Libav
|
00001 /* 00002 * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at> 00003 * 00004 * This file is part of FFmpeg. 00005 * 00006 * FFmpeg is free software; you can redistribute it and/or 00007 * modify it under the terms of the GNU Lesser General Public 00008 * License as published by the Free Software Foundation; either 00009 * version 2.1 of the License, or (at your option) any later version. 00010 * 00011 * FFmpeg is distributed in the hope that it will be useful, 00012 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00013 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00014 * Lesser General Public License for more details. 00015 * 00016 * You should have received a copy of the GNU Lesser General Public 00017 * License along with FFmpeg; if not, write to the Free Software 00018 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00019 */ 00020 00021 #include "libavutil/attributes.h" 00022 #include "dsputil.h" 00023 #include "dwt.h" 00024 00025 void ff_slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer) 00026 { 00027 int i; 00028 00029 buf->base_buffer = base_buffer; 00030 buf->line_count = line_count; 00031 buf->line_width = line_width; 00032 buf->data_count = max_allocated_lines; 00033 buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count); 00034 buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines); 00035 00036 for(i = 0; i < max_allocated_lines; i++){ 00037 buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width); 00038 } 00039 00040 buf->data_stack_top = max_allocated_lines - 1; 00041 } 00042 00043 IDWTELEM * ff_slice_buffer_load_line(slice_buffer * buf, int line) 00044 { 00045 IDWTELEM * buffer; 00046 00047 assert(buf->data_stack_top >= 0); 00048 // assert(!buf->line[line]); 00049 if (buf->line[line]) 00050 return buf->line[line]; 00051 00052 buffer = buf->data_stack[buf->data_stack_top]; 00053 buf->data_stack_top--; 00054 buf->line[line] = buffer; 00055 00056 return buffer; 00057 } 00058 00059 void ff_slice_buffer_release(slice_buffer * buf, int line) 00060 { 00061 IDWTELEM * buffer; 00062 00063 assert(line >= 0 && line < buf->line_count); 00064 assert(buf->line[line]); 00065 00066 buffer = buf->line[line]; 00067 buf->data_stack_top++; 00068 buf->data_stack[buf->data_stack_top] = buffer; 00069 buf->line[line] = NULL; 00070 } 00071 00072 void ff_slice_buffer_flush(slice_buffer * buf) 00073 { 00074 int i; 00075 for(i = 0; i < buf->line_count; i++){ 00076 if (buf->line[i]) 00077 ff_slice_buffer_release(buf, i); 00078 } 00079 } 00080 00081 void ff_slice_buffer_destroy(slice_buffer * buf) 00082 { 00083 int i; 00084 ff_slice_buffer_flush(buf); 00085 00086 for(i = buf->data_count - 1; i >= 0; i--){ 00087 av_freep(&buf->data_stack[i]); 00088 } 00089 av_freep(&buf->data_stack); 00090 av_freep(&buf->line); 00091 } 00092 00093 static inline int mirror(int v, int m){ 00094 while((unsigned)v > (unsigned)m){ 00095 v=-v; 00096 if(v<0) v+= 2*m; 00097 } 00098 return v; 00099 } 00100 00101 static av_always_inline void 00102 lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, 00103 int dst_step, int src_step, int ref_step, 00104 int width, int mul, int add, int shift, 00105 int highpass, int inverse){ 00106 const int mirror_left= !highpass; 00107 const int mirror_right= (width&1) ^ highpass; 00108 const int w= (width>>1) - 1 + (highpass & width); 00109 int i; 00110 00111 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) 00112 if(mirror_left){ 00113 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); 00114 dst += dst_step; 00115 src += src_step; 00116 } 00117 00118 for(i=0; i<w; i++){ 00119 dst[i*dst_step] = 00120 LIFT(src[i*src_step], 00121 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), 00122 inverse); 00123 } 00124 00125 if(mirror_right){ 00126 dst[w*dst_step] = 00127 LIFT(src[w*src_step], 00128 ((mul*2*ref[w*ref_step]+add)>>shift), 00129 inverse); 00130 } 00131 } 00132 00133 static av_always_inline void 00134 inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, 00135 int dst_step, int src_step, int ref_step, 00136 int width, int mul, int add, int shift, 00137 int highpass, int inverse){ 00138 const int mirror_left= !highpass; 00139 const int mirror_right= (width&1) ^ highpass; 00140 const int w= (width>>1) - 1 + (highpass & width); 00141 int i; 00142 00143 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) 00144 if(mirror_left){ 00145 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); 00146 dst += dst_step; 00147 src += src_step; 00148 } 00149 00150 for(i=0; i<w; i++){ 00151 dst[i*dst_step] = 00152 LIFT(src[i*src_step], 00153 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), 00154 inverse); 00155 } 00156 00157 if(mirror_right){ 00158 dst[w*dst_step] = 00159 LIFT(src[w*src_step], 00160 ((mul*2*ref[w*ref_step]+add)>>shift), 00161 inverse); 00162 } 00163 } 00164 00165 #ifndef liftS 00166 static av_always_inline void 00167 liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, 00168 int dst_step, int src_step, int ref_step, 00169 int width, int mul, int add, int shift, 00170 int highpass, int inverse){ 00171 const int mirror_left= !highpass; 00172 const int mirror_right= (width&1) ^ highpass; 00173 const int w= (width>>1) - 1 + (highpass & width); 00174 int i; 00175 00176 assert(shift == 4); 00177 #define LIFTS(src, ref, inv) \ 00178 ((inv) ? \ 00179 (src) + (((ref) + 4*(src))>>shift): \ 00180 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23))) 00181 if(mirror_left){ 00182 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); 00183 dst += dst_step; 00184 src += src_step; 00185 } 00186 00187 for(i=0; i<w; i++){ 00188 dst[i*dst_step] = 00189 LIFTS(src[i*src_step], 00190 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, 00191 inverse); 00192 } 00193 00194 if(mirror_right){ 00195 dst[w*dst_step] = 00196 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); 00197 } 00198 } 00199 static av_always_inline void 00200 inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, 00201 int dst_step, int src_step, int ref_step, 00202 int width, int mul, int add, int shift, 00203 int highpass, int inverse){ 00204 const int mirror_left= !highpass; 00205 const int mirror_right= (width&1) ^ highpass; 00206 const int w= (width>>1) - 1 + (highpass & width); 00207 int i; 00208 00209 assert(shift == 4); 00210 #define LIFTS(src, ref, inv) \ 00211 ((inv) ? \ 00212 (src) + (((ref) + 4*(src))>>shift): \ 00213 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23))) 00214 if(mirror_left){ 00215 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); 00216 dst += dst_step; 00217 src += src_step; 00218 } 00219 00220 for(i=0; i<w; i++){ 00221 dst[i*dst_step] = 00222 LIFTS(src[i*src_step], 00223 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, 00224 inverse); 00225 } 00226 00227 if(mirror_right){ 00228 dst[w*dst_step] = 00229 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); 00230 } 00231 } 00232 #endif /* ! liftS */ 00233 00234 static void horizontal_decompose53i(DWTELEM *b, int width){ 00235 DWTELEM temp[width]; 00236 const int width2= width>>1; 00237 int x; 00238 const int w2= (width+1)>>1; 00239 00240 for(x=0; x<width2; x++){ 00241 temp[x ]= b[2*x ]; 00242 temp[x+w2]= b[2*x + 1]; 00243 } 00244 if(width&1) 00245 temp[x ]= b[2*x ]; 00246 #if 0 00247 { 00248 int A1,A2,A3,A4; 00249 A2= temp[1 ]; 00250 A4= temp[0 ]; 00251 A1= temp[0+width2]; 00252 A1 -= (A2 + A4)>>1; 00253 A4 += (A1 + 1)>>1; 00254 b[0+width2] = A1; 00255 b[0 ] = A4; 00256 for(x=1; x+1<width2; x+=2){ 00257 A3= temp[x+width2]; 00258 A4= temp[x+1 ]; 00259 A3 -= (A2 + A4)>>1; 00260 A2 += (A1 + A3 + 2)>>2; 00261 b[x+width2] = A3; 00262 b[x ] = A2; 00263 00264 A1= temp[x+1+width2]; 00265 A2= temp[x+2 ]; 00266 A1 -= (A2 + A4)>>1; 00267 A4 += (A1 + A3 + 2)>>2; 00268 b[x+1+width2] = A1; 00269 b[x+1 ] = A4; 00270 } 00271 A3= temp[width-1]; 00272 A3 -= A2; 00273 A2 += (A1 + A3 + 2)>>2; 00274 b[width -1] = A3; 00275 b[width2-1] = A2; 00276 } 00277 #else 00278 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0); 00279 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0); 00280 #endif /* 0 */ 00281 } 00282 00283 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00284 int i; 00285 00286 for(i=0; i<width; i++){ 00287 b1[i] -= (b0[i] + b2[i])>>1; 00288 } 00289 } 00290 00291 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00292 int i; 00293 00294 for(i=0; i<width; i++){ 00295 b1[i] += (b0[i] + b2[i] + 2)>>2; 00296 } 00297 } 00298 00299 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){ 00300 int y; 00301 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride; 00302 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride; 00303 00304 for(y=-2; y<height; y+=2){ 00305 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; 00306 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; 00307 00308 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width); 00309 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width); 00310 00311 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width); 00312 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width); 00313 00314 b0=b2; 00315 b1=b3; 00316 } 00317 } 00318 00319 static void horizontal_decompose97i(DWTELEM *b, int width){ 00320 DWTELEM temp[width]; 00321 const int w2= (width+1)>>1; 00322 00323 lift (temp+w2, b +1, b , 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1); 00324 liftS(temp , b , temp+w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0); 00325 lift (b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0); 00326 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0); 00327 } 00328 00329 00330 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00331 int i; 00332 00333 for(i=0; i<width; i++){ 00334 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; 00335 } 00336 } 00337 00338 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00339 int i; 00340 00341 for(i=0; i<width; i++){ 00342 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; 00343 } 00344 } 00345 00346 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00347 int i; 00348 00349 for(i=0; i<width; i++){ 00350 #ifdef liftS 00351 b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; 00352 #else 00353 b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23); 00354 #endif 00355 } 00356 } 00357 00358 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ 00359 int i; 00360 00361 for(i=0; i<width; i++){ 00362 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; 00363 } 00364 } 00365 00366 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){ 00367 int y; 00368 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride; 00369 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride; 00370 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride; 00371 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride; 00372 00373 for(y=-4; y<height; y+=2){ 00374 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; 00375 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; 00376 00377 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width); 00378 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width); 00379 00380 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width); 00381 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width); 00382 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width); 00383 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width); 00384 00385 b0=b2; 00386 b1=b3; 00387 b2=b4; 00388 b3=b5; 00389 } 00390 } 00391 00392 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ 00393 int level; 00394 00395 for(level=0; level<decomposition_count; level++){ 00396 switch(type){ 00397 case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break; 00398 case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break; 00399 } 00400 } 00401 } 00402 00403 static void horizontal_compose53i(IDWTELEM *b, int width){ 00404 IDWTELEM temp[width]; 00405 const int width2= width>>1; 00406 const int w2= (width+1)>>1; 00407 int x; 00408 00409 for(x=0; x<width2; x++){ 00410 temp[2*x ]= b[x ]; 00411 temp[2*x + 1]= b[x+w2]; 00412 } 00413 if(width&1) 00414 temp[2*x ]= b[x ]; 00415 00416 b[0] = temp[0] - ((temp[1]+1)>>1); 00417 for(x=2; x<width-1; x+=2){ 00418 b[x ] = temp[x ] - ((temp[x-1] + temp[x+1]+2)>>2); 00419 b[x-1] = temp[x-1] + ((b [x-2] + b [x ]+1)>>1); 00420 } 00421 if(width&1){ 00422 b[x ] = temp[x ] - ((temp[x-1]+1)>>1); 00423 b[x-1] = temp[x-1] + ((b [x-2] + b [x ]+1)>>1); 00424 }else 00425 b[x-1] = temp[x-1] + b[x-2]; 00426 } 00427 00428 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00429 int i; 00430 00431 for(i=0; i<width; i++){ 00432 b1[i] += (b0[i] + b2[i])>>1; 00433 } 00434 } 00435 00436 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00437 int i; 00438 00439 for(i=0; i<width; i++){ 00440 b1[i] -= (b0[i] + b2[i] + 2)>>2; 00441 } 00442 } 00443 00444 static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){ 00445 cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line); 00446 cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line); 00447 cs->y = -1; 00448 } 00449 00450 static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){ 00451 cs->b0 = buffer + mirror(-1-1, height-1)*stride; 00452 cs->b1 = buffer + mirror(-1 , height-1)*stride; 00453 cs->y = -1; 00454 } 00455 00456 static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){ 00457 int y= cs->y; 00458 00459 IDWTELEM *b0= cs->b0; 00460 IDWTELEM *b1= cs->b1; 00461 IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line); 00462 IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line); 00463 00464 if(y+1<(unsigned)height && y<(unsigned)height){ 00465 int x; 00466 00467 for(x=0; x<width; x++){ 00468 b2[x] -= (b1[x] + b3[x] + 2)>>2; 00469 b1[x] += (b0[x] + b2[x])>>1; 00470 } 00471 }else{ 00472 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); 00473 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); 00474 } 00475 00476 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); 00477 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); 00478 00479 cs->b0 = b2; 00480 cs->b1 = b3; 00481 cs->y += 2; 00482 } 00483 00484 static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){ 00485 int y= cs->y; 00486 IDWTELEM *b0= cs->b0; 00487 IDWTELEM *b1= cs->b1; 00488 IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride; 00489 IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride; 00490 00491 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); 00492 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); 00493 00494 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); 00495 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); 00496 00497 cs->b0 = b2; 00498 cs->b1 = b3; 00499 cs->y += 2; 00500 } 00501 00502 static void av_unused spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){ 00503 DWTCompose cs; 00504 spatial_compose53i_init(&cs, buffer, height, stride); 00505 while(cs.y <= height) 00506 spatial_compose53i_dy(&cs, buffer, width, height, stride); 00507 } 00508 00509 00510 void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){ 00511 IDWTELEM temp[width]; 00512 const int w2= (width+1)>>1; 00513 00514 #if 0 //maybe more understadable but slower 00515 inv_lift (temp , b , b +w2, 2, 1, 1, width, W_DM, W_DO, W_DS, 0, 1); 00516 inv_lift (temp+1 , b +w2, temp , 2, 1, 2, width, W_CM, W_CO, W_CS, 1, 1); 00517 00518 inv_liftS(b , temp , temp+1 , 2, 2, 2, width, W_BM, W_BO, W_BS, 0, 1); 00519 inv_lift (b+1 , temp+1 , b , 2, 2, 2, width, W_AM, W_AO, W_AS, 1, 0); 00520 #else 00521 int x; 00522 temp[0] = b[0] - ((3*b[w2]+2)>>2); 00523 for(x=1; x<(width>>1); x++){ 00524 temp[2*x ] = b[x ] - ((3*(b [x+w2-1] + b[x+w2])+4)>>3); 00525 temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x]; 00526 } 00527 if(width&1){ 00528 temp[2*x ] = b[x ] - ((3*b [x+w2-1]+2)>>2); 00529 temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x]; 00530 }else 00531 temp[2*x-1] = b[x+w2-1] - 2*temp[2*x-2]; 00532 00533 b[0] = temp[0] + ((2*temp[0] + temp[1]+4)>>3); 00534 for(x=2; x<width-1; x+=2){ 00535 b[x ] = temp[x ] + ((4*temp[x ] + temp[x-1] + temp[x+1]+8)>>4); 00536 b[x-1] = temp[x-1] + ((3*(b [x-2] + b [x ] ))>>1); 00537 } 00538 if(width&1){ 00539 b[x ] = temp[x ] + ((2*temp[x ] + temp[x-1]+4)>>3); 00540 b[x-1] = temp[x-1] + ((3*(b [x-2] + b [x ] ))>>1); 00541 }else 00542 b[x-1] = temp[x-1] + 3*b [x-2]; 00543 #endif 00544 } 00545 00546 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00547 int i; 00548 00549 for(i=0; i<width; i++){ 00550 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; 00551 } 00552 } 00553 00554 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00555 int i; 00556 00557 for(i=0; i<width; i++){ 00558 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; 00559 } 00560 } 00561 00562 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00563 int i; 00564 00565 for(i=0; i<width; i++){ 00566 #ifdef liftS 00567 b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; 00568 #else 00569 b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS; 00570 #endif 00571 } 00572 } 00573 00574 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ 00575 int i; 00576 00577 for(i=0; i<width; i++){ 00578 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; 00579 } 00580 } 00581 00582 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){ 00583 int i; 00584 00585 for(i=0; i<width; i++){ 00586 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS; 00587 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS; 00588 #ifdef liftS 00589 b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS; 00590 #else 00591 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS; 00592 #endif 00593 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; 00594 } 00595 } 00596 00597 static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){ 00598 cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line); 00599 cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line); 00600 cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line); 00601 cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line); 00602 cs->y = -3; 00603 } 00604 00605 static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){ 00606 cs->b0 = buffer + mirror(-3-1, height-1)*stride; 00607 cs->b1 = buffer + mirror(-3 , height-1)*stride; 00608 cs->b2 = buffer + mirror(-3+1, height-1)*stride; 00609 cs->b3 = buffer + mirror(-3+2, height-1)*stride; 00610 cs->y = -3; 00611 } 00612 00613 static void spatial_compose97i_dy_buffered(DWTContext *dsp, DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){ 00614 int y = cs->y; 00615 00616 IDWTELEM *b0= cs->b0; 00617 IDWTELEM *b1= cs->b1; 00618 IDWTELEM *b2= cs->b2; 00619 IDWTELEM *b3= cs->b3; 00620 IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line); 00621 IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line); 00622 00623 if(y>0 && y+4<height){ 00624 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width); 00625 }else{ 00626 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); 00627 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); 00628 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); 00629 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); 00630 } 00631 00632 if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width); 00633 if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width); 00634 00635 cs->b0=b2; 00636 cs->b1=b3; 00637 cs->b2=b4; 00638 cs->b3=b5; 00639 cs->y += 2; 00640 } 00641 00642 static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){ 00643 int y = cs->y; 00644 IDWTELEM *b0= cs->b0; 00645 IDWTELEM *b1= cs->b1; 00646 IDWTELEM *b2= cs->b2; 00647 IDWTELEM *b3= cs->b3; 00648 IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride; 00649 IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride; 00650 00651 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); 00652 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); 00653 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); 00654 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); 00655 00656 if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width); 00657 if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width); 00658 00659 cs->b0=b2; 00660 cs->b1=b3; 00661 cs->b2=b4; 00662 cs->b3=b5; 00663 cs->y += 2; 00664 } 00665 00666 static void av_unused spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){ 00667 DWTCompose cs; 00668 spatial_compose97i_init(&cs, buffer, height, stride); 00669 while(cs.y <= height) 00670 spatial_compose97i_dy(&cs, buffer, width, height, stride); 00671 } 00672 00673 void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){ 00674 int level; 00675 for(level=decomposition_count-1; level>=0; level--){ 00676 switch(type){ 00677 case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; 00678 case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; 00679 } 00680 } 00681 } 00682 00683 void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ 00684 const int support = type==1 ? 3 : 5; 00685 int level; 00686 if(type==2) return; 00687 00688 for(level=decomposition_count-1; level>=0; level--){ 00689 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ 00690 switch(type){ 00691 case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level); 00692 break; 00693 case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level); 00694 break; 00695 } 00696 } 00697 } 00698 } 00699 00700 void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ 00701 int level; 00702 for(level=decomposition_count-1; level>=0; level--){ 00703 switch(type){ 00704 case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break; 00705 case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break; 00706 } 00707 } 00708 } 00709 00710 void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){ 00711 const int support = type==1 ? 3 : 5; 00712 int level; 00713 if(type==2) return; 00714 00715 for(level=decomposition_count-1; level>=0; level--){ 00716 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ 00717 switch(type){ 00718 case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); 00719 break; 00720 case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); 00721 break; 00722 } 00723 } 00724 } 00725 } 00726 00727 void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ 00728 DWTCompose cs[MAX_DECOMPOSITIONS]; 00729 int y; 00730 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count); 00731 for(y=0; y<height; y+=4) 00732 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y); 00733 } 00734 00735 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ 00736 int s, i, j; 00737 const int dec_count= w==8 ? 3 : 4; 00738 int tmp[32*32]; 00739 int level, ori; 00740 static const int scale[2][2][4][4]={ 00741 { 00742 { 00743 // 9/7 8x8 dec=3 00744 {268, 239, 239, 213}, 00745 { 0, 224, 224, 152}, 00746 { 0, 135, 135, 110}, 00747 },{ 00748 // 9/7 16x16 or 32x32 dec=4 00749 {344, 310, 310, 280}, 00750 { 0, 320, 320, 228}, 00751 { 0, 175, 175, 136}, 00752 { 0, 129, 129, 102}, 00753 } 00754 },{ 00755 { 00756 // 5/3 8x8 dec=3 00757 {275, 245, 245, 218}, 00758 { 0, 230, 230, 156}, 00759 { 0, 138, 138, 113}, 00760 },{ 00761 // 5/3 16x16 or 32x32 dec=4 00762 {352, 317, 317, 286}, 00763 { 0, 328, 328, 233}, 00764 { 0, 180, 180, 140}, 00765 { 0, 132, 132, 105}, 00766 } 00767 } 00768 }; 00769 00770 for (i = 0; i < h; i++) { 00771 for (j = 0; j < w; j+=4) { 00772 tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4; 00773 tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4; 00774 tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4; 00775 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; 00776 } 00777 pix1 += line_size; 00778 pix2 += line_size; 00779 } 00780 00781 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); 00782 00783 s=0; 00784 assert(w==h); 00785 for(level=0; level<dec_count; level++){ 00786 for(ori= level ? 1 : 0; ori<4; ori++){ 00787 int size= w>>(dec_count-level); 00788 int sx= (ori&1) ? size : 0; 00789 int stride= 32<<(dec_count-level); 00790 int sy= (ori&2) ? stride>>1 : 0; 00791 00792 for(i=0; i<size; i++){ 00793 for(j=0; j<size; j++){ 00794 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; 00795 s += FFABS(v); 00796 } 00797 } 00798 } 00799 } 00800 assert(s>=0); 00801 return s>>9; 00802 } 00803 00804 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00805 return w_c(v, pix1, pix2, line_size, 8, h, 1); 00806 } 00807 00808 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00809 return w_c(v, pix1, pix2, line_size, 8, h, 0); 00810 } 00811 00812 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00813 return w_c(v, pix1, pix2, line_size, 16, h, 1); 00814 } 00815 00816 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00817 return w_c(v, pix1, pix2, line_size, 16, h, 0); 00818 } 00819 00820 int ff_w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00821 return w_c(v, pix1, pix2, line_size, 32, h, 1); 00822 } 00823 00824 int ff_w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ 00825 return w_c(v, pix1, pix2, line_size, 32, h, 0); 00826 } 00827 00828 void ff_dsputil_init_dwt(DSPContext *c) 00829 { 00830 c->w53[0]= w53_16_c; 00831 c->w53[1]= w53_8_c; 00832 c->w97[0]= w97_16_c; 00833 c->w97[1]= w97_8_c; 00834 } 00835 00836 void ff_dwt_init(DWTContext *c) 00837 { 00838 c->vertical_compose97i = ff_snow_vertical_compose97i; 00839 c->horizontal_compose97i = ff_snow_horizontal_compose97i; 00840 c->inner_add_yblock = ff_snow_inner_add_yblock; 00841 00842 if (HAVE_MMX) ff_dwt_init_x86(c); 00843 }