libmpcodecs/vf_spp.c

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