| 1 | /* |
| 2 | * This file is part of FFmpeg. |
| 3 | * |
| 4 | * FFmpeg is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU Lesser General Public |
| 6 | * License as published by the Free Software Foundation; either |
| 7 | * version 2.1 of the License, or (at your option) any later version. |
| 8 | * |
| 9 | * FFmpeg 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 GNU |
| 12 | * Lesser General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU Lesser General Public |
| 15 | * License along with FFmpeg; if not, write to the Free Software |
| 16 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 17 | */ |
| 18 | |
| 19 | /** |
| 20 | * @file |
| 21 | *@brief IntraX8 frame subdecoder image manipulation routines |
| 22 | */ |
| 23 | |
| 24 | #include "intrax8dsp.h" |
| 25 | #include "libavutil/common.h" |
| 26 | |
| 27 | /* |
| 28 | area positions, #3 is 1 pixel only, other are 8 pixels |
| 29 | |66666666| |
| 30 | 3|44444444|55555555| |
| 31 | - -+--------+--------+ |
| 32 | 1 2|XXXXXXXX| |
| 33 | 1 2|XXXXXXXX| |
| 34 | 1 2|XXXXXXXX| |
| 35 | 1 2|XXXXXXXX| |
| 36 | 1 2|XXXXXXXX| |
| 37 | 1 2|XXXXXXXX| |
| 38 | 1 2|XXXXXXXX| |
| 39 | 1 2|XXXXXXXX| |
| 40 | ^-start |
| 41 | */ |
| 42 | |
| 43 | #define area1 (0) |
| 44 | #define area2 (8) |
| 45 | #define area3 (8+8) |
| 46 | #define area4 (8+8+1) |
| 47 | #define area5 (8+8+1+8) |
| 48 | #define area6 (8+8+1+16) |
| 49 | |
| 50 | /** |
| 51 | Collect statistics and prepare the edge pixels required by the other spatial compensation functions. |
| 52 | |
| 53 | * @param src pointer to the beginning of the processed block |
| 54 | * @param dst pointer to emu_edge, edge pixels are stored the way other compensation routines do. |
| 55 | * @param linesize byte offset between 2 vertical pixels in the source image |
| 56 | * @param range pointer to the variable where the edge pixel range is to be stored (max-min values) |
| 57 | * @param psum pointer to the variable where the edge pixel sum is to be stored |
| 58 | * @param edges Informs this routine that the block is on an image border, so it has to interpolate the missing edge pixels. |
| 59 | and some of the edge pixels should be interpolated, the flag has the following meaning: |
| 60 | 1 - mb_x==0 - first block in the row, interpolate area #1,#2,#3; |
| 61 | 2 - mb_y==0 - first row, interpolate area #3,#4,#5,#6; |
| 62 | note: 1|2 - mb_x==mb_y==0 - first block, use 0x80 value for all areas; |
| 63 | 4 - mb_x>= (mb_width-1) last block in the row, interpolate area #5; |
| 64 | */ |
| 65 | static void x8_setup_spatial_compensation(uint8_t *src, uint8_t *dst, int linesize, |
| 66 | int * range, int * psum, int edges){ |
| 67 | uint8_t * ptr; |
| 68 | int sum; |
| 69 | int i; |
| 70 | int min_pix,max_pix; |
| 71 | uint8_t c; |
| 72 | |
| 73 | if((edges&3)==3){ |
| 74 | *psum=0x80*(8+1+8+2); |
| 75 | *range=0; |
| 76 | memset(dst,0x80,16+1+16+8); |
| 77 | //this triggers flat_dc for sure. |
| 78 | //flat_dc avoids all (other) prediction modes, but requires dc_level decoding. |
| 79 | return; |
| 80 | } |
| 81 | |
| 82 | min_pix=256; |
| 83 | max_pix=-1; |
| 84 | |
| 85 | sum=0; |
| 86 | |
| 87 | if(!(edges&1)){//(mb_x!=0)//there is previous block on this row |
| 88 | ptr=src-1;//left column, area 2 |
| 89 | for(i=7;i>=0;i--){ |
| 90 | c=*(ptr-1);//area1, same mb as area2, no need to check |
| 91 | dst[area1+i]=c; |
| 92 | c=*(ptr); |
| 93 | |
| 94 | sum+=c; |
| 95 | min_pix=FFMIN(min_pix,c); |
| 96 | max_pix=FFMAX(max_pix,c); |
| 97 | dst[area2+i]=c; |
| 98 | |
| 99 | ptr+=linesize; |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | if(!(edges&2)){ //(mb_y!=0)//there is row above |
| 104 | ptr=src-linesize;//top line |
| 105 | for(i=0;i<8;i++){ |
| 106 | c=*(ptr+i); |
| 107 | sum+=c; |
| 108 | min_pix=FFMIN(min_pix, c); |
| 109 | max_pix=FFMAX(max_pix, c); |
| 110 | } |
| 111 | if(edges&4){//last block on the row? |
| 112 | memset(dst+area5,c,8);//set with last pixel fr |
| 113 | memcpy(dst+area4, ptr, 8); |
| 114 | }else{ |
| 115 | memcpy(dst+area4, ptr, 16);//both area4 and 5 |
| 116 | } |
| 117 | memcpy(dst+area6, ptr-linesize, 8);//area6 always present in the above block |
| 118 | } |
| 119 | //now calculate the stuff we need |
| 120 | if(edges&3){//mb_x==0 || mb_y==0){ |
| 121 | int avg=(sum+4)>>3; |
| 122 | if(edges&1){ //(mb_x==0) {//implies mb_y!=0 |
| 123 | memset(dst+area1,avg,8+8+1);//areas 1,2 and 3 are averaged |
| 124 | }else{//implies y==0 x!=0 |
| 125 | memset(dst+area3,avg, 1+16+8);//areas 3, 4,5,6 |
| 126 | } |
| 127 | sum+=avg*9; |
| 128 | }else{ |
| 129 | uint8_t c=*(src-1-linesize);//the edge pixel, in the top line and left column |
| 130 | dst[area3]=c; |
| 131 | sum+=c; |
| 132 | //edge pixel is not part of min/max |
| 133 | } |
| 134 | (*range) = max_pix - min_pix; |
| 135 | sum += *(dst+area5) + *(dst+area5+1); |
| 136 | *psum = sum; |
| 137 | } |
| 138 | |
| 139 | |
| 140 | static const uint16_t zero_prediction_weights[64*2] = { |
| 141 | 640, 640, 669, 480, 708, 354, 748, 257, 792, 198, 760, 143, 808, 101, 772, 72, |
| 142 | 480, 669, 537, 537, 598, 416, 661, 316, 719, 250, 707, 185, 768, 134, 745, 97, |
| 143 | 354, 708, 416, 598, 488, 488, 564, 388, 634, 317, 642, 241, 716, 179, 706, 132, |
| 144 | 257, 748, 316, 661, 388, 564, 469, 469, 543, 395, 571, 311, 655, 238, 660, 180, |
| 145 | 198, 792, 250, 719, 317, 634, 395, 543, 469, 469, 507, 380, 597, 299, 616, 231, |
| 146 | 161, 855, 206, 788, 266, 710, 340, 623, 411, 548, 455, 455, 548, 366, 576, 288, |
| 147 | 122, 972, 159, 914, 211, 842, 276, 758, 341, 682, 389, 584, 483, 483, 520, 390, |
| 148 | 110, 1172, 144, 1107, 193, 1028, 254, 932, 317, 846, 366, 731, 458, 611, 499, 499 |
| 149 | }; |
| 150 | |
| 151 | static void spatial_compensation_0(uint8_t *src , uint8_t *dst, int linesize){ |
| 152 | int i,j; |
| 153 | int x,y; |
| 154 | unsigned int p;//power divided by 2 |
| 155 | int a; |
| 156 | uint16_t left_sum[2][8] = { { 0 } }; |
| 157 | uint16_t top_sum[2][8] = { { 0 } }; |
| 158 | |
| 159 | for(i=0;i<8;i++){ |
| 160 | a=src[area2+7-i]<<4; |
| 161 | for(j=0;j<8;j++){ |
| 162 | p=abs(i-j); |
| 163 | left_sum[p&1][j]+= a>>(p>>1); |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | for(i=0;i<8;i++){ |
| 168 | a=src[area4+i]<<4; |
| 169 | for(j=0;j<8;j++){ |
| 170 | p=abs(i-j); |
| 171 | top_sum[p&1][j]+= a>>(p>>1); |
| 172 | } |
| 173 | } |
| 174 | for(;i<10;i++){ |
| 175 | a=src[area4+i]<<4; |
| 176 | for(j=5;j<8;j++){ |
| 177 | p=abs(i-j); |
| 178 | top_sum[p&1][j]+= a>>(p>>1); |
| 179 | } |
| 180 | } |
| 181 | for(;i<12;i++){ |
| 182 | a=src[area4+i]<<4; |
| 183 | for(j=7;j<8;j++){ |
| 184 | p=abs(i-j); |
| 185 | top_sum[p&1][j]+= a>>(p>>1); |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | for(i=0;i<8;i++){ |
| 190 | top_sum [0][i]+=(top_sum [1][i]*181 + 128 )>>8;//181 is sqrt(2)/2 |
| 191 | left_sum[0][i]+=(left_sum[1][i]*181 + 128 )>>8; |
| 192 | } |
| 193 | for(y=0;y<8;y++){ |
| 194 | for(x=0;x<8;x++){ |
| 195 | dst[x] = ( |
| 196 | (uint32_t)top_sum [0][x]*zero_prediction_weights[y*16+x*2+0] + |
| 197 | (uint32_t)left_sum[0][y]*zero_prediction_weights[y*16+x*2+1] + |
| 198 | 0x8000 |
| 199 | )>>16; |
| 200 | } |
| 201 | dst+=linesize; |
| 202 | } |
| 203 | } |
| 204 | static void spatial_compensation_1(uint8_t *src , uint8_t *dst, int linesize){ |
| 205 | int x,y; |
| 206 | |
| 207 | for(y=0;y<8;y++){ |
| 208 | for(x=0;x<8;x++){ |
| 209 | dst[x]=src[area4 + FFMIN(2*y+x+2, 15) ]; |
| 210 | } |
| 211 | dst+=linesize; |
| 212 | } |
| 213 | } |
| 214 | static void spatial_compensation_2(uint8_t *src , uint8_t *dst, int linesize){ |
| 215 | int x,y; |
| 216 | |
| 217 | for(y=0;y<8;y++){ |
| 218 | for(x=0;x<8;x++){ |
| 219 | dst[x]=src[area4 +1+y+x]; |
| 220 | } |
| 221 | dst+=linesize; |
| 222 | } |
| 223 | } |
| 224 | static void spatial_compensation_3(uint8_t *src , uint8_t *dst, int linesize){ |
| 225 | int x,y; |
| 226 | |
| 227 | for(y=0;y<8;y++){ |
| 228 | for(x=0;x<8;x++){ |
| 229 | dst[x]=src[area4 +((y+1)>>1)+x]; |
| 230 | } |
| 231 | dst+=linesize; |
| 232 | } |
| 233 | } |
| 234 | static void spatial_compensation_4(uint8_t *src , uint8_t *dst, int linesize){ |
| 235 | int x,y; |
| 236 | |
| 237 | for(y=0;y<8;y++){ |
| 238 | for(x=0;x<8;x++){ |
| 239 | dst[x]=( src[area4+x] + src[area6+x] + 1 )>>1; |
| 240 | } |
| 241 | dst+=linesize; |
| 242 | } |
| 243 | } |
| 244 | static void spatial_compensation_5(uint8_t *src , uint8_t *dst, int linesize){ |
| 245 | int x,y; |
| 246 | |
| 247 | for(y=0;y<8;y++){ |
| 248 | for(x=0;x<8;x++){ |
| 249 | if(2*x-y<0){ |
| 250 | dst[x]=src[area2+9+2*x-y]; |
| 251 | }else{ |
| 252 | dst[x]=src[area4 +x-((y+1)>>1)]; |
| 253 | } |
| 254 | } |
| 255 | dst+=linesize; |
| 256 | } |
| 257 | } |
| 258 | static void spatial_compensation_6(uint8_t *src , uint8_t *dst, int linesize){ |
| 259 | int x,y; |
| 260 | |
| 261 | for(y=0;y<8;y++){ |
| 262 | for(x=0;x<8;x++){ |
| 263 | dst[x]=src[area3+x-y]; |
| 264 | } |
| 265 | dst+=linesize; |
| 266 | } |
| 267 | } |
| 268 | static void spatial_compensation_7(uint8_t *src , uint8_t *dst, int linesize){ |
| 269 | int x,y; |
| 270 | |
| 271 | for(y=0;y<8;y++){ |
| 272 | for(x=0;x<8;x++){ |
| 273 | if(x-2*y>0){ |
| 274 | dst[x]=( src[area3-1+x-2*y] + src[area3+x-2*y] + 1)>>1; |
| 275 | }else{ |
| 276 | dst[x]=src[area2+8-y +(x>>1)]; |
| 277 | } |
| 278 | } |
| 279 | dst+=linesize; |
| 280 | } |
| 281 | } |
| 282 | static void spatial_compensation_8(uint8_t *src , uint8_t *dst, int linesize){ |
| 283 | int x,y; |
| 284 | |
| 285 | for(y=0;y<8;y++){ |
| 286 | for(x=0;x<8;x++){ |
| 287 | dst[x]=( src[area1+7-y] + src[area2+7-y] + 1 )>>1; |
| 288 | } |
| 289 | dst+=linesize; |
| 290 | } |
| 291 | } |
| 292 | static void spatial_compensation_9(uint8_t *src , uint8_t *dst, int linesize){ |
| 293 | int x,y; |
| 294 | |
| 295 | for(y=0;y<8;y++){ |
| 296 | for(x=0;x<8;x++){ |
| 297 | dst[x]=src[area2+6-FFMIN(x+y,6)]; |
| 298 | } |
| 299 | dst+=linesize; |
| 300 | } |
| 301 | } |
| 302 | static void spatial_compensation_10(uint8_t *src , uint8_t *dst, int linesize){ |
| 303 | int x,y; |
| 304 | |
| 305 | for(y=0;y<8;y++){ |
| 306 | for(x=0;x<8;x++){ |
| 307 | dst[x]=(src[area2+7-y]*(8-x)+src[area4+x]*x+4)>>3; |
| 308 | } |
| 309 | dst+=linesize; |
| 310 | } |
| 311 | } |
| 312 | static void spatial_compensation_11(uint8_t *src , uint8_t *dst, int linesize){ |
| 313 | int x,y; |
| 314 | |
| 315 | for(y=0;y<8;y++){ |
| 316 | for(x=0;x<8;x++){ |
| 317 | dst[x]=(src[area2+7-y]*y+src[area4+x]*(8-y)+4)>>3; |
| 318 | } |
| 319 | dst+=linesize; |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | static void x8_loop_filter(uint8_t * ptr, const int a_stride, const int b_stride, int quant){ |
| 324 | int i,t; |
| 325 | int p0,p1,p2,p3,p4,p5,p6,p7,p8,p9; |
| 326 | int ql=(quant+10)>>3; |
| 327 | |
| 328 | for(i=0; i<8; i++,ptr+=b_stride){ |
| 329 | p0=ptr[-5*a_stride]; |
| 330 | p1=ptr[-4*a_stride]; |
| 331 | p2=ptr[-3*a_stride]; |
| 332 | p3=ptr[-2*a_stride]; |
| 333 | p4=ptr[-1*a_stride]; |
| 334 | p5=ptr[ 0 ]; |
| 335 | p6=ptr[ 1*a_stride]; |
| 336 | p7=ptr[ 2*a_stride]; |
| 337 | p8=ptr[ 3*a_stride]; |
| 338 | p9=ptr[ 4*a_stride]; |
| 339 | |
| 340 | t= |
| 341 | (FFABS(p1-p2) <= ql) + |
| 342 | (FFABS(p2-p3) <= ql) + |
| 343 | (FFABS(p3-p4) <= ql) + |
| 344 | (FFABS(p4-p5) <= ql); |
| 345 | if(t>0){//You need at least 1 to be able to reach a total score of 6. |
| 346 | t+= |
| 347 | (FFABS(p5-p6) <= ql) + |
| 348 | (FFABS(p6-p7) <= ql) + |
| 349 | (FFABS(p7-p8) <= ql) + |
| 350 | (FFABS(p8-p9) <= ql) + |
| 351 | (FFABS(p0-p1) <= ql); |
| 352 | if(t>=6){ |
| 353 | int min,max; |
| 354 | |
| 355 | min=max=p1; |
| 356 | min=FFMIN(min,p3); max=FFMAX(max,p3); |
| 357 | min=FFMIN(min,p5); max=FFMAX(max,p5); |
| 358 | min=FFMIN(min,p8); max=FFMAX(max,p8); |
| 359 | if(max-min<2*quant){//early stop |
| 360 | min=FFMIN(min,p2); max=FFMAX(max,p2); |
| 361 | min=FFMIN(min,p4); max=FFMAX(max,p4); |
| 362 | min=FFMIN(min,p6); max=FFMAX(max,p6); |
| 363 | min=FFMIN(min,p7); max=FFMAX(max,p7); |
| 364 | if(max-min<2*quant){ |
| 365 | ptr[-2*a_stride]=(4*p2 + 3*p3 + 1*p7 + 4)>>3; |
| 366 | ptr[-1*a_stride]=(3*p2 + 3*p4 + 2*p7 + 4)>>3; |
| 367 | ptr[ 0 ]=(2*p2 + 3*p5 + 3*p7 + 4)>>3; |
| 368 | ptr[ 1*a_stride]=(1*p2 + 3*p6 + 4*p7 + 4)>>3; |
| 369 | continue; |
| 370 | }; |
| 371 | } |
| 372 | } |
| 373 | } |
| 374 | { |
| 375 | int x,x0,x1,x2; |
| 376 | int m; |
| 377 | |
| 378 | x0 = (2*p3 - 5*p4 + 5*p5 - 2*p6 + 4)>>3; |
| 379 | if(FFABS(x0) < quant){ |
| 380 | x1=(2*p1 - 5*p2 + 5*p3 - 2*p4 + 4)>>3; |
| 381 | x2=(2*p5 - 5*p6 + 5*p7 - 2*p8 + 4)>>3; |
| 382 | |
| 383 | x=FFABS(x0) - FFMIN( FFABS(x1), FFABS(x2) ); |
| 384 | m=p4-p5; |
| 385 | |
| 386 | if( x > 0 && (m^x0) <0){ |
| 387 | int32_t sign; |
| 388 | |
| 389 | sign=m>>31; |
| 390 | m=(m^sign)-sign;//abs(m) |
| 391 | m>>=1; |
| 392 | |
| 393 | x=(5*x)>>3; |
| 394 | |
| 395 | if(x>m) x=m; |
| 396 | |
| 397 | x=(x^sign)-sign; |
| 398 | |
| 399 | ptr[-1*a_stride] -= x; |
| 400 | ptr[ 0] += x; |
| 401 | } |
| 402 | } |
| 403 | } |
| 404 | } |
| 405 | } |
| 406 | |
| 407 | static void x8_h_loop_filter(uint8_t *src, int stride, int qscale){ |
| 408 | x8_loop_filter(src, stride, 1, qscale); |
| 409 | } |
| 410 | |
| 411 | static void x8_v_loop_filter(uint8_t *src, int stride, int qscale){ |
| 412 | x8_loop_filter(src, 1, stride, qscale); |
| 413 | } |
| 414 | |
| 415 | av_cold void ff_intrax8dsp_init(IntraX8DSPContext *dsp) |
| 416 | { |
| 417 | dsp->h_loop_filter=x8_h_loop_filter; |
| 418 | dsp->v_loop_filter=x8_v_loop_filter; |
| 419 | dsp->setup_spatial_compensation=x8_setup_spatial_compensation; |
| 420 | dsp->spatial_compensation[0]=spatial_compensation_0; |
| 421 | dsp->spatial_compensation[1]=spatial_compensation_1; |
| 422 | dsp->spatial_compensation[2]=spatial_compensation_2; |
| 423 | dsp->spatial_compensation[3]=spatial_compensation_3; |
| 424 | dsp->spatial_compensation[4]=spatial_compensation_4; |
| 425 | dsp->spatial_compensation[5]=spatial_compensation_5; |
| 426 | dsp->spatial_compensation[6]=spatial_compensation_6; |
| 427 | dsp->spatial_compensation[7]=spatial_compensation_7; |
| 428 | dsp->spatial_compensation[8]=spatial_compensation_8; |
| 429 | dsp->spatial_compensation[9]=spatial_compensation_9; |
| 430 | dsp->spatial_compensation[10]=spatial_compensation_10; |
| 431 | dsp->spatial_compensation[11]=spatial_compensation_11; |
| 432 | } |