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"
31 #include "img_format.h"
34 #include "libvo/fastmemcpy.h"
35 #include "libavutil/common.h"
37 //===========================================================================//
45 mp_image_t
*buffered_mpi
;
51 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
);
53 static void store_ref(struct vf_priv_s
*p
, uint8_t *src
[3], int src_stride
[3], int width
, int height
){
56 memcpy (p
->ref
[3], p
->ref
[0], sizeof(uint8_t *)*3);
57 memmove(p
->ref
[0], p
->ref
[1], sizeof(uint8_t *)*3*3);
62 memcpy_pic(p
->ref
[2][i
], src
[i
], width
>>is_chroma
, height
>>is_chroma
, p
->stride
[i
], src_stride
[i
]);
66 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
68 #define LOAD4(mem,dst) \
69 "movd "mem", "#dst" \n\t"\
70 "punpcklbw %%mm7, "#dst" \n\t"
72 #define PABS(tmp,dst) \
73 "pxor "#tmp", "#tmp" \n\t"\
74 "psubw "#dst", "#tmp" \n\t"\
75 "pmaxsw "#tmp", "#dst" \n\t"
77 #define CHECK(pj,mj) \
78 "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
79 "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
80 "movq %%mm2, %%mm4 \n\t"\
81 "movq %%mm2, %%mm5 \n\t"\
82 "pxor %%mm3, %%mm4 \n\t"\
83 "pavgb %%mm3, %%mm5 \n\t"\
84 "pand %[pb1], %%mm4 \n\t"\
85 "psubusb %%mm4, %%mm5 \n\t"\
86 "psrlq $8, %%mm5 \n\t"\
87 "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
88 "movq %%mm2, %%mm4 \n\t"\
89 "psubusb %%mm3, %%mm2 \n\t"\
90 "psubusb %%mm4, %%mm3 \n\t"\
91 "pmaxub %%mm3, %%mm2 \n\t"\
92 "movq %%mm2, %%mm3 \n\t"\
93 "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
94 "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
95 "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
96 "punpcklbw %%mm7, %%mm2 \n\t"\
97 "punpcklbw %%mm7, %%mm3 \n\t"\
98 "punpcklbw %%mm7, %%mm4 \n\t"\
99 "paddw %%mm3, %%mm2 \n\t"\
100 "paddw %%mm4, %%mm2 \n\t" /* score */
103 "movq %%mm0, %%mm3 \n\t"\
104 "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
105 "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
106 "movq %%mm3, %%mm6 \n\t"\
107 "pand %%mm3, %%mm5 \n\t"\
108 "pandn %%mm1, %%mm3 \n\t"\
109 "por %%mm5, %%mm3 \n\t"\
110 "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
112 #define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
113 hurts both quality and speed, but matches the C version. */\
114 "paddw %[pw1], %%mm6 \n\t"\
115 "psllw $14, %%mm6 \n\t"\
116 "paddsw %%mm6, %%mm2 \n\t"\
117 "movq %%mm0, %%mm3 \n\t"\
118 "pcmpgtw %%mm2, %%mm3 \n\t"\
119 "pminsw %%mm2, %%mm0 \n\t"\
120 "pand %%mm3, %%mm5 \n\t"\
121 "pandn %%mm1, %%mm3 \n\t"\
122 "por %%mm5, %%mm3 \n\t"\
123 "movq %%mm3, %%mm1 \n\t"
125 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
){
126 static const uint64_t pw_1
= 0x0001000100010001ULL
;
127 static const uint64_t pb_1
= 0x0101010101010101ULL
;
128 const int mode
= p
->mode
;
129 uint64_t tmp0
, tmp1
, tmp2
, tmp3
;
133 for(x=0; x<w; x+=4){\
135 "pxor %%mm7, %%mm7 \n\t"\
136 LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
137 LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
138 LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
139 LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
140 "movq %%mm3, %%mm4 \n\t"\
141 "paddw %%mm2, %%mm3 \n\t"\
142 "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
143 "movq %%mm0, %[tmp0] \n\t" /* c */\
144 "movq %%mm3, %[tmp1] \n\t" /* d */\
145 "movq %%mm1, %[tmp2] \n\t" /* e */\
146 "psubw %%mm4, %%mm2 \n\t"\
147 PABS( %%mm4, %%mm2) /* temporal_diff0 */\
148 LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
149 LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
150 "psubw %%mm0, %%mm3 \n\t"\
151 "psubw %%mm1, %%mm4 \n\t"\
154 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
155 "psrlw $1, %%mm2 \n\t"\
156 "psrlw $1, %%mm3 \n\t"\
157 "pmaxsw %%mm3, %%mm2 \n\t"\
158 LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
159 LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
160 "psubw %%mm0, %%mm3 \n\t"\
161 "psubw %%mm1, %%mm4 \n\t"\
164 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
165 "psrlw $1, %%mm3 \n\t"\
166 "pmaxsw %%mm3, %%mm2 \n\t"\
167 "movq %%mm2, %[tmp3] \n\t" /* diff */\
169 "paddw %%mm0, %%mm1 \n\t"\
170 "paddw %%mm0, %%mm0 \n\t"\
171 "psubw %%mm1, %%mm0 \n\t"\
172 "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
173 PABS( %%mm2, %%mm0) /* ABS(c-e) */\
175 "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
176 "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
177 "movq %%mm2, %%mm4 \n\t"\
178 "psubusb %%mm3, %%mm2 \n\t"\
179 "psubusb %%mm4, %%mm3 \n\t"\
180 "pmaxub %%mm3, %%mm2 \n\t"\
181 "pshufw $9,%%mm2, %%mm3 \n\t"\
182 "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
183 "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
184 "paddw %%mm2, %%mm0 \n\t"\
185 "paddw %%mm3, %%mm0 \n\t"\
186 "psubw %[pw1], %%mm0 \n\t" /* spatial_score */\
197 /* if(p->mode<2) ... */\
198 "movq %[tmp3], %%mm6 \n\t" /* diff */\
199 "cmp $2, %[mode] \n\t"\
201 LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
202 LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
203 LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
204 LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
205 "paddw %%mm4, %%mm2 \n\t"\
206 "paddw %%mm5, %%mm3 \n\t"\
207 "psrlw $1, %%mm2 \n\t" /* b */\
208 "psrlw $1, %%mm3 \n\t" /* f */\
209 "movq %[tmp0], %%mm4 \n\t" /* c */\
210 "movq %[tmp1], %%mm5 \n\t" /* d */\
211 "movq %[tmp2], %%mm7 \n\t" /* e */\
212 "psubw %%mm4, %%mm2 \n\t" /* b-c */\
213 "psubw %%mm7, %%mm3 \n\t" /* f-e */\
214 "movq %%mm5, %%mm0 \n\t"\
215 "psubw %%mm4, %%mm5 \n\t" /* d-c */\
216 "psubw %%mm7, %%mm0 \n\t" /* d-e */\
217 "movq %%mm2, %%mm4 \n\t"\
218 "pminsw %%mm3, %%mm2 \n\t"\
219 "pmaxsw %%mm4, %%mm3 \n\t"\
220 "pmaxsw %%mm5, %%mm2 \n\t"\
221 "pminsw %%mm5, %%mm3 \n\t"\
222 "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
223 "pminsw %%mm0, %%mm3 \n\t" /* min */\
224 "pxor %%mm4, %%mm4 \n\t"\
225 "pmaxsw %%mm3, %%mm6 \n\t"\
226 "psubw %%mm2, %%mm4 \n\t" /* -max */\
227 "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
230 "movq %[tmp1], %%mm2 \n\t" /* d */\
231 "movq %%mm2, %%mm3 \n\t"\
232 "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
233 "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
234 "pmaxsw %%mm2, %%mm1 \n\t"\
235 "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
236 "packuswb %%mm1, %%mm1 \n\t"\
245 [prefs]"r"((x86_reg)refs),\
246 [mrefs]"r"((x86_reg)-refs),\
251 __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
279 #endif /* HAVE_MMX && defined(NAMED_ASM_ARGS) */
281 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
){
283 uint8_t *prev2
= parity
? prev
: cur
;
284 uint8_t *next2
= parity
? cur
: next
;
287 int d
= (prev2
[0] + next2
[0])>>1;
289 int temporal_diff0
= FFABS(prev2
[0] - next2
[0]);
290 int temporal_diff1
=( FFABS(prev
[-refs
] - c
) + FFABS(prev
[+refs
] - e
) )>>1;
291 int temporal_diff2
=( FFABS(next
[-refs
] - c
) + FFABS(next
[+refs
] - e
) )>>1;
292 int diff
= FFMAX3(temporal_diff0
>>1, temporal_diff1
, temporal_diff2
);
293 int spatial_pred
= (c
+e
)>>1;
294 int spatial_score
= FFABS(cur
[-refs
-1] - cur
[+refs
-1]) + FFABS(c
-e
)
295 + FFABS(cur
[-refs
+1] - cur
[+refs
+1]) - 1;
298 { int score= FFABS(cur[-refs-1+j] - cur[+refs-1-j])\
299 + FFABS(cur[-refs +j] - cur[+refs -j])\
300 + FFABS(cur[-refs+1+j] - cur[+refs+1-j]);\
301 if(score < spatial_score){\
302 spatial_score= score;\
303 spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
305 CHECK(-1) CHECK(-2) }} }}
306 CHECK( 1) CHECK( 2) }} }}
309 int b
= (prev2
[-2*refs
] + next2
[-2*refs
])>>1;
310 int f
= (prev2
[+2*refs
] + next2
[+2*refs
])>>1;
314 int max
= FFMAX3(d
-e
, d
-c
, FFMIN3(FFMAX(b
-c
,f
-e
),FFMAX(b
-c
,b
-a
),FFMAX(f
-g
,f
-e
)) );
315 int min
= FFMIN3(d
-e
, d
-c
, FFMAX3(FFMIN(b
-c
,f
-e
),FFMIN(b
-c
,b
-a
),FFMIN(f
-g
,f
-e
)) );
317 int max
= FFMAX3(d
-e
, d
-c
, FFMIN(b
-c
, f
-e
));
318 int min
= FFMIN3(d
-e
, d
-c
, FFMAX(b
-c
, f
-e
));
321 diff
= FFMAX3(diff
, min
, -max
);
324 if(spatial_pred
> d
+ diff
)
325 spatial_pred
= d
+ diff
;
326 else if(spatial_pred
< d
- diff
)
327 spatial_pred
= d
- diff
;
329 dst
[0] = spatial_pred
;
340 static void filter(struct vf_priv_s
*p
, uint8_t *dst
[3], int dst_stride
[3], int width
, int height
, int parity
, int tff
){
345 int w
= width
>>is_chroma
;
346 int h
= height
>>is_chroma
;
347 int refs
= p
->stride
[i
];
350 if((y
^ parity
) & 1){
351 uint8_t *prev
= &p
->ref
[0][i
][y
*refs
];
352 uint8_t *cur
= &p
->ref
[1][i
][y
*refs
];
353 uint8_t *next
= &p
->ref
[2][i
][y
*refs
];
354 uint8_t *dst2
= &dst
[i
][y
*dst_stride
[i
]];
355 filter_line(p
, dst2
, prev
, cur
, next
, w
, refs
, parity
^ tff
);
357 fast_memcpy(&dst
[i
][y
*dst_stride
[i
]], &p
->ref
[1][i
][y
*refs
], w
);
361 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
362 if(gCpuCaps
.hasMMX2
) __asm__
volatile("emms \n\t" : : : "memory");
366 static int config(struct vf_instance_s
* vf
,
367 int width
, int height
, int d_width
, int d_height
,
368 unsigned int flags
, unsigned int outfmt
){
373 int w
= ((width
+ 31) & (~31))>>is_chroma
;
374 int h
= ((height
+6+ 31) & (~31))>>is_chroma
;
376 vf
->priv
->stride
[i
]= w
;
378 vf
->priv
->ref
[j
][i
]= malloc(w
*h
*sizeof(uint8_t))+3*w
;
381 return vf_next_config(vf
,width
,height
,d_width
,d_height
,flags
,outfmt
);
384 static int continue_buffered_image(struct vf_instance_s
*vf
);
385 extern int correct_pts
;
387 static int put_image(struct vf_instance_s
* vf
, mp_image_t
*mpi
, double pts
){
390 if(vf
->priv
->parity
< 0) {
391 if (mpi
->fields
& MP_IMGFIELD_ORDERED
)
392 tff
= !!(mpi
->fields
& MP_IMGFIELD_TOP_FIRST
);
396 else tff
= (vf
->priv
->parity
&1)^1;
398 store_ref(vf
->priv
, mpi
->planes
, mpi
->stride
, mpi
->w
, mpi
->h
);
400 vf
->priv
->buffered_mpi
= mpi
;
401 vf
->priv
->buffered_tff
= tff
;
402 vf
->priv
->buffered_i
= 0;
403 vf
->priv
->buffered_pts
= pts
;
405 if(vf
->priv
->do_deinterlace
== 0)
406 return vf_next_put_image(vf
, mpi
, pts
);
407 else if(vf
->priv
->do_deinterlace
== 1){
408 vf
->priv
->do_deinterlace
= 2;
411 return continue_buffered_image(vf
);
414 static int continue_buffered_image(struct vf_instance_s
*vf
)
416 mp_image_t
*mpi
= vf
->priv
->buffered_mpi
;
417 int tff
= vf
->priv
->buffered_tff
;
418 double pts
= vf
->priv
->buffered_pts
;
423 pts
+= vf
->priv
->buffered_i
* .02; // XXX not right
425 for(i
= vf
->priv
->buffered_i
; i
<=(vf
->priv
->mode
&1); i
++){
426 dmpi
=vf_get_image(vf
->next
,mpi
->imgfmt
,
428 MP_IMGFLAG_ACCEPT_STRIDE
|MP_IMGFLAG_PREFER_ALIGNED_STRIDE
,
429 mpi
->width
,mpi
->height
);
430 vf_clone_mpi_attributes(dmpi
, mpi
);
431 filter(vf
->priv
, dmpi
->planes
, dmpi
->stride
, mpi
->w
, mpi
->h
, i
^ tff
^ 1, tff
);
432 if (correct_pts
&& i
< (vf
->priv
->mode
& 1))
433 vf_queue_frame(vf
, continue_buffered_image
);
434 ret
|= vf_next_put_image(vf
, dmpi
, pts
/*FIXME*/);
437 if(i
<(vf
->priv
->mode
&1))
440 vf
->priv
->buffered_i
= 1;
444 static void uninit(struct vf_instance_s
* vf
){
446 if(!vf
->priv
) return;
448 for(i
=0; i
<3*3; i
++){
449 uint8_t **p
= &vf
->priv
->ref
[i
%3][i
/3];
450 if(*p
) free(*p
- 3*vf
->priv
->stride
[i
/3]);
457 //===========================================================================//
458 static int query_format(struct vf_instance_s
* vf
, unsigned int fmt
){
465 return vf_next_query_format(vf
,fmt
);
470 static int control(struct vf_instance_s
* vf
, int request
, void* data
){
472 case VFCTRL_GET_DEINTERLACE
:
473 *(int*)data
= vf
->priv
->do_deinterlace
;
475 case VFCTRL_SET_DEINTERLACE
:
476 vf
->priv
->do_deinterlace
= 2*!!*(int*)data
;
479 return vf_next_control (vf
, request
, data
);
482 static int open(vf_instance_t
*vf
, char* args
){
485 vf
->put_image
=put_image
;
486 vf
->query_format
=query_format
;
488 vf
->priv
=malloc(sizeof(struct vf_priv_s
));
490 memset(vf
->priv
, 0, sizeof(struct vf_priv_s
));
493 vf
->priv
->parity
= -1;
494 vf
->priv
->do_deinterlace
=1;
496 if (args
) sscanf(args
, "%d:%d", &vf
->priv
->mode
, &vf
->priv
->parity
);
498 filter_line
= filter_line_c
;
499 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
500 if(gCpuCaps
.hasMMX2
) filter_line
= filter_line_mmx2
;
506 const vf_info_t vf_info_yadif
= {
507 "Yet Another DeInterlacing Filter",
509 "Michael Niedermayer",