libmpcodecs/vf_pp7.c

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