?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gl_common: minor cleanup/refactor
[mplayer.git] / libmpcodecs / vf_pp7.c
blob1c0274d24f2cb09163307461d638c6498d2c9864
1 /*
2 * Copyright (C) 2005 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 #include <stdio.h>
23 #include <stdlib.h>
24 #include <string.h>
25 #include <inttypes.h>
26 #include <math.h>
27
28 #include "config.h"
29
30 #include "mp_msg.h"
31 #include "cpudetect.h"
32
33 #if HAVE_MALLOC_H
34 #include <malloc.h>
35 #endif
36
37 #include "libavutil/mem.h"
38
39 #include "img_format.h"
40 #include "mp_image.h"
41 #include "vf.h"
42 #include "libvo/fastmemcpy.h"
43
44 #define XMIN(a,b) ((a) < (b) ? (a) : (b))
45 #define XMAX(a,b) ((a) > (b) ? (a) : (b))
46
47 typedef short DCTELEM;
48
49 //===========================================================================//
50 static const uint8_t __attribute__((aligned(8))) dither[8][8]={
51 { 0, 48, 12, 60, 3, 51, 15, 63, },
52 { 32, 16, 44, 28, 35, 19, 47, 31, },
53 { 8, 56, 4, 52, 11, 59, 7, 55, },
54 { 40, 24, 36, 20, 43, 27, 39, 23, },
55 { 2, 50, 14, 62, 1, 49, 13, 61, },
56 { 34, 18, 46, 30, 33, 17, 45, 29, },
57 { 10, 58, 6, 54, 9, 57, 5, 53, },
58 { 42, 26, 38, 22, 41, 25, 37, 21, },
59 };
60
61 struct vf_priv_s {
62 int qp;
63 int mode;
64 int mpeg2;
65 int temp_stride;
66 uint8_t *src;
67 };
68 #if 0
69 static inline void dct7_c(DCTELEM *dst, int s0, int s1, int s2, int s3, int step){
70 int s, d;
71 int dst2[64];
72 //#define S0 (1024/0.37796447300922719759)
73 #define C0 ((int)(1024*0.37796447300922719759+0.5)) //sqrt(1/7)
74 #define C1 ((int)(1024*0.53452248382484879308/6+0.5)) //sqrt(2/7)/6
75
76 #define C2 ((int)(1024*0.45221175985034745004/2+0.5))
77 #define C3 ((int)(1024*0.36264567479870879474/2+0.5))
78
79 //0.1962505182412941918 0.0149276808419397944-0.2111781990832339584
80 #define C4 ((int)(1024*0.1962505182412941918+0.5))
81 #define C5 ((int)(1024*0.0149276808419397944+0.5))
82 //#define C6 ((int)(1024*0.2111781990832339584+0.5))
83 #if 0
84 s= s0 + s1 + s2;
85 dst[0*step] = ((s + s3)*C0 + 512) >> 10;
86 s= (s - 6*s3)*C1 + 512;
87 d= (s0-s2)*C4 + (s1-s2)*C5;
88 dst[1*step] = (s + 2*d)>>10;
89 s -= d;
90 d= (s1-s0)*C2 + (s1-s2)*C3;
91 dst[2*step] = (s + d)>>10;
92 dst[3*step] = (s - d)>>10;
93 #elif 1
94 s = s3+s3;
95 s3= s-s0;
96 s0= s+s0;
97 s = s2+s1;
98 s2= s2-s1;
99 dst[0*step]= s0 + s;
100 dst[2*step]= s0 - s;
101 dst[1*step]= 2*s3 + s2;
102 dst[3*step]= s3 - 2*s2;
103 #else
104 int i,j,n=7;
105 for(i=0; i<7; i+=2){
106 dst2[i*step/2]= 0;
107 for(j=0; j<4; j++)
108 dst2[i*step/2] += src[j*step] * cos(i*M_PI/n*(j+0.5)) * sqrt((i?2.0:1.0)/n);
109 if(fabs(dst2[i*step/2] - dst[i*step/2]) > 20)
110 printf("%d %d %d (%d %d %d %d) -> (%d %d %d %d)\n", i,dst2[i*step/2], dst[i*step/2],src[0*step], src[1*step], src[2*step], src[3*step], dst[0*step], dst[1*step],dst[2*step],dst[3*step]);
111 }
112 #endif
113 }
114 #endif
115
116 static inline void dctA_c(DCTELEM *dst, uint8_t *src, int stride){
117 int i;
118
119 for(i=0; i<4; i++){
120 int s0= src[0*stride] + src[6*stride];
121 int s1= src[1*stride] + src[5*stride];
122 int s2= src[2*stride] + src[4*stride];
123 int s3= src[3*stride];
124 int s= s3+s3;
125 s3= s-s0;
126 s0= s+s0;
127 s = s2+s1;
128 s2= s2-s1;
129 dst[0]= s0 + s;
130 dst[2]= s0 - s;
131 dst[1]= 2*s3 + s2;
132 dst[3]= s3 - 2*s2;
133 src++;
134 dst+=4;
135 }
136 }
137
138 static void dctB_c(DCTELEM *dst, DCTELEM *src){
139 int i;
140
141 for(i=0; i<4; i++){
142 int s0= src[0*4] + src[6*4];
143 int s1= src[1*4] + src[5*4];
144 int s2= src[2*4] + src[4*4];
145 int s3= src[3*4];
146 int s= s3+s3;
147 s3= s-s0;
148 s0= s+s0;
149 s = s2+s1;
150 s2= s2-s1;
151 dst[0*4]= s0 + s;
152 dst[2*4]= s0 - s;
153 dst[1*4]= 2*s3 + s2;
154 dst[3*4]= s3 - 2*s2;
155 src++;
156 dst++;
157 }
158 }
159
160 #if HAVE_MMX
161 static void dctB_mmx(DCTELEM *dst, DCTELEM *src){
162 __asm__ volatile (
163 "movq (%0), %%mm0 \n\t"
164 "movq 1*4*2(%0), %%mm1 \n\t"
165 "paddw 6*4*2(%0), %%mm0 \n\t"
166 "paddw 5*4*2(%0), %%mm1 \n\t"
167 "movq 2*4*2(%0), %%mm2 \n\t"
168 "movq 3*4*2(%0), %%mm3 \n\t"
169 "paddw 4*4*2(%0), %%mm2 \n\t"
170 "paddw %%mm3, %%mm3 \n\t" //s
171 "movq %%mm3, %%mm4 \n\t" //s
172 "psubw %%mm0, %%mm3 \n\t" //s-s0
173 "paddw %%mm0, %%mm4 \n\t" //s+s0
174 "movq %%mm2, %%mm0 \n\t" //s2
175 "psubw %%mm1, %%mm2 \n\t" //s2-s1
176 "paddw %%mm1, %%mm0 \n\t" //s2+s1
177 "movq %%mm4, %%mm1 \n\t" //s0'
178 "psubw %%mm0, %%mm4 \n\t" //s0'-s'
179 "paddw %%mm0, %%mm1 \n\t" //s0'+s'
180 "movq %%mm3, %%mm0 \n\t" //s3'
181 "psubw %%mm2, %%mm3 \n\t"
182 "psubw %%mm2, %%mm3 \n\t"
183 "paddw %%mm0, %%mm2 \n\t"
184 "paddw %%mm0, %%mm2 \n\t"
185 "movq %%mm1, (%1) \n\t"
186 "movq %%mm4, 2*4*2(%1) \n\t"
187 "movq %%mm2, 1*4*2(%1) \n\t"
188 "movq %%mm3, 3*4*2(%1) \n\t"
189 :: "r" (src), "r"(dst)
190 );
191 }
192 #endif
193
194 static void (*dctB)(DCTELEM *dst, DCTELEM *src)= dctB_c;
195
196 #define N0 4
197 #define N1 5
198 #define N2 10
199 #define SN0 2
200 #define SN1 2.2360679775
201 #define SN2 3.16227766017
202 #define N (1<<16)
203
204 static const int factor[16]={
205 N/(N0*N0), N/(N0*N1), N/(N0*N0),N/(N0*N2),
206 N/(N1*N0), N/(N1*N1), N/(N1*N0),N/(N1*N2),
207 N/(N0*N0), N/(N0*N1), N/(N0*N0),N/(N0*N2),
208 N/(N2*N0), N/(N2*N1), N/(N2*N0),N/(N2*N2),
209 };
210
211 static const int thres[16]={
212 N/(SN0*SN0), N/(SN0*SN2), N/(SN0*SN0),N/(SN0*SN2),
213 N/(SN2*SN0), N/(SN2*SN2), N/(SN2*SN0),N/(SN2*SN2),
214 N/(SN0*SN0), N/(SN0*SN2), N/(SN0*SN0),N/(SN0*SN2),
215 N/(SN2*SN0), N/(SN2*SN2), N/(SN2*SN0),N/(SN2*SN2),
216 };
217
218 static int thres2[99][16];
219
220 static void init_thres2(void){
221 int qp, i;
222 int bias= 0; //FIXME
223
224 for(qp=0; qp<99; qp++){
225 for(i=0; i<16; i++){
226 thres2[qp][i]= ((i&1)?SN2:SN0) * ((i&4)?SN2:SN0) * XMAX(1,qp) * (1<<2) - 1 - bias;
227 }
228 }
229 }
230
231 static int hardthresh_c(DCTELEM *src, int qp){
232 int i;
233 int a;
234
235 a= src[0] * factor[0];
236 for(i=1; i<16; i++){
237 unsigned int threshold1= thres2[qp][i];
238 unsigned int threshold2= (threshold1<<1);
239 int level= src[i];
240 if(((unsigned)(level+threshold1))>threshold2){
241 a += level * factor[i];
242 }
243 }
244 return (a + (1<<11))>>12;
245 }
246
247 static int mediumthresh_c(DCTELEM *src, int qp){
248 int i;
249 int a;
250
251 a= src[0] * factor[0];
252 for(i=1; i<16; i++){
253 unsigned int threshold1= thres2[qp][i];
254 unsigned int threshold2= (threshold1<<1);
255 int level= src[i];
256 if(((unsigned)(level+threshold1))>threshold2){
257 if(((unsigned)(level+2*threshold1))>2*threshold2){
258 a += level * factor[i];
259 }else{
260 if(level>0) a+= 2*(level - (int)threshold1)*factor[i];
261 else a+= 2*(level + (int)threshold1)*factor[i];
262 }
263 }
264 }
265 return (a + (1<<11))>>12;
266 }
267
268 static int softthresh_c(DCTELEM *src, int qp){
269 int i;
270 int a;
271
272 a= src[0] * factor[0];
273 for(i=1; i<16; i++){
274 unsigned int threshold1= thres2[qp][i];
275 unsigned int threshold2= (threshold1<<1);
276 int level= src[i];
277 if(((unsigned)(level+threshold1))>threshold2){
278 if(level>0) a+= (level - (int)threshold1)*factor[i];
279 else a+= (level + (int)threshold1)*factor[i];
280 }
281 }
282 return (a + (1<<11))>>12;
283 }
284
285 static int (*requantize)(DCTELEM *src, int qp)= hardthresh_c;
286
287 static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, uint8_t *qp_store, int qp_stride, int is_luma){
288 int x, y;
289 const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15));
290 uint8_t *p_src= p->src + 8*stride;
291 DCTELEM *block= (DCTELEM *)p->src;
292 DCTELEM *temp= (DCTELEM *)(p->src + 32);
293
294 if (!src || !dst) return; // HACK avoid crash for Y8 colourspace
295 for(y=0; y<height; y++){
296 int index= 8 + 8*stride + y*stride;
297 fast_memcpy(p_src + index, src + y*src_stride, width);
298 for(x=0; x<8; x++){
299 p_src[index - x - 1]= p_src[index + x ];
300 p_src[index + width + x ]= p_src[index + width - x - 1];
301 }
302 }
303 for(y=0; y<8; y++){
304 fast_memcpy(p_src + ( 7-y)*stride, p_src + ( y+8)*stride, stride);
305 fast_memcpy(p_src + (height+8+y)*stride, p_src + (height-y+7)*stride, stride);
306 }
307 //FIXME (try edge emu)
308
309 for(y=0; y<height; y++){
310 for(x=-8; x<0; x+=4){
311 const int index= x + y*stride + (8-3)*(1+stride) + 8; //FIXME silly offset
312 uint8_t *src = p_src + index;
313 DCTELEM *tp= temp+4*x;
314
315 dctA_c(tp+4*8, src, stride);
316 }
317 for(x=0; x<width; ){
318 const int qps= 3 + is_luma;
319 int qp;
320 int end= XMIN(x+8, width);
321
322 if(p->qp)
323 qp= p->qp;
324 else{
325 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride];
326 qp=norm_qscale(qp, p->mpeg2);
327 }
328 for(; x<end; x++){
329 const int index= x + y*stride + (8-3)*(1+stride) + 8; //FIXME silly offset
330 uint8_t *src = p_src + index;
331 DCTELEM *tp= temp+4*x;
332 int v;
333
334 if((x&3)==0)
335 dctA_c(tp+4*8, src, stride);
336
337 dctB(block, tp);
338
339 v= requantize(block, qp);
340 v= (v + dither[y&7][x&7])>>6;
341 if((unsigned)v > 255)
342 v= (-v)>>31;
343 dst[x + y*dst_stride]= v;
344 }
345 }
346 }
347 }
348
349 static int config(struct vf_instance *vf,
350 int width, int height, int d_width, int d_height,
351 unsigned int flags, unsigned int outfmt){
352 int h= (height+16+15)&(~15);
353
354 vf->priv->temp_stride= (width+16+15)&(~15);
355 vf->priv->src = av_malloc(vf->priv->temp_stride*(h+8)*sizeof(uint8_t));
356
357 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
358 }
359
360 static void get_image(struct vf_instance *vf, mp_image_t *mpi){
361 if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
362 // ok, we can do pp in-place (or pp disabled):
363 vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
364 mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
365 mpi->planes[0]=vf->dmpi->planes[0];
366 mpi->stride[0]=vf->dmpi->stride[0];
367 mpi->width=vf->dmpi->width;
368 if(mpi->flags&MP_IMGFLAG_PLANAR){
369 mpi->planes[1]=vf->dmpi->planes[1];
370 mpi->planes[2]=vf->dmpi->planes[2];
371 mpi->stride[1]=vf->dmpi->stride[1];
372 mpi->stride[2]=vf->dmpi->stride[2];
373 }
374 mpi->flags|=MP_IMGFLAG_DIRECT;
375 }
376
377 static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts){
378 mp_image_t *dmpi;
379
380 if(mpi->flags&MP_IMGFLAG_DIRECT){
381 dmpi=vf->dmpi;
382 }else{
383 // no DR, so get a new image! hope we'll get DR buffer:
384 dmpi=vf_get_image(vf->next,mpi->imgfmt,
385 MP_IMGTYPE_TEMP,
386 MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
387 mpi->width,mpi->height);
388 vf_clone_mpi_attributes(dmpi, mpi);
389 }
390
391 vf->priv->mpeg2= mpi->qscale_type;
392 if(mpi->qscale || vf->priv->qp){
393 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, mpi->qscale, mpi->qstride, 1);
394 filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, mpi->qscale, mpi->qstride, 0);
395 filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, mpi->qscale, mpi->qstride, 0);
396 }else{
397 memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]);
398 memcpy_pic(dmpi->planes[1], mpi->planes[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[1], mpi->stride[1]);
399 memcpy_pic(dmpi->planes[2], mpi->planes[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[2], mpi->stride[2]);
400 }
401
402 #if HAVE_MMX
403 if(gCpuCaps.hasMMX) __asm__ volatile ("emms\n\t");
404 #endif
405 #if HAVE_MMX2
406 if(gCpuCaps.hasMMX2) __asm__ volatile ("sfence\n\t");
407 #endif
408
409 return vf_next_put_image(vf,dmpi, pts);
410 }
411
412 static void uninit(struct vf_instance *vf){
413 if(!vf->priv) return;
414
415 av_free(vf->priv->src);
416 vf->priv->src= NULL;
417
418 free(vf->priv);
419 vf->priv=NULL;
420 }
421
422 //===========================================================================//
423 static int query_format(struct vf_instance *vf, unsigned int fmt){
424 switch(fmt){
425 case IMGFMT_YVU9:
426 case IMGFMT_IF09:
427 case IMGFMT_YV12:
428 case IMGFMT_I420:
429 case IMGFMT_IYUV:
430 case IMGFMT_CLPL:
431 case IMGFMT_Y800:
432 case IMGFMT_Y8:
433 case IMGFMT_444P:
434 case IMGFMT_422P:
435 case IMGFMT_411P:
436 return vf_next_query_format(vf,fmt);
437 }
438 return 0;
439 }
440
441 static int control(struct vf_instance *vf, int request, void* data){
442 return vf_next_control(vf,request,data);
443 }
444
445 static int vf_open(vf_instance_t *vf, char *args){
446 vf->config=config;
447 vf->put_image=put_image;
448 vf->get_image=get_image;
449 vf->query_format=query_format;
450 vf->uninit=uninit;
451 vf->control= control;
452 vf->priv=malloc(sizeof(struct vf_priv_s));
453 memset(vf->priv, 0, sizeof(struct vf_priv_s));
454
455 if (args) sscanf(args, "%d:%d", &vf->priv->qp, &vf->priv->mode);
456
457 if(vf->priv->qp < 0)
458 vf->priv->qp = 0;
459
460 init_thres2();
461
462 switch(vf->priv->mode){
463 case 0: requantize= hardthresh_c; break;
464 case 1: requantize= softthresh_c; break;
465 default:
466 case 2: requantize= mediumthresh_c; break;
467 }
468
469 #if HAVE_MMX
470 if(gCpuCaps.hasMMX){
471 dctB= dctB_mmx;
472 }
473 #endif
474 #if 0
475 if(gCpuCaps.hasMMX){
476 switch(vf->priv->mode){
477 case 0: requantize= hardthresh_mmx; break;
478 case 1: requantize= softthresh_mmx; break;
479 }
480 }
481 #endif
482
483 return 1;
484 }
485
486 const vf_info_t vf_info_pp7 = {
487 "postprocess 7",
488 "pp7",
489 "Michael Niedermayer",
490 "",
491 vf_open,
492 NULL
493 };
MPlayer development branch