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