Add support for VDPAU video out, including hardware decoding.
[mplayer/glamo.git] / libmpcodecs / vf_yadif.c
blob5a36b3caeae63bc96f8ca02593645d8bed3e1fde
1 /*
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.
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <inttypes.h>
25 #include <math.h>
27 #include "config.h"
28 #include "cpudetect.h"
30 #include "mp_msg.h"
32 #ifdef HAVE_MALLOC_H
33 #include <malloc.h>
34 #endif
36 #include "img_format.h"
37 #include "mp_image.h"
38 #include "vf.h"
39 #include "libvo/fastmemcpy.h"
40 #include "libavutil/common.h"
42 //===========================================================================//
44 struct vf_priv_s {
45 int mode;
46 int parity;
47 int buffered_i;
48 int buffered_tff;
49 double buffered_pts;
50 mp_image_t *buffered_mpi;
51 int stride[3];
52 uint8_t *ref[4][3];
53 int do_deinterlace;
56 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);
58 static void store_ref(struct vf_priv_s *p, uint8_t *src[3], int src_stride[3], int width, int height){
59 int i;
61 memcpy (p->ref[3], p->ref[0], sizeof(uint8_t *)*3);
62 memmove(p->ref[0], p->ref[1], sizeof(uint8_t *)*3*3);
64 for(i=0; i<3; i++){
65 int is_chroma= !!i;
67 memcpy_pic(p->ref[2][i], src[i], width>>is_chroma, height>>is_chroma, p->stride[i], src_stride[i]);
71 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
73 #define LOAD4(mem,dst) \
74 "movd "mem", "#dst" \n\t"\
75 "punpcklbw %%mm7, "#dst" \n\t"
77 #define PABS(tmp,dst) \
78 "pxor "#tmp", "#tmp" \n\t"\
79 "psubw "#dst", "#tmp" \n\t"\
80 "pmaxsw "#tmp", "#dst" \n\t"
82 #define CHECK(pj,mj) \
83 "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
84 "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
85 "movq %%mm2, %%mm4 \n\t"\
86 "movq %%mm2, %%mm5 \n\t"\
87 "pxor %%mm3, %%mm4 \n\t"\
88 "pavgb %%mm3, %%mm5 \n\t"\
89 "pand %[pb1], %%mm4 \n\t"\
90 "psubusb %%mm4, %%mm5 \n\t"\
91 "psrlq $8, %%mm5 \n\t"\
92 "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
93 "movq %%mm2, %%mm4 \n\t"\
94 "psubusb %%mm3, %%mm2 \n\t"\
95 "psubusb %%mm4, %%mm3 \n\t"\
96 "pmaxub %%mm3, %%mm2 \n\t"\
97 "movq %%mm2, %%mm3 \n\t"\
98 "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
99 "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
100 "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
101 "punpcklbw %%mm7, %%mm2 \n\t"\
102 "punpcklbw %%mm7, %%mm3 \n\t"\
103 "punpcklbw %%mm7, %%mm4 \n\t"\
104 "paddw %%mm3, %%mm2 \n\t"\
105 "paddw %%mm4, %%mm2 \n\t" /* score */
107 #define CHECK1 \
108 "movq %%mm0, %%mm3 \n\t"\
109 "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
110 "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
111 "movq %%mm3, %%mm6 \n\t"\
112 "pand %%mm3, %%mm5 \n\t"\
113 "pandn %%mm1, %%mm3 \n\t"\
114 "por %%mm5, %%mm3 \n\t"\
115 "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
117 #define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
118 hurts both quality and speed, but matches the C version. */\
119 "paddw %[pw1], %%mm6 \n\t"\
120 "psllw $14, %%mm6 \n\t"\
121 "paddsw %%mm6, %%mm2 \n\t"\
122 "movq %%mm0, %%mm3 \n\t"\
123 "pcmpgtw %%mm2, %%mm3 \n\t"\
124 "pminsw %%mm2, %%mm0 \n\t"\
125 "pand %%mm3, %%mm5 \n\t"\
126 "pandn %%mm1, %%mm3 \n\t"\
127 "por %%mm5, %%mm3 \n\t"\
128 "movq %%mm3, %%mm1 \n\t"
130 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){
131 static const uint64_t pw_1 = 0x0001000100010001ULL;
132 static const uint64_t pb_1 = 0x0101010101010101ULL;
133 const int mode = p->mode;
134 uint64_t tmp0, tmp1, tmp2, tmp3;
135 int x;
137 #define FILTER\
138 for(x=0; x<w; x+=4){\
139 __asm__ volatile(\
140 "pxor %%mm7, %%mm7 \n\t"\
141 LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
142 LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
143 LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
144 LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
145 "movq %%mm3, %%mm4 \n\t"\
146 "paddw %%mm2, %%mm3 \n\t"\
147 "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
148 "movq %%mm0, %[tmp0] \n\t" /* c */\
149 "movq %%mm3, %[tmp1] \n\t" /* d */\
150 "movq %%mm1, %[tmp2] \n\t" /* e */\
151 "psubw %%mm4, %%mm2 \n\t"\
152 PABS( %%mm4, %%mm2) /* temporal_diff0 */\
153 LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
154 LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
155 "psubw %%mm0, %%mm3 \n\t"\
156 "psubw %%mm1, %%mm4 \n\t"\
157 PABS( %%mm5, %%mm3)\
158 PABS( %%mm5, %%mm4)\
159 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
160 "psrlw $1, %%mm2 \n\t"\
161 "psrlw $1, %%mm3 \n\t"\
162 "pmaxsw %%mm3, %%mm2 \n\t"\
163 LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
164 LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
165 "psubw %%mm0, %%mm3 \n\t"\
166 "psubw %%mm1, %%mm4 \n\t"\
167 PABS( %%mm5, %%mm3)\
168 PABS( %%mm5, %%mm4)\
169 "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
170 "psrlw $1, %%mm3 \n\t"\
171 "pmaxsw %%mm3, %%mm2 \n\t"\
172 "movq %%mm2, %[tmp3] \n\t" /* diff */\
174 "paddw %%mm0, %%mm1 \n\t"\
175 "paddw %%mm0, %%mm0 \n\t"\
176 "psubw %%mm1, %%mm0 \n\t"\
177 "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
178 PABS( %%mm2, %%mm0) /* ABS(c-e) */\
180 "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
181 "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
182 "movq %%mm2, %%mm4 \n\t"\
183 "psubusb %%mm3, %%mm2 \n\t"\
184 "psubusb %%mm4, %%mm3 \n\t"\
185 "pmaxub %%mm3, %%mm2 \n\t"\
186 "pshufw $9,%%mm2, %%mm3 \n\t"\
187 "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
188 "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
189 "paddw %%mm2, %%mm0 \n\t"\
190 "paddw %%mm3, %%mm0 \n\t"\
191 "psubw %[pw1], %%mm0 \n\t" /* spatial_score */\
193 CHECK(-2,0)\
194 CHECK1\
195 CHECK(-3,1)\
196 CHECK2\
197 CHECK(0,-2)\
198 CHECK1\
199 CHECK(1,-3)\
200 CHECK2\
202 /* if(p->mode<2) ... */\
203 "movq %[tmp3], %%mm6 \n\t" /* diff */\
204 "cmp $2, %[mode] \n\t"\
205 "jge 1f \n\t"\
206 LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
207 LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
208 LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
209 LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
210 "paddw %%mm4, %%mm2 \n\t"\
211 "paddw %%mm5, %%mm3 \n\t"\
212 "psrlw $1, %%mm2 \n\t" /* b */\
213 "psrlw $1, %%mm3 \n\t" /* f */\
214 "movq %[tmp0], %%mm4 \n\t" /* c */\
215 "movq %[tmp1], %%mm5 \n\t" /* d */\
216 "movq %[tmp2], %%mm7 \n\t" /* e */\
217 "psubw %%mm4, %%mm2 \n\t" /* b-c */\
218 "psubw %%mm7, %%mm3 \n\t" /* f-e */\
219 "movq %%mm5, %%mm0 \n\t"\
220 "psubw %%mm4, %%mm5 \n\t" /* d-c */\
221 "psubw %%mm7, %%mm0 \n\t" /* d-e */\
222 "movq %%mm2, %%mm4 \n\t"\
223 "pminsw %%mm3, %%mm2 \n\t"\
224 "pmaxsw %%mm4, %%mm3 \n\t"\
225 "pmaxsw %%mm5, %%mm2 \n\t"\
226 "pminsw %%mm5, %%mm3 \n\t"\
227 "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
228 "pminsw %%mm0, %%mm3 \n\t" /* min */\
229 "pxor %%mm4, %%mm4 \n\t"\
230 "pmaxsw %%mm3, %%mm6 \n\t"\
231 "psubw %%mm2, %%mm4 \n\t" /* -max */\
232 "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
233 "1: \n\t"\
235 "movq %[tmp1], %%mm2 \n\t" /* d */\
236 "movq %%mm2, %%mm3 \n\t"\
237 "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
238 "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
239 "pmaxsw %%mm2, %%mm1 \n\t"\
240 "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
241 "packuswb %%mm1, %%mm1 \n\t"\
243 :[tmp0]"=m"(tmp0),\
244 [tmp1]"=m"(tmp1),\
245 [tmp2]"=m"(tmp2),\
246 [tmp3]"=m"(tmp3)\
247 :[prev] "r"(prev),\
248 [cur] "r"(cur),\
249 [next] "r"(next),\
250 [prefs]"r"((long)refs),\
251 [mrefs]"r"((long)-refs),\
252 [pw1] "m"(pw_1),\
253 [pb1] "m"(pb_1),\
254 [mode] "g"(mode)\
256 __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
257 dst += 4;\
258 prev+= 4;\
259 cur += 4;\
260 next+= 4;\
263 if(parity){
264 #define prev2 "prev"
265 #define next2 "cur"
266 FILTER
267 #undef prev2
268 #undef next2
269 }else{
270 #define prev2 "cur"
271 #define next2 "next"
272 FILTER
273 #undef prev2
274 #undef next2
277 #undef LOAD4
278 #undef PABS
279 #undef CHECK
280 #undef CHECK1
281 #undef CHECK2
282 #undef FILTER
284 #endif /* HAVE_MMX && defined(NAMED_ASM_ARGS) */
286 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){
287 int x;
288 uint8_t *prev2= parity ? prev : cur ;
289 uint8_t *next2= parity ? cur : next;
290 for(x=0; x<w; x++){
291 int c= cur[-refs];
292 int d= (prev2[0] + next2[0])>>1;
293 int e= cur[+refs];
294 int temporal_diff0= FFABS(prev2[0] - next2[0]);
295 int temporal_diff1=( FFABS(prev[-refs] - c) + FFABS(prev[+refs] - e) )>>1;
296 int temporal_diff2=( FFABS(next[-refs] - c) + FFABS(next[+refs] - e) )>>1;
297 int diff= FFMAX3(temporal_diff0>>1, temporal_diff1, temporal_diff2);
298 int spatial_pred= (c+e)>>1;
299 int spatial_score= FFABS(cur[-refs-1] - cur[+refs-1]) + FFABS(c-e)
300 + FFABS(cur[-refs+1] - cur[+refs+1]) - 1;
302 #define CHECK(j)\
303 { int score= FFABS(cur[-refs-1+j] - cur[+refs-1-j])\
304 + FFABS(cur[-refs +j] - cur[+refs -j])\
305 + FFABS(cur[-refs+1+j] - cur[+refs+1-j]);\
306 if(score < spatial_score){\
307 spatial_score= score;\
308 spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
310 CHECK(-1) CHECK(-2) }} }}
311 CHECK( 1) CHECK( 2) }} }}
313 if(p->mode<2){
314 int b= (prev2[-2*refs] + next2[-2*refs])>>1;
315 int f= (prev2[+2*refs] + next2[+2*refs])>>1;
316 #if 0
317 int a= cur[-3*refs];
318 int g= cur[+3*refs];
319 int max= FFMAX3(d-e, d-c, FFMIN3(FFMAX(b-c,f-e),FFMAX(b-c,b-a),FFMAX(f-g,f-e)) );
320 int min= FFMIN3(d-e, d-c, FFMAX3(FFMIN(b-c,f-e),FFMIN(b-c,b-a),FFMIN(f-g,f-e)) );
321 #else
322 int max= FFMAX3(d-e, d-c, FFMIN(b-c, f-e));
323 int min= FFMIN3(d-e, d-c, FFMAX(b-c, f-e));
324 #endif
326 diff= FFMAX3(diff, min, -max);
329 if(spatial_pred > d + diff)
330 spatial_pred = d + diff;
331 else if(spatial_pred < d - diff)
332 spatial_pred = d - diff;
334 dst[0] = spatial_pred;
336 dst++;
337 cur++;
338 prev++;
339 next++;
340 prev2++;
341 next2++;
345 static void filter(struct vf_priv_s *p, uint8_t *dst[3], int dst_stride[3], int width, int height, int parity, int tff){
346 int y, i;
348 for(i=0; i<3; i++){
349 int is_chroma= !!i;
350 int w= width >>is_chroma;
351 int h= height>>is_chroma;
352 int refs= p->stride[i];
354 for(y=0; y<h; y++){
355 if((y ^ parity) & 1){
356 uint8_t *prev= &p->ref[0][i][y*refs];
357 uint8_t *cur = &p->ref[1][i][y*refs];
358 uint8_t *next= &p->ref[2][i][y*refs];
359 uint8_t *dst2= &dst[i][y*dst_stride[i]];
360 filter_line(p, dst2, prev, cur, next, w, refs, parity ^ tff);
361 }else{
362 fast_memcpy(&dst[i][y*dst_stride[i]], &p->ref[1][i][y*refs], w);
366 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
367 if(gCpuCaps.hasMMX2) __asm__ volatile("emms \n\t" : : : "memory");
368 #endif
371 static int config(struct vf_instance_s* vf,
372 int width, int height, int d_width, int d_height,
373 unsigned int flags, unsigned int outfmt){
374 int i, j;
376 for(i=0; i<3; i++){
377 int is_chroma= !!i;
378 int w= ((width + 31) & (~31))>>is_chroma;
379 int h= ((height+6+ 31) & (~31))>>is_chroma;
381 vf->priv->stride[i]= w;
382 for(j=0; j<3; j++)
383 vf->priv->ref[j][i]= malloc(w*h*sizeof(uint8_t))+3*w;
386 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
389 static int continue_buffered_image(struct vf_instance_s *vf);
390 extern int correct_pts;
392 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts){
393 int tff;
395 if(vf->priv->parity < 0) {
396 if (mpi->fields & MP_IMGFIELD_ORDERED)
397 tff = !!(mpi->fields & MP_IMGFIELD_TOP_FIRST);
398 else
399 tff = 1;
401 else tff = (vf->priv->parity&1)^1;
403 store_ref(vf->priv, mpi->planes, mpi->stride, mpi->w, mpi->h);
405 vf->priv->buffered_mpi = mpi;
406 vf->priv->buffered_tff = tff;
407 vf->priv->buffered_i = 0;
408 vf->priv->buffered_pts = pts;
410 if(vf->priv->do_deinterlace == 0)
411 return vf_next_put_image(vf, mpi, pts);
412 else if(vf->priv->do_deinterlace == 1){
413 vf->priv->do_deinterlace= 2;
414 return 0;
415 }else
416 return continue_buffered_image(vf);
419 static int continue_buffered_image(struct vf_instance_s *vf)
421 mp_image_t *mpi = vf->priv->buffered_mpi;
422 int tff = vf->priv->buffered_tff;
423 double pts = vf->priv->buffered_pts;
424 int i;
425 int ret=0;
426 mp_image_t *dmpi;
428 pts += vf->priv->buffered_i * .02; // XXX not right
430 for(i = vf->priv->buffered_i; i<=(vf->priv->mode&1); i++){
431 dmpi=vf_get_image(vf->next,mpi->imgfmt,
432 MP_IMGTYPE_TEMP,
433 MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
434 mpi->width,mpi->height);
435 vf_clone_mpi_attributes(dmpi, mpi);
436 filter(vf->priv, dmpi->planes, dmpi->stride, mpi->w, mpi->h, i ^ tff ^ 1, tff);
437 if (correct_pts && i < (vf->priv->mode & 1))
438 vf_queue_frame(vf, continue_buffered_image);
439 ret |= vf_next_put_image(vf, dmpi, pts /*FIXME*/);
440 if (correct_pts)
441 break;
442 if(i<(vf->priv->mode&1))
443 vf_extra_flip(vf);
445 vf->priv->buffered_i = 1;
446 return ret;
449 static void uninit(struct vf_instance_s* vf){
450 int i;
451 if(!vf->priv) return;
453 for(i=0; i<3*3; i++){
454 uint8_t **p= &vf->priv->ref[i%3][i/3];
455 if(*p) free(*p - 3*vf->priv->stride[i/3]);
456 *p= NULL;
458 free(vf->priv);
459 vf->priv=NULL;
462 //===========================================================================//
463 static int query_format(struct vf_instance_s* vf, unsigned int fmt){
464 switch(fmt){
465 case IMGFMT_YV12:
466 case IMGFMT_I420:
467 case IMGFMT_IYUV:
468 case IMGFMT_Y800:
469 case IMGFMT_Y8:
470 return vf_next_query_format(vf,fmt);
472 return 0;
475 static int control(struct vf_instance_s* vf, int request, void* data){
476 switch (request){
477 case VFCTRL_GET_DEINTERLACE:
478 *(int*)data = vf->priv->do_deinterlace;
479 return CONTROL_OK;
480 case VFCTRL_SET_DEINTERLACE:
481 vf->priv->do_deinterlace = 2*!!*(int*)data;
482 return CONTROL_OK;
484 return vf_next_control (vf, request, data);
487 static int open(vf_instance_t *vf, char* args){
489 vf->config=config;
490 vf->put_image=put_image;
491 vf->query_format=query_format;
492 vf->uninit=uninit;
493 vf->priv=malloc(sizeof(struct vf_priv_s));
494 vf->control=control;
495 memset(vf->priv, 0, sizeof(struct vf_priv_s));
497 vf->priv->mode=0;
498 vf->priv->parity= -1;
499 vf->priv->do_deinterlace=1;
501 if (args) sscanf(args, "%d:%d", &vf->priv->mode, &vf->priv->parity);
503 filter_line = filter_line_c;
504 #if HAVE_MMX && defined(NAMED_ASM_ARGS)
505 if(gCpuCaps.hasMMX2) filter_line = filter_line_mmx2;
506 #endif
508 return 1;
511 const vf_info_t vf_info_yadif = {
512 "Yet Another DeInterlacing Filter",
513 "yadif",
514 "Michael Niedermayer",
516 open,
517 NULL