libmpcodecs/vf_pp7.c

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