libmpcodecs/vf_noise.c

   1 /*
   2  * Copyright (C) 2002 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 #include <stdio.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <inttypes.h>
  25 #include <math.h>
  26
  27 #include "config.h"
  28 #include "mp_msg.h"
  29 #include "cpudetect.h"
  30
  31 #if HAVE_MALLOC_H
  32 #include <malloc.h>
  33 #endif
  34
  35 #include "img_format.h"
  36 #include "mp_image.h"
  37 #include "vf.h"
  38 #include "libvo/fastmemcpy.h"
  39
  40 #define MAX_NOISE 4096
  41 #define MAX_SHIFT 1024
  42 #define MAX_RES (MAX_NOISE-MAX_SHIFT)
  43
  44 //===========================================================================//
  45
  46 static inline void lineNoise_C(uint8_t *dst, uint8_t *src, int8_t *noise, int len, int shift);
  47 static inline void lineNoiseAvg_C(uint8_t *dst, uint8_t *src, int len, int8_t **shift);
  48
  49 static void (*lineNoise)(uint8_t *dst, uint8_t *src, int8_t *noise, int len, int shift)= lineNoise_C;
  50 static void (*lineNoiseAvg)(uint8_t *dst, uint8_t *src, int len, int8_t **shift)= lineNoiseAvg_C;
  51
  52 typedef struct FilterParam{
  53         int strength;
  54         int uniform;
  55         int temporal;
  56         int quality;
  57         int averaged;
  58         int pattern;
  59         int shiftptr;
  60         int8_t *noise;
  61         int8_t *prev_shift[MAX_RES][3];
  62 }FilterParam;
  63
  64 struct vf_priv_s {
  65         FilterParam lumaParam;
  66         FilterParam chromaParam;
  67         unsigned int outfmt;
  68 };
  69
  70 static int nonTempRandShift_init;
  71 static int nonTempRandShift[MAX_RES];
  72
  73 static int patt[4] = {
  74     -1,0,1,0
  75 };
  76
  77 #define RAND_N(range) ((int) ((double)range*rand()/(RAND_MAX+1.0)))
  78 static int8_t *initNoise(FilterParam *fp){
  79         int strength= fp->strength;
  80         int uniform= fp->uniform;
  81         int averaged= fp->averaged;
  82         int pattern= fp->pattern;
  83         int8_t *noise= memalign(16, MAX_NOISE*sizeof(int8_t));
  84         int i, j;
  85
  86         srand(123457);
  87
  88         for(i=0,j=0; i<MAX_NOISE; i++,j++)
  89         {
  90                 if(uniform) {
  91                         if (averaged) {
  92                                 if (pattern) {
  93                                         noise[i]= (RAND_N(strength) - strength/2)/6
  94                                                 +patt[j%4]*strength*0.25/3;
  95                                 } else {
  96                                         noise[i]= (RAND_N(strength) - strength/2)/3;
  97                                 }
  98                         } else {
  99                                 if (pattern) {
 100                                     noise[i]= (RAND_N(strength) - strength/2)/2
 101                                             + patt[j%4]*strength*0.25;
 102                                 } else {
 103                                         noise[i]= RAND_N(strength) - strength/2;
 104                                 }
 105                         }
 106                 } else {
 107                         double x1, x2, w, y1;
 108                         do {
 109                                 x1 = 2.0 * rand()/(float)RAND_MAX - 1.0;
 110                                 x2 = 2.0 * rand()/(float)RAND_MAX - 1.0;
 111                                 w = x1 * x1 + x2 * x2;
 112                         } while ( w >= 1.0 );
 113
 114                         w = sqrt( (-2.0 * log( w ) ) / w );
 115                         y1= x1 * w;
 116                         y1*= strength / sqrt(3.0);
 117                         if (pattern) {
 118                             y1 /= 2;
 119                             y1 += patt[j%4]*strength*0.35;
 120                         }
 121                         if     (y1<-128) y1=-128;
 122                         else if(y1> 127) y1= 127;
 123                         if (averaged) y1 /= 3.0;
 124                         noise[i]= (int)y1;
 125                 }
 126                 if (RAND_N(6) == 0) j--;
 127         }
 128
 129
 130         for (i = 0; i < MAX_RES; i++)
 131             for (j = 0; j < 3; j++)
 132                 fp->prev_shift[i][j] = noise + (rand()&(MAX_SHIFT-1));
 133
 134         if(!nonTempRandShift_init){
 135                 for(i=0; i<MAX_RES; i++){
 136                         nonTempRandShift[i]= rand()&(MAX_SHIFT-1);
 137                 }
 138                 nonTempRandShift_init = 1;
 139         }
 140
 141         fp->noise= noise;
 142         fp->shiftptr= 0;
 143         return noise;
 144 }
 145
 146 /***************************************************************************/
 147
 148 #if HAVE_MMX
 149 static inline void lineNoise_MMX(uint8_t *dst, uint8_t *src, int8_t *noise, int len, int shift){
 150         x86_reg mmx_len= len&(~7);
 151         noise+=shift;
 152
 153         __asm__ volatile(
 154                 "mov %3, %%"REG_a"              \n\t"
 155                 "pcmpeqb %%mm7, %%mm7           \n\t"
 156                 "psllw $15, %%mm7               \n\t"
 157                 "packsswb %%mm7, %%mm7          \n\t"
 158                 ASMALIGN(4)
 159                 "1:                             \n\t"
 160                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
 161                 "movq (%1, %%"REG_a"), %%mm1    \n\t"
 162                 "pxor %%mm7, %%mm0              \n\t"
 163                 "paddsb %%mm1, %%mm0            \n\t"
 164                 "pxor %%mm7, %%mm0              \n\t"
 165                 "movq %%mm0, (%2, %%"REG_a")    \n\t"
 166                 "add $8, %%"REG_a"              \n\t"
 167                 " js 1b                         \n\t"
 168                 :: "r" (src+mmx_len), "r" (noise+mmx_len), "r" (dst+mmx_len), "g" (-mmx_len)
 169                 : "%"REG_a
 170         );
 171         if(mmx_len!=len)
 172                 lineNoise_C(dst+mmx_len, src+mmx_len, noise+mmx_len, len-mmx_len, 0);
 173 }
 174 #endif
 175
 176 //duplicate of previous except movntq
 177 #if HAVE_MMX2
 178 static inline void lineNoise_MMX2(uint8_t *dst, uint8_t *src, int8_t *noise, int len, int shift){
 179         x86_reg mmx_len= len&(~7);
 180         noise+=shift;
 181
 182         __asm__ volatile(
 183                 "mov %3, %%"REG_a"              \n\t"
 184                 "pcmpeqb %%mm7, %%mm7           \n\t"
 185                 "psllw $15, %%mm7               \n\t"
 186                 "packsswb %%mm7, %%mm7          \n\t"
 187                 ASMALIGN(4)
 188                 "1:                             \n\t"
 189                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
 190                 "movq (%1, %%"REG_a"), %%mm1    \n\t"
 191                 "pxor %%mm7, %%mm0              \n\t"
 192                 "paddsb %%mm1, %%mm0            \n\t"
 193                 "pxor %%mm7, %%mm0              \n\t"
 194                 "movntq %%mm0, (%2, %%"REG_a")  \n\t"
 195                 "add $8, %%"REG_a"              \n\t"
 196                 " js 1b                         \n\t"
 197                 :: "r" (src+mmx_len), "r" (noise+mmx_len), "r" (dst+mmx_len), "g" (-mmx_len)
 198                 : "%"REG_a
 199         );
 200         if(mmx_len!=len)
 201                 lineNoise_C(dst+mmx_len, src+mmx_len, noise+mmx_len, len-mmx_len, 0);
 202 }
 203 #endif
 204
 205 static inline void lineNoise_C(uint8_t *dst, uint8_t *src, int8_t *noise, int len, int shift){
 206         int i;
 207         noise+= shift;
 208         for(i=0; i<len; i++)
 209         {
 210                 int v= src[i]+ noise[i];
 211                 if(v>255)       dst[i]=255; //FIXME optimize
 212                 else if(v<0)    dst[i]=0;
 213                 else            dst[i]=v;
 214         }
 215 }
 216
 217 /***************************************************************************/
 218
 219 #if HAVE_MMX
 220 static inline void lineNoiseAvg_MMX(uint8_t *dst, uint8_t *src, int len, int8_t **shift){
 221         x86_reg mmx_len= len&(~7);
 222
 223         __asm__ volatile(
 224                 "mov %5, %%"REG_a"              \n\t"
 225                 ASMALIGN(4)
 226                 "1:                             \n\t"
 227                 "movq (%1, %%"REG_a"), %%mm1    \n\t"
 228                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
 229                 "paddb (%2, %%"REG_a"), %%mm1   \n\t"
 230                 "paddb (%3, %%"REG_a"), %%mm1   \n\t"
 231                 "movq %%mm0, %%mm2              \n\t"
 232                 "movq %%mm1, %%mm3              \n\t"
 233                 "punpcklbw %%mm0, %%mm0         \n\t"
 234                 "punpckhbw %%mm2, %%mm2         \n\t"
 235                 "punpcklbw %%mm1, %%mm1         \n\t"
 236                 "punpckhbw %%mm3, %%mm3         \n\t"
 237                 "pmulhw %%mm0, %%mm1            \n\t"
 238                 "pmulhw %%mm2, %%mm3            \n\t"
 239                 "paddw %%mm1, %%mm1             \n\t"
 240                 "paddw %%mm3, %%mm3             \n\t"
 241                 "paddw %%mm0, %%mm1             \n\t"
 242                 "paddw %%mm2, %%mm3             \n\t"
 243                 "psrlw $8, %%mm1                \n\t"
 244                 "psrlw $8, %%mm3                \n\t"
 245                 "packuswb %%mm3, %%mm1          \n\t"
 246                 "movq %%mm1, (%4, %%"REG_a")    \n\t"
 247                 "add $8, %%"REG_a"              \n\t"
 248                 " js 1b                         \n\t"
 249                 :: "r" (src+mmx_len), "r" (shift[0]+mmx_len), "r" (shift[1]+mmx_len), "r" (shift[2]+mmx_len),
 250                    "r" (dst+mmx_len), "g" (-mmx_len)
 251                 : "%"REG_a
 252         );
 253
 254         if(mmx_len!=len){
 255                 int8_t *shift2[3]={shift[0]+mmx_len, shift[1]+mmx_len, shift[2]+mmx_len};
 256                 lineNoiseAvg_C(dst+mmx_len, src+mmx_len, len-mmx_len, shift2);
 257         }
 258 }
 259 #endif
 260
 261 static inline void lineNoiseAvg_C(uint8_t *dst, uint8_t *src, int len, int8_t **shift){
 262         int i;
 263         int8_t *src2= (int8_t*)src;
 264
 265         for(i=0; i<len; i++)
 266         {
 267             const int n= shift[0][i] + shift[1][i] + shift[2][i];
 268             dst[i]= src2[i]+((n*src2[i])>>7);
 269         }
 270 }
 271
 272 /***************************************************************************/
 273
 274 static void noise(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int width, int height, FilterParam *fp){
 275         int8_t *noise= fp->noise;
 276         int y;
 277         int shift=0;
 278
 279         if(!noise)
 280         {
 281                 if(src==dst) return;
 282
 283                 if(dstStride==srcStride) fast_memcpy(dst, src, srcStride*height);
 284                 else
 285                 {
 286                         for(y=0; y<height; y++)
 287                         {
 288                                 fast_memcpy(dst, src, width);
 289                                 dst+= dstStride;
 290                                 src+= srcStride;
 291                         }
 292                 }
 293                 return;
 294         }
 295
 296         for(y=0; y<height; y++)
 297         {
 298                 if(fp->temporal)        shift=  rand()&(MAX_SHIFT  -1);
 299                 else                    shift= nonTempRandShift[y];
 300
 301                 if(fp->quality==0) shift&= ~7;
 302                 if (fp->averaged) {
 303                     lineNoiseAvg(dst, src, width, fp->prev_shift[y]);
 304                     fp->prev_shift[y][fp->shiftptr] = noise + shift;
 305                 } else {
 306                     lineNoise(dst, src, noise, width, shift);
 307                 }
 308                 dst+= dstStride;
 309                 src+= srcStride;
 310         }
 311         fp->shiftptr++;
 312         if (fp->shiftptr == 3) fp->shiftptr = 0;
 313 }
 314
 315 static int config(struct vf_instance_s* vf,
 316         int width, int height, int d_width, int d_height,
 317         unsigned int flags, unsigned int outfmt){
 318
 319         return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
 320 }
 321
 322 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){
 323     if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
 324     if(mpi->imgfmt!=vf->priv->outfmt) return; // colorspace differ
 325     // ok, we can do pp in-place (or pp disabled):
 326     vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
 327         mpi->type, mpi->flags, mpi->w, mpi->h);
 328     mpi->planes[0]=vf->dmpi->planes[0];
 329     mpi->stride[0]=vf->dmpi->stride[0];
 330     mpi->width=vf->dmpi->width;
 331     if(mpi->flags&MP_IMGFLAG_PLANAR){
 332         mpi->planes[1]=vf->dmpi->planes[1];
 333         mpi->planes[2]=vf->dmpi->planes[2];
 334         mpi->stride[1]=vf->dmpi->stride[1];
 335         mpi->stride[2]=vf->dmpi->stride[2];
 336     }
 337     mpi->flags|=MP_IMGFLAG_DIRECT;
 338 }
 339
 340 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts){
 341         mp_image_t *dmpi;
 342
 343         if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
 344                 // no DR, so get a new image! hope we'll get DR buffer:
 345                 vf->dmpi=vf_get_image(vf->next,vf->priv->outfmt,
 346                 MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE,
 347                 mpi->w,mpi->h);
 348 //printf("nodr\n");
 349         }
 350 //else printf("dr\n");
 351         dmpi= vf->dmpi;
 352
 353         noise(dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, &vf->priv->lumaParam);
 354         noise(dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w/2, mpi->h/2, &vf->priv->chromaParam);
 355         noise(dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w/2, mpi->h/2, &vf->priv->chromaParam);
 356
 357         vf_clone_mpi_attributes(dmpi, mpi);
 358
 359 #if HAVE_MMX
 360         if(gCpuCaps.hasMMX) __asm__ volatile ("emms\n\t");
 361 #endif
 362 #if HAVE_MMX2
 363         if(gCpuCaps.hasMMX2) __asm__ volatile ("sfence\n\t");
 364 #endif
 365
 366         return vf_next_put_image(vf,dmpi, pts);
 367 }
 368
 369 static void uninit(struct vf_instance_s* vf){
 370         if(!vf->priv) return;
 371
 372         if(vf->priv->chromaParam.noise) free(vf->priv->chromaParam.noise);
 373         vf->priv->chromaParam.noise= NULL;
 374
 375         if(vf->priv->lumaParam.noise) free(vf->priv->lumaParam.noise);
 376         vf->priv->lumaParam.noise= NULL;
 377
 378         free(vf->priv);
 379         vf->priv=NULL;
 380 }
 381
 382 //===========================================================================//
 383
 384 static int query_format(struct vf_instance_s* vf, unsigned int fmt){
 385         switch(fmt)
 386         {
 387         case IMGFMT_YV12:
 388         case IMGFMT_I420:
 389         case IMGFMT_IYUV:
 390                 return vf_next_query_format(vf,vf->priv->outfmt);
 391         }
 392         return 0;
 393 }
 394
 395 static void parse(FilterParam *fp, char* args){
 396         char *pos;
 397         char *max= strchr(args, ':');
 398
 399         if(!max) max= args + strlen(args);
 400
 401         fp->strength= atoi(args);
 402         pos= strchr(args, 'u');
 403         if(pos && pos<max) fp->uniform=1;
 404         pos= strchr(args, 't');
 405         if(pos && pos<max) fp->temporal=1;
 406         pos= strchr(args, 'h');
 407         if(pos && pos<max) fp->quality=1;
 408         pos= strchr(args, 'p');
 409         if(pos && pos<max) fp->pattern=1;
 410         pos= strchr(args, 'a');
 411         if(pos && pos<max) {
 412             fp->temporal=1;
 413             fp->averaged=1;
 414         }
 415
 416         if(fp->strength) initNoise(fp);
 417 }
 418
 419 static unsigned int fmt_list[]={
 420     IMGFMT_YV12,
 421     IMGFMT_I420,
 422     IMGFMT_IYUV,
 423     0
 424 };
 425
 426 static int open(vf_instance_t *vf, char* args){
 427     vf->config=config;
 428     vf->put_image=put_image;
 429     vf->get_image=get_image;
 430     vf->query_format=query_format;
 431     vf->uninit=uninit;
 432     vf->priv=malloc(sizeof(struct vf_priv_s));
 433     memset(vf->priv, 0, sizeof(struct vf_priv_s));
 434     if(args)
 435     {
 436         char *arg2= strchr(args,':');
 437         if(arg2) parse(&vf->priv->chromaParam, arg2+1);
 438         parse(&vf->priv->lumaParam, args);
 439     }
 440
 441     // check csp:
 442     vf->priv->outfmt=vf_match_csp(&vf->next,fmt_list,IMGFMT_YV12);
 443     if(!vf->priv->outfmt)
 444     {
 445         uninit(vf);
 446         return 0; // no csp match :(
 447     }
 448
 449
 450 #if HAVE_MMX
 451     if(gCpuCaps.hasMMX){
 452         lineNoise= lineNoise_MMX;
 453         lineNoiseAvg= lineNoiseAvg_MMX;
 454     }
 455 #endif
 456 #if HAVE_MMX2
 457     if(gCpuCaps.hasMMX2) lineNoise= lineNoise_MMX2;
 458 //    if(gCpuCaps.hasMMX) lineNoiseAvg= lineNoiseAvg_MMX2;
 459 #endif
 460
 461     return 1;
 462 }
 463
 464 const vf_info_t vf_info_noise = {
 465     "noise generator",
 466     "noise",
 467     "Michael Niedermayer",
 468     "",
 469     open,
 470     NULL
 471 };
 472
 473 //===========================================================================//