libmpcodecs/vf_spp.c

   1 /*
   2  * Copyright (C) 2003 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  * This implementation is based on an algorithm described in
  23  * "Aria Nosratinia Embedded Post-Processing for
  24  * Enhancement of Compressed Images (1999)"
  25  * (http://citeseer.nj.nec.com/nosratinia99embedded.html)
  26  */
  27
  28
  29 #include <stdio.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <inttypes.h>
  33 #include <math.h>
  34
  35 #include "config.h"
  36
  37 #include "mp_msg.h"
  38 #include "cpudetect.h"
  39
  40 #include "libavutil/internal.h"
  41 #include "libavutil/intreadwrite.h"
  42 #include "libavcodec/avcodec.h"
  43 #include "libavcodec/dsputil.h"
  44
  45 #undef fprintf
  46 #undef free
  47 #undef malloc
  48
  49 #ifdef HAVE_MALLOC_H
  50 #include <malloc.h>
  51 #endif
  52
  53 #include "img_format.h"
  54 #include "mp_image.h"
  55 #include "vf.h"
  56 #include "libvo/fastmemcpy.h"
  57
  58 #define XMIN(a,b) ((a) < (b) ? (a) : (b))
  59
  60 //===========================================================================//
  61 static const uint8_t  __attribute__((aligned(8))) dither[8][8]={
  62 {  0,  48,  12,  60,   3,  51,  15,  63, },
  63 { 32,  16,  44,  28,  35,  19,  47,  31, },
  64 {  8,  56,   4,  52,  11,  59,   7,  55, },
  65 { 40,  24,  36,  20,  43,  27,  39,  23, },
  66 {  2,  50,  14,  62,   1,  49,  13,  61, },
  67 { 34,  18,  46,  30,  33,  17,  45,  29, },
  68 { 10,  58,   6,  54,   9,  57,   5,  53, },
  69 { 42,  26,  38,  22,  41,  25,  37,  21, },
  70 };
  71
  72 static const uint8_t offset[127][2]= {
  73 {0,0},
  74 {0,0}, {4,4},
  75 {0,0}, {2,2}, {6,4}, {4,6},
  76 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7},
  77
  78 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3},
  79 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
  80
  81 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7},
  82 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
  83 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
  84 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
  85
  86 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2},
  87 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
  88 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
  89 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
  90 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
  91 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
  92 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
  93 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
  94 };
  95
  96 struct vf_priv_s {
  97         int log2_count;
  98         int qp;
  99         int mode;
 100         int mpeg2;
 101         int temp_stride;
 102         uint8_t *src;
 103         int16_t *temp;
 104         AVCodecContext *avctx;
 105         DSPContext dsp;
 106         char *non_b_qp;
 107 };
 108
 109 #define SHIFT 22
 110
 111 static void hardthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
 112         int i;
 113         int bias= 0; //FIXME
 114         unsigned int threshold1, threshold2;
 115
 116         threshold1= qp*((1<<4) - bias) - 1;
 117         threshold2= (threshold1<<1);
 118
 119         memset(dst, 0, 64*sizeof(DCTELEM));
 120         dst[0]= (src[0] + 4)>>3;
 121
 122         for(i=1; i<64; i++){
 123                 int level= src[i];
 124                 if(((unsigned)(level+threshold1))>threshold2){
 125                         const int j= permutation[i];
 126                         dst[j]= (level + 4)>>3;
 127                 }
 128         }
 129 }
 130
 131 static void softthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
 132         int i;
 133         int bias= 0; //FIXME
 134         unsigned int threshold1, threshold2;
 135
 136         threshold1= qp*((1<<4) - bias) - 1;
 137         threshold2= (threshold1<<1);
 138
 139         memset(dst, 0, 64*sizeof(DCTELEM));
 140         dst[0]= (src[0] + 4)>>3;
 141
 142         for(i=1; i<64; i++){
 143                 int level= src[i];
 144                 if(((unsigned)(level+threshold1))>threshold2){
 145                         const int j= permutation[i];
 146                         if(level>0)
 147                                 dst[j]= (level - threshold1 + 4)>>3;
 148                         else
 149                                 dst[j]= (level + threshold1 + 4)>>3;
 150                 }
 151         }
 152 }
 153
 154 #if HAVE_MMX
 155 static void hardthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
 156         int bias= 0; //FIXME
 157         unsigned int threshold1;
 158
 159         threshold1= qp*((1<<4) - bias) - 1;
 160
 161         __asm__ volatile(
 162 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
 163                 "movq " #src0 ", %%mm0  \n\t"\
 164                 "movq " #src1 ", %%mm1  \n\t"\
 165                 "movq " #src2 ", %%mm2  \n\t"\
 166                 "movq " #src3 ", %%mm3  \n\t"\
 167                 "psubw %%mm4, %%mm0     \n\t"\
 168                 "psubw %%mm4, %%mm1     \n\t"\
 169                 "psubw %%mm4, %%mm2     \n\t"\
 170                 "psubw %%mm4, %%mm3     \n\t"\
 171                 "paddusw %%mm5, %%mm0   \n\t"\
 172                 "paddusw %%mm5, %%mm1   \n\t"\
 173                 "paddusw %%mm5, %%mm2   \n\t"\
 174                 "paddusw %%mm5, %%mm3   \n\t"\
 175                 "paddw %%mm6, %%mm0     \n\t"\
 176                 "paddw %%mm6, %%mm1     \n\t"\
 177                 "paddw %%mm6, %%mm2     \n\t"\
 178                 "paddw %%mm6, %%mm3     \n\t"\
 179                 "psubusw %%mm6, %%mm0   \n\t"\
 180                 "psubusw %%mm6, %%mm1   \n\t"\
 181                 "psubusw %%mm6, %%mm2   \n\t"\
 182                 "psubusw %%mm6, %%mm3   \n\t"\
 183                 "psraw $3, %%mm0        \n\t"\
 184                 "psraw $3, %%mm1        \n\t"\
 185                 "psraw $3, %%mm2        \n\t"\
 186                 "psraw $3, %%mm3        \n\t"\
 187 \
 188                 "movq %%mm0, %%mm7      \n\t"\
 189                 "punpcklwd %%mm2, %%mm0 \n\t" /*A*/\
 190                 "punpckhwd %%mm2, %%mm7 \n\t" /*C*/\
 191                 "movq %%mm1, %%mm2      \n\t"\
 192                 "punpcklwd %%mm3, %%mm1 \n\t" /*B*/\
 193                 "punpckhwd %%mm3, %%mm2 \n\t" /*D*/\
 194                 "movq %%mm0, %%mm3      \n\t"\
 195                 "punpcklwd %%mm1, %%mm0 \n\t" /*A*/\
 196                 "punpckhwd %%mm7, %%mm3 \n\t" /*C*/\
 197                 "punpcklwd %%mm2, %%mm7 \n\t" /*B*/\
 198                 "punpckhwd %%mm2, %%mm1 \n\t" /*D*/\
 199 \
 200                 "movq %%mm0, " #dst0 "  \n\t"\
 201                 "movq %%mm7, " #dst1 "  \n\t"\
 202                 "movq %%mm3, " #dst2 "  \n\t"\
 203                 "movq %%mm1, " #dst3 "  \n\t"
 204
 205                 "movd %2, %%mm4         \n\t"
 206                 "movd %3, %%mm5         \n\t"
 207                 "movd %4, %%mm6         \n\t"
 208                 "packssdw %%mm4, %%mm4  \n\t"
 209                 "packssdw %%mm5, %%mm5  \n\t"
 210                 "packssdw %%mm6, %%mm6  \n\t"
 211                 "packssdw %%mm4, %%mm4  \n\t"
 212                 "packssdw %%mm5, %%mm5  \n\t"
 213                 "packssdw %%mm6, %%mm6  \n\t"
 214                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
 215                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
 216                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
 217                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
 218                 : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed?
 219         );
 220         dst[0]= (src[0] + 4)>>3;
 221 }
 222
 223 static void softthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
 224         int bias= 0; //FIXME
 225         unsigned int threshold1;
 226
 227         threshold1= qp*((1<<4) - bias) - 1;
 228
 229         __asm__ volatile(
 230 #undef REQUANT_CORE
 231 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
 232                 "movq " #src0 ", %%mm0  \n\t"\
 233                 "movq " #src1 ", %%mm1  \n\t"\
 234                 "pxor %%mm6, %%mm6      \n\t"\
 235                 "pxor %%mm7, %%mm7      \n\t"\
 236                 "pcmpgtw %%mm0, %%mm6   \n\t"\
 237                 "pcmpgtw %%mm1, %%mm7   \n\t"\
 238                 "pxor %%mm6, %%mm0      \n\t"\
 239                 "pxor %%mm7, %%mm1      \n\t"\
 240                 "psubusw %%mm4, %%mm0   \n\t"\
 241                 "psubusw %%mm4, %%mm1   \n\t"\
 242                 "pxor %%mm6, %%mm0      \n\t"\
 243                 "pxor %%mm7, %%mm1      \n\t"\
 244                 "movq " #src2 ", %%mm2  \n\t"\
 245                 "movq " #src3 ", %%mm3  \n\t"\
 246                 "pxor %%mm6, %%mm6      \n\t"\
 247                 "pxor %%mm7, %%mm7      \n\t"\
 248                 "pcmpgtw %%mm2, %%mm6   \n\t"\
 249                 "pcmpgtw %%mm3, %%mm7   \n\t"\
 250                 "pxor %%mm6, %%mm2      \n\t"\
 251                 "pxor %%mm7, %%mm3      \n\t"\
 252                 "psubusw %%mm4, %%mm2   \n\t"\
 253                 "psubusw %%mm4, %%mm3   \n\t"\
 254                 "pxor %%mm6, %%mm2      \n\t"\
 255                 "pxor %%mm7, %%mm3      \n\t"\
 256 \
 257                 "paddsw %%mm5, %%mm0    \n\t"\
 258                 "paddsw %%mm5, %%mm1    \n\t"\
 259                 "paddsw %%mm5, %%mm2    \n\t"\
 260                 "paddsw %%mm5, %%mm3    \n\t"\
 261                 "psraw $3, %%mm0        \n\t"\
 262                 "psraw $3, %%mm1        \n\t"\
 263                 "psraw $3, %%mm2        \n\t"\
 264                 "psraw $3, %%mm3        \n\t"\
 265 \
 266                 "movq %%mm0, %%mm7      \n\t"\
 267                 "punpcklwd %%mm2, %%mm0 \n\t" /*A*/\
 268                 "punpckhwd %%mm2, %%mm7 \n\t" /*C*/\
 269                 "movq %%mm1, %%mm2      \n\t"\
 270                 "punpcklwd %%mm3, %%mm1 \n\t" /*B*/\
 271                 "punpckhwd %%mm3, %%mm2 \n\t" /*D*/\
 272                 "movq %%mm0, %%mm3      \n\t"\
 273                 "punpcklwd %%mm1, %%mm0 \n\t" /*A*/\
 274                 "punpckhwd %%mm7, %%mm3 \n\t" /*C*/\
 275                 "punpcklwd %%mm2, %%mm7 \n\t" /*B*/\
 276                 "punpckhwd %%mm2, %%mm1 \n\t" /*D*/\
 277 \
 278                 "movq %%mm0, " #dst0 "  \n\t"\
 279                 "movq %%mm7, " #dst1 "  \n\t"\
 280                 "movq %%mm3, " #dst2 "  \n\t"\
 281                 "movq %%mm1, " #dst3 "  \n\t"
 282
 283                 "movd %2, %%mm4         \n\t"
 284                 "movd %3, %%mm5         \n\t"
 285                 "packssdw %%mm4, %%mm4  \n\t"
 286                 "packssdw %%mm5, %%mm5  \n\t"
 287                 "packssdw %%mm4, %%mm4  \n\t"
 288                 "packssdw %%mm5, %%mm5  \n\t"
 289                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
 290                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
 291                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
 292                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
 293                 : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed?
 294         );
 295
 296         dst[0]= (src[0] + 4)>>3;
 297 }
 298 #endif
 299
 300 static inline void add_block(int16_t *dst, int stride, DCTELEM block[64]){
 301         int y;
 302
 303         for(y=0; y<8; y++){
 304                 *(uint32_t*)&dst[0 + y*stride]+= *(uint32_t*)&block[0 + y*8];
 305                 *(uint32_t*)&dst[2 + y*stride]+= *(uint32_t*)&block[2 + y*8];
 306                 *(uint32_t*)&dst[4 + y*stride]+= *(uint32_t*)&block[4 + y*8];
 307                 *(uint32_t*)&dst[6 + y*stride]+= *(uint32_t*)&block[6 + y*8];
 308         }
 309 }
 310
 311 static void store_slice_c(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
 312         int y, x;
 313
 314 #define STORE(pos) \
 315         temp= ((src[x + y*src_stride + pos]<<log2_scale) + d[pos])>>6;\
 316         if(temp & 0x100) temp= ~(temp>>31);\
 317         dst[x + y*dst_stride + pos]= temp;
 318
 319         for(y=0; y<height; y++){
 320                 const uint8_t *d= dither[y];
 321                 for(x=0; x<width; x+=8){
 322                         int temp;
 323                         STORE(0);
 324                         STORE(1);
 325                         STORE(2);
 326                         STORE(3);
 327                         STORE(4);
 328                         STORE(5);
 329                         STORE(6);
 330                         STORE(7);
 331                 }
 332         }
 333 }
 334
 335 #if HAVE_MMX
 336 static void store_slice_mmx(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
 337         int y;
 338
 339         for(y=0; y<height; y++){
 340                 uint8_t *dst1= dst;
 341                 int16_t *src1= src;
 342                 __asm__ volatile(
 343                         "movq (%3), %%mm3       \n\t"
 344                         "movq (%3), %%mm4       \n\t"
 345                         "movd %4, %%mm2         \n\t"
 346                         "pxor %%mm0, %%mm0      \n\t"
 347                         "punpcklbw %%mm0, %%mm3 \n\t"
 348                         "punpckhbw %%mm0, %%mm4 \n\t"
 349                         "psraw %%mm2, %%mm3     \n\t"
 350                         "psraw %%mm2, %%mm4     \n\t"
 351                         "movd %5, %%mm2         \n\t"
 352                         "1:                     \n\t"
 353                         "movq (%0), %%mm0       \n\t"
 354                         "movq 8(%0), %%mm1      \n\t"
 355                         "paddw %%mm3, %%mm0     \n\t"
 356                         "paddw %%mm4, %%mm1     \n\t"
 357                         "psraw %%mm2, %%mm0     \n\t"
 358                         "psraw %%mm2, %%mm1     \n\t"
 359                         "packuswb %%mm1, %%mm0  \n\t"
 360                         "movq %%mm0, (%1)       \n\t"
 361                         "add $16, %0            \n\t"
 362                         "add $8, %1             \n\t"
 363                         "cmp %2, %1             \n\t"
 364                         " jb 1b                 \n\t"
 365                         : "+r" (src1), "+r"(dst1)
 366                         : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(6-log2_scale)
 367                 );
 368                 src += src_stride;
 369                 dst += dst_stride;
 370         }
 371 //      if(width != mmxw)
 372 //              store_slice_c(dst + mmxw, src + mmxw, dst_stride, src_stride, width - mmxw, log2_scale);
 373 }
 374 #endif
 375
 376 static void (*store_slice)(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale)= store_slice_c;
 377
 378 static void (*requantize)(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation)= hardthresh_c;
 379
 380 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){
 381         int x, y, i;
 382         const int count= 1<<p->log2_count;
 383         const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15));
 384         uint64_t __attribute__((aligned(16))) block_align[32];
 385         DCTELEM *block = (DCTELEM *)block_align;
 386         DCTELEM *block2= (DCTELEM *)(block_align+16);
 387
 388         if (!src || !dst) return; // HACK avoid crash for Y8 colourspace
 389         for(y=0; y<height; y++){
 390                 int index= 8 + 8*stride + y*stride;
 391                 fast_memcpy(p->src + index, src + y*src_stride, width);
 392                 for(x=0; x<8; x++){
 393                         p->src[index         - x - 1]= p->src[index +         x    ];
 394                         p->src[index + width + x    ]= p->src[index + width - x - 1];
 395                 }
 396         }
 397         for(y=0; y<8; y++){
 398                 fast_memcpy(p->src + (      7-y)*stride, p->src + (      y+8)*stride, stride);
 399                 fast_memcpy(p->src + (height+8+y)*stride, p->src + (height-y+7)*stride, stride);
 400         }
 401         //FIXME (try edge emu)
 402
 403         for(y=0; y<height+8; y+=8){
 404                 memset(p->temp + (8+y)*stride, 0, 8*stride*sizeof(int16_t));
 405                 for(x=0; x<width+8; x+=8){
 406                         const int qps= 3 + is_luma;
 407                         int qp;
 408
 409                         if(p->qp)
 410                                 qp= p->qp;
 411                         else{
 412                                 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride];
 413                                 if(p->mpeg2) qp = FFMAX(1, qp>>1);
 414                         }
 415                         for(i=0; i<count; i++){
 416                                 const int x1= x + offset[i+count-1][0];
 417                                 const int y1= y + offset[i+count-1][1];
 418                                 const int index= x1 + y1*stride;
 419                                 p->dsp.get_pixels(block, p->src + index, stride);
 420                                 p->dsp.fdct(block);
 421                                 requantize(block2, block, qp, p->dsp.idct_permutation);
 422                                 p->dsp.idct(block2);
 423                                 add_block(p->temp + index, stride, block2);
 424                         }
 425                 }
 426                 if(y)
 427                         store_slice(dst + (y-8)*dst_stride, p->temp + 8 + y*stride, dst_stride, stride, width, XMIN(8, height+8-y), 6-p->log2_count);
 428         }
 429 #if 0
 430         for(y=0; y<height; y++){
 431                 for(x=0; x<width; x++){
 432                         if((((x>>6) ^ (y>>6)) & 1) == 0)
 433                                 dst[x + y*dst_stride]= p->src[8 + 8*stride  + x + y*stride];
 434                         if((x&63) == 0 || (y&63)==0)
 435                                 dst[x + y*dst_stride] += 128;
 436                 }
 437         }
 438 #endif
 439         //FIXME reorder for better caching
 440 }
 441
 442 static int config(struct vf_instance_s* vf,
 443         int width, int height, int d_width, int d_height,
 444         unsigned int flags, unsigned int outfmt){
 445         int h= (height+16+15)&(~15);
 446
 447         vf->priv->temp_stride= (width+16+15)&(~15);
 448         vf->priv->temp= malloc(vf->priv->temp_stride*h*sizeof(int16_t));
 449         vf->priv->src = malloc(vf->priv->temp_stride*h*sizeof(uint8_t));
 450
 451         return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
 452 }
 453
 454 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){
 455     if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
 456     // ok, we can do pp in-place (or pp disabled):
 457     vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
 458         mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
 459     mpi->planes[0]=vf->dmpi->planes[0];
 460     mpi->stride[0]=vf->dmpi->stride[0];
 461     mpi->width=vf->dmpi->width;
 462     if(mpi->flags&MP_IMGFLAG_PLANAR){
 463         mpi->planes[1]=vf->dmpi->planes[1];
 464         mpi->planes[2]=vf->dmpi->planes[2];
 465         mpi->stride[1]=vf->dmpi->stride[1];
 466         mpi->stride[2]=vf->dmpi->stride[2];
 467     }
 468     mpi->flags|=MP_IMGFLAG_DIRECT;
 469 }
 470
 471 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts){
 472         mp_image_t *dmpi;
 473
 474         if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
 475                 // no DR, so get a new image! hope we'll get DR buffer:
 476                 dmpi=vf_get_image(vf->next,mpi->imgfmt,
 477                     MP_IMGTYPE_TEMP,
 478                     MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
 479                     mpi->width,mpi->height);
 480                 vf_clone_mpi_attributes(dmpi, mpi);
 481         }else{
 482            dmpi=vf->dmpi;
 483         }
 484
 485         vf->priv->mpeg2= mpi->qscale_type;
 486         if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
 487             if(!vf->priv->non_b_qp)
 488                 vf->priv->non_b_qp= malloc(mpi->qstride * ((mpi->h + 15) >> 4));
 489             fast_memcpy(vf->priv->non_b_qp, mpi->qscale, mpi->qstride * ((mpi->h + 15) >> 4));
 490         }
 491         if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
 492             char *qp_tab= vf->priv->non_b_qp;
 493             if((vf->priv->mode&4) || !qp_tab)
 494                 qp_tab= mpi->qscale;
 495
 496             if(qp_tab || vf->priv->qp){
 497                 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, qp_tab, mpi->qstride, 1);
 498                 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, qp_tab, mpi->qstride, 0);
 499                 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, qp_tab, mpi->qstride, 0);
 500             }else{
 501                 memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]);
 502                 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]);
 503                 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]);
 504             }
 505         }
 506
 507 #if HAVE_MMX
 508         if(gCpuCaps.hasMMX) __asm__ volatile ("emms\n\t");
 509 #endif
 510 #if HAVE_MMX2
 511         if(gCpuCaps.hasMMX2) __asm__ volatile ("sfence\n\t");
 512 #endif
 513
 514         return vf_next_put_image(vf,dmpi, pts);
 515 }
 516
 517 static void uninit(struct vf_instance_s* vf){
 518         if(!vf->priv) return;
 519
 520         if(vf->priv->temp) free(vf->priv->temp);
 521         vf->priv->temp= NULL;
 522         if(vf->priv->src) free(vf->priv->src);
 523         vf->priv->src= NULL;
 524         if(vf->priv->avctx) free(vf->priv->avctx);
 525         vf->priv->avctx= NULL;
 526         if(vf->priv->non_b_qp) free(vf->priv->non_b_qp);
 527         vf->priv->non_b_qp= NULL;
 528
 529         free(vf->priv);
 530         vf->priv=NULL;
 531 }
 532
 533 //===========================================================================//
 534 static int query_format(struct vf_instance_s* vf, unsigned int fmt){
 535     switch(fmt){
 536         case IMGFMT_YVU9:
 537         case IMGFMT_IF09:
 538         case IMGFMT_YV12:
 539         case IMGFMT_I420:
 540         case IMGFMT_IYUV:
 541         case IMGFMT_CLPL:
 542         case IMGFMT_Y800:
 543         case IMGFMT_Y8:
 544         case IMGFMT_444P:
 545         case IMGFMT_422P:
 546         case IMGFMT_411P:
 547             return vf_next_query_format(vf,fmt);
 548     }
 549     return 0;
 550 }
 551
 552 static int control(struct vf_instance_s* vf, int request, void* data){
 553     switch(request){
 554     case VFCTRL_QUERY_MAX_PP_LEVEL:
 555         return 6;
 556     case VFCTRL_SET_PP_LEVEL:
 557         vf->priv->log2_count= *((unsigned int*)data);
 558         return CONTROL_TRUE;
 559     }
 560     return vf_next_control(vf,request,data);
 561 }
 562
 563 static int open(vf_instance_t *vf, char* args){
 564
 565     int log2c=-1;
 566
 567     vf->config=config;
 568     vf->put_image=put_image;
 569     vf->get_image=get_image;
 570     vf->query_format=query_format;
 571     vf->uninit=uninit;
 572     vf->control= control;
 573     vf->priv=malloc(sizeof(struct vf_priv_s));
 574     memset(vf->priv, 0, sizeof(struct vf_priv_s));
 575
 576     avcodec_init();
 577
 578     vf->priv->avctx= avcodec_alloc_context();
 579     dsputil_init(&vf->priv->dsp, vf->priv->avctx);
 580
 581     vf->priv->log2_count= 3;
 582
 583     if (args) sscanf(args, "%d:%d:%d", &log2c, &vf->priv->qp, &vf->priv->mode);
 584
 585     if( log2c >=0 && log2c <=6 )
 586         vf->priv->log2_count = log2c;
 587
 588     if(vf->priv->qp < 0)
 589         vf->priv->qp = 0;
 590
 591     switch(vf->priv->mode&3){
 592         default:
 593         case 0: requantize= hardthresh_c; break;
 594         case 1: requantize= softthresh_c; break;
 595     }
 596
 597 #if HAVE_MMX
 598     if(gCpuCaps.hasMMX){
 599         store_slice= store_slice_mmx;
 600         switch(vf->priv->mode&3){
 601             case 0: requantize= hardthresh_mmx; break;
 602             case 1: requantize= softthresh_mmx; break;
 603         }
 604     }
 605 #endif
 606
 607     return 1;
 608 }
 609
 610 const vf_info_t vf_info_spp = {
 611     "simple postprocess",
 612     "spp",
 613     "Michael Niedermayer",
 614     "",
 615     open,
 616     NULL
 617 };