0 */\
157 LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
158 LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
159 "psubw %%mm0, %%mm3 \n\t"\
160 "psubw %%mm1, %%mm4 \n\t"\
163 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
164 "psrlw $1, %%mm2 \n\t"\
165 "psrlw $1, %%mm3 \n\t"\
166 "pmaxsw %%mm3, %%mm2 \n\t"\
167 LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
168 LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
169 "psubw %%mm0, %%mm3 \n\t"\
170 "psubw %%mm1, %%mm4 \n\t"\
173 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
174 "psrlw $1, %%mm3 \n\t"\
175 "pmaxsw %%mm3, %%mm2 \n\t"\
176 "movq %%mm2, %[tmp3] \n\t" /* diff */\
178 "paddw %%mm0, %%mm1 \n\t"\
179 "paddw %%mm0, %%mm0 \n\t"\
180 "psubw %%mm1, %%mm0 \n\t"\
181 "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
182 PABS( %%mm2, %%mm0) /* ABS(c-e) */\
184 "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
185 "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
186 "movq %%mm2, %%mm4 \n\t"\
187 "psubusb %%mm3, %%mm2 \n\t"\
188 "psubusb %%mm4, %%mm3 \n\t"\
189 "pmaxub %%mm3, %%mm2 \n\t"\
190 "pshufw $9,%%mm2, %%mm3 \n\t"\
191 "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
192 "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
193 "paddw %%mm2, %%mm0 \n\t"\
194 "paddw %%mm3, %%mm0 \n\t"\
195 "psubw %[pw1], %%mm0 \n\t" /* spatial_score */\
206 /* if(p->mode<2) ... */\
207 "movq %[tmp3], %%mm6 \n\t" /* diff */\
208 "cmp $2, %[mode] \n\t"\
210 LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
211 LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
212 LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
213 LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
214 "paddw %%mm4, %%mm2 \n\t"\
215 "paddw %%mm5, %%mm3 \n\t"\
216 "psrlw $1, %%mm2 \n\t" /* b */\
217 "psrlw $1, %%mm3 \n\t" /* f */\
218 "movq %[tmp0], %%mm4 \n\t" /* c */\
219 "movq %[tmp1], %%mm5 \n\t" /* d */\
220 "movq %[tmp2], %%mm7 \n\t" /* e */\
221 "psubw %%mm4, %%mm2 \n\t" /* b-c */\
222 "psubw %%mm7, %%mm3 \n\t" /* f-e */\
223 "movq %%mm5, %%mm0 \n\t"\
224 "psubw %%mm4, %%mm5 \n\t" /* d-c */\
225 "psubw %%mm7, %%mm0 \n\t" /* d-e */\
226 "movq %%mm2, %%mm4 \n\t"\
227 "pminsw %%mm3, %%mm2 \n\t"\
228 "pmaxsw %%mm4, %%mm3 \n\t"\
229 "pmaxsw %%mm5, %%mm2 \n\t"\
230 "pminsw %%mm5, %%mm3 \n\t"\
231 "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
232 "pminsw %%mm0, %%mm3 \n\t" /* min */\
233 "pxor %%mm4, %%mm4 \n\t"\
234 "pmaxsw %%mm3, %%mm6 \n\t"\
235 "psubw %%mm2, %%mm4 \n\t" /* -max */\
236 "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
239 "movq %[tmp1], %%mm2 \n\t" /* d */\
240 "movq %%mm2, %%mm3 \n\t"\
241 "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
242 "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
243 "pmaxsw %%mm2, %%mm1 \n\t"\
244 "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
245 "packuswb %%mm1, %%mm1 \n\t"\
254 [prefs]"r"((long)refs),\
255 [mrefs]"r"((long)-refs),\
260 asm volatile("movd %%mm1, %0" :"=m"(*dst));\
288 #endif /* defined(HAVE_MMX) && defined(NAMED_ASM_ARGS) */
290 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
){
292 uint8_t *prev2
= parity
? prev
: cur
;
293 uint8_t *next2
= parity
? cur
: next
;
296 int d
= (prev2
[0] + next2
[0])>>1;
298 int temporal_diff0
= ABS(prev2
[0] - next2
[0]);
299 int temporal_diff1
=( ABS(prev
[-refs
] - c
) + ABS(prev
[+refs
] - e
) )>>1;
300 int temporal_diff2
=( ABS(next
[-refs
] - c
) + ABS(next
[+refs
] - e
) )>>1;
301 int diff
= MAX3(temporal_diff0
>>1, temporal_diff1
, temporal_diff2
);
302 int spatial_pred
= (c
+e
)>>1;
303 int spatial_score
= ABS(cur
[-refs
-1] - cur
[+refs
-1]) + ABS(c
-e
)
304 + ABS(cur
[-refs
+1] - cur
[+refs
+1]) - 1;
307 { int score= ABS(cur[-refs-1+j] - cur[+refs-1-j])\
308 + ABS(cur[-refs +j] - cur[+refs -j])\
309 + ABS(cur[-refs+1+j] - cur[+refs+1-j]);\
310 if(score < spatial_score){\
311 spatial_score= score;\
312 spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
314 CHECK(-1) CHECK(-2) }} }}
315 CHECK( 1) CHECK( 2) }} }}
318 int b
= (prev2
[-2*refs
] + next2
[-2*refs
])>>1;
319 int f
= (prev2
[+2*refs
] + next2
[+2*refs
])>>1;
323 int max
= MAX3(d
-e
, d
-c
, MIN3(MAX(b
-c
,f
-e
),MAX(b
-c
,b
-a
),MAX(f
-g
,f
-e
)) );
324 int min
= MIN3(d
-e
, d
-c
, MAX3(MIN(b
-c
,f
-e
),MIN(b
-c
,b
-a
),MIN(f
-g
,f
-e
)) );
326 int max
= MAX3(d
-e
, d
-c
, MIN(b
-c
, f
-e
));
327 int min
= MIN3(d
-e
, d
-c
, MAX(b
-c
, f
-e
));
330 diff
= MAX3(diff
, min
, -max
);
333 if(spatial_pred
> d
+ diff
)
334 spatial_pred
= d
+ diff
;
335 else if(spatial_pred
< d
- diff
)
336 spatial_pred
= d
- diff
;
338 dst
[0] = spatial_pred
;
349 static void filter(struct vf_priv_s
*p
, uint8_t *dst
[3], int dst_stride
[3], int width
, int height
, int parity
, int tff
){
354 int w
= width
>>is_chroma
;
355 int h
= height
>>is_chroma
;
356 int refs
= p
->stride
[i
];
359 if((y
^ parity
) & 1){
360 uint8_t *prev
= &p
->ref
[0][i
][y
*refs
];
361 uint8_t *cur
= &p
->ref
[1][i
][y
*refs
];
362 uint8_t *next
= &p
->ref
[2][i
][y
*refs
];
363 uint8_t *dst2
= &dst
[i
][y
*dst_stride
[i
]];
364 filter_line(p
, dst2
, prev
, cur
, next
, w
, refs
, parity
^ tff
);
366 fast_memcpy(&dst
[i
][y
*dst_stride
[i
]], &p
->ref
[1][i
][y
*refs
], w
);
370 #if defined(HAVE_MMX) && defined(NAMED_ASM_ARGS)
371 if(gCpuCaps
.hasMMX2
) asm volatile("emms \n\t" : : : "memory");
375 static int config(struct vf_instance_s
* vf
,
376 int width
, int height
, int d_width
, int d_height
,
377 unsigned int flags
, unsigned int outfmt
){
382 int w
= ((width
+ 31) & (~31))>>is_chroma
;
383 int h
= ((height
+6+ 31) & (~31))>>is_chroma
;
385 vf
->priv
->stride
[i
]= w
;
387 vf
->priv
->ref
[j
][i
]= malloc(w
*h
*sizeof(uint8_t))+3*w
;
390 return vf_next_config(vf
,width
,height
,d_width
,d_height
,flags
,outfmt
);
393 static int continue_buffered_image(struct vf_instance_s
*vf
);
394 extern int correct_pts
;
396 static int put_image(struct vf_instance_s
* vf
, mp_image_t
*mpi
, double pts
){
399 if(vf
->priv
->parity
< 0) {
400 if (mpi
->fields
& MP_IMGFIELD_ORDERED
)
401 tff
= !!(mpi
->fields
& MP_IMGFIELD_TOP_FIRST
);
405 else tff
= (vf
->priv
->parity
&1)^1;
407 store_ref(vf
->priv
, mpi
->planes
, mpi
->stride
, mpi
->w
, mpi
->h
);
409 vf
->priv
->buffered_mpi
= mpi
;
410 vf
->priv
->buffered_tff
= tff
;
411 vf
->priv
->buffered_i
= 0;
412 vf
->priv
->buffered_pts
= pts
;
414 if(vf
->priv
->do_deinterlace
== 0)
415 return vf_next_put_image(vf
, mpi
, pts
);
416 else if(vf
->priv
->do_deinterlace
== 1){
417 vf
->priv
->do_deinterlace
= 2;
420 return continue_buffered_image(vf
);
423 static int continue_buffered_image(struct vf_instance_s
*vf
)
425 mp_image_t
*mpi
= vf
->priv
->buffered_mpi
;
426 int tff
= vf
->priv
->buffered_tff
;
427 double pts
= vf
->priv
->buffered_pts
;
432 pts
+= vf
->priv
->buffered_i
* .02; // XXX not right
434 for(i
= vf
->priv
->buffered_i
; i
<=(vf
->priv
->mode
&1); i
++){
435 dmpi
=vf_get_image(vf
->next
,mpi
->imgfmt
,
437 MP_IMGFLAG_ACCEPT_STRIDE
|MP_IMGFLAG_PREFER_ALIGNED_STRIDE
,
438 mpi
->width
,mpi
->height
);
439 vf_clone_mpi_attributes(dmpi
, mpi
);
440 filter(vf
->priv
, dmpi
->planes
, dmpi
->stride
, mpi
->w
, mpi
->h
, i
^ tff
^ 1, tff
);
441 if (correct_pts
&& i
< (vf
->priv
->mode
& 1))
442 vf_queue_frame(vf
, continue_buffered_image
);
443 ret
|= vf_next_put_image(vf
, dmpi
, pts
/*FIXME*/);
446 if(i
<(vf
->priv
->mode
&1))
447 vf_next_control(vf
, VFCTRL_FLIP_PAGE
, NULL
);
449 vf
->priv
->buffered_i
= 1;
453 static void uninit(struct vf_instance_s
* 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_s
* vf
, unsigned int fmt
){
474 return vf_next_query_format(vf
,fmt
);
479 static int control(struct vf_instance_s
* 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 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
;
508 #if defined(HAVE_MMX) && defined(NAMED_ASM_ARGS)
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",