Sascha Haeberling | 8bddf8c | 2013-08-14 11:20:34 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (C) 2011 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | /*$Id: db_feature_detection.cpp,v 1.4 2011/06/17 14:03:30 mbansal Exp $*/ |
| 18 | |
| 19 | /***************************************************************** |
| 20 | * Lean and mean begins here * |
| 21 | *****************************************************************/ |
| 22 | |
| 23 | #include "db_utilities.h" |
| 24 | #include "db_feature_detection.h" |
| 25 | #ifdef _VERBOSE_ |
| 26 | #include <iostream> |
| 27 | #endif |
| 28 | #include <float.h> |
| 29 | |
| 30 | #define DB_SUB_PIXEL |
| 31 | |
| 32 | #define BORDER 10 // 5 |
| 33 | |
| 34 | float** db_AllocStrengthImage_f(float **im,int w,int h) |
| 35 | { |
| 36 | int i,n,aw; |
| 37 | long c,size; |
| 38 | float **img,*aim,*p; |
| 39 | |
| 40 | /*Determine number of 124 element chunks needed*/ |
| 41 | n=(db_maxi(1,w-6)+123)/124; |
| 42 | /*Determine the total allocation width aw*/ |
| 43 | aw=n*124+8; |
| 44 | /*Allocate*/ |
| 45 | size=aw*h+16; |
| 46 | *im=new float [size]; |
| 47 | /*Clean up*/ |
| 48 | p=(*im); |
| 49 | for(c=0;c<size;c++) p[c]=0.0; |
| 50 | /*Get a 16 byte aligned pointer*/ |
| 51 | aim=db_AlignPointer_f(*im,16); |
| 52 | /*Allocate pointer table*/ |
| 53 | img=new float* [h]; |
| 54 | /*Initialize the pointer table*/ |
| 55 | for(i=0;i<h;i++) |
| 56 | { |
| 57 | img[i]=aim+aw*i+1; |
| 58 | } |
| 59 | |
| 60 | return(img); |
| 61 | } |
| 62 | |
| 63 | void db_FreeStrengthImage_f(float *im,float **img,int h) |
| 64 | { |
| 65 | delete [] im; |
| 66 | delete [] img; |
| 67 | } |
| 68 | |
| 69 | /*Compute derivatives Ix,Iy for a subrow of img with upper left (i,j) and width chunk_width |
| 70 | Memory references occur one pixel outside the subrow*/ |
| 71 | inline void db_IxIyRow_f(float *Ix,float *Iy,const float * const *img,int i,int j,int chunk_width) |
| 72 | { |
| 73 | int c; |
| 74 | |
| 75 | for(c=0;c<chunk_width;c++) |
| 76 | { |
| 77 | Ix[c]=img[i][j+c-1]-img[i][j+c+1]; |
| 78 | Iy[c]=img[i-1][j+c]-img[i+1][j+c]; |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | /*Compute derivatives Ix,Iy for a subrow of img with upper left (i,j) and width 128 |
| 83 | Memory references occur one pixel outside the subrow*/ |
| 84 | inline void db_IxIyRow_u(int *dxx,const unsigned char * const *img,int i,int j,int nc) |
| 85 | { |
| 86 | #ifdef DB_USE_MMX |
| 87 | const unsigned char *r1,*r2,*r3; |
| 88 | |
| 89 | r1=img[i-1]+j; r2=img[i]+j; r3=img[i+1]+j; |
| 90 | |
| 91 | _asm |
| 92 | { |
| 93 | mov esi,16 |
| 94 | mov eax,r1 |
| 95 | mov ebx,r2 |
| 96 | mov ecx,r3 |
| 97 | mov edx,dxx |
| 98 | |
| 99 | /*Get bitmask into mm7*/ |
| 100 | mov edi,7F7F7F7Fh |
| 101 | movd mm7,edi |
| 102 | punpckldq mm7,mm7 |
| 103 | |
| 104 | loopstart: |
| 105 | /***************dx part 1-12*********************************/ |
| 106 | movq mm0,[eax] /*1 Get upper*/ |
| 107 | pxor mm6,mm6 /*2 Set to zero*/ |
| 108 | movq mm1,[ecx] /*3 Get lower*/ |
| 109 | psrlq mm0,1 /*4 Shift*/ |
| 110 | psrlq mm1,1 /*5 Shift*/ |
| 111 | pand mm0,mm7 /*6 And*/ |
| 112 | movq mm2,[ebx-1] /*13 Get left*/ |
| 113 | pand mm1,mm7 /*7 And*/ |
| 114 | psubb mm0,mm1 /*8 Subtract*/ |
| 115 | pxor mm5,mm5 /*14 Set to zero*/ |
| 116 | movq mm1,mm0 /*9 Copy*/ |
| 117 | pcmpgtb mm6,mm0 /*10 Create unpack mask*/ |
| 118 | movq mm3,[ebx+1] /*15 Get right*/ |
| 119 | punpcklbw mm0,mm6 /*11 Unpack low*/ |
| 120 | punpckhbw mm1,mm6 /*12 Unpack high*/ |
| 121 | /***************dy part 13-24*********************************/ |
| 122 | movq mm4,mm0 /*25 Copy dx*/ |
| 123 | psrlq mm2,1 /*16 Shift*/ |
| 124 | pmullw mm0,mm0 /*26 Multiply dx*dx*/ |
| 125 | psrlq mm3,1 /*17 Shift*/ |
| 126 | pand mm2,mm7 /*18 And*/ |
| 127 | pand mm3,mm7 /*19 And*/ |
| 128 | /*Stall*/ |
| 129 | psubb mm2,mm3 /*20 Subtract*/ |
| 130 | /*Stall*/ |
| 131 | movq mm3,mm2 /*21 Copy*/ |
| 132 | pcmpgtb mm5,mm2 /*22 Create unpack mask*/ |
| 133 | punpcklbw mm2,mm5 /*23 Unpack low*/ |
| 134 | /*Stall*/ |
| 135 | punpckhbw mm3,mm5 /*24 Unpack high*/ |
| 136 | /***************dxx dxy dyy low part 25-49*********************************/ |
| 137 | pmullw mm4,mm2 /*27 Multiply dx*dy*/ |
| 138 | pmullw mm2,mm2 /*28 Multiply dy*dy*/ |
| 139 | pxor mm6,mm6 /*29 Set to zero*/ |
| 140 | movq mm5,mm0 /*30 Copy dx*dx*/ |
| 141 | pcmpgtw mm6,mm0 /*31 Create unpack mask for dx*dx*/ |
| 142 | punpcklwd mm0,mm6 /*32 Unpack dx*dx lows*/ |
| 143 | /*Stall*/ |
| 144 | punpckhwd mm5,mm6 /*33 Unpack dx*dx highs*/ |
| 145 | pxor mm6,mm6 /*36 Set to zero*/ |
| 146 | movq [edx],mm0 /*34 Store dx*dx lows*/ |
| 147 | movq mm0,mm4 /*37 Copy dx*dy*/ |
| 148 | movq [edx+8],mm5 /*35 Store dx*dx highs*/ |
| 149 | pcmpgtw mm6,mm4 /*38 Create unpack mask for dx*dy*/ |
| 150 | punpcklwd mm4,mm6 /*39 Unpack dx*dy lows*/ |
| 151 | /*Stall*/ |
| 152 | punpckhwd mm0,mm6 /*40 Unpack dx*dy highs*/ |
| 153 | pxor mm6,mm6 /*43 Set to zero*/ |
| 154 | movq [edx+512],mm4 /*41 Store dx*dy lows*/ |
| 155 | movq mm5,mm2 /*44 Copy dy*dy*/ |
| 156 | movq [edx+520],mm0 /*42 Store dx*dy highs*/ |
| 157 | pcmpgtw mm6,mm2 /*45 Create unpack mask for dy*dy*/ |
| 158 | punpcklwd mm2,mm6 /*46 Unpack dy*dy lows*/ |
| 159 | movq mm4,mm1 /*50 Copy dx*/ |
| 160 | punpckhwd mm5,mm6 /*47 Unpack dy*dy highs*/ |
| 161 | pmullw mm1,mm1 /*51 Multiply dx*dx*/ |
| 162 | movq [edx+1024],mm2 /*48 Store dy*dy lows*/ |
| 163 | pmullw mm4,mm3 /*52 Multiply dx*dy*/ |
| 164 | movq [edx+1032],mm5 /*49 Store dy*dy highs*/ |
| 165 | /***************dxx dxy dyy high part 50-79*********************************/ |
| 166 | pmullw mm3,mm3 /*53 Multiply dy*dy*/ |
| 167 | pxor mm6,mm6 /*54 Set to zero*/ |
| 168 | movq mm5,mm1 /*55 Copy dx*dx*/ |
| 169 | pcmpgtw mm6,mm1 /*56 Create unpack mask for dx*dx*/ |
| 170 | pxor mm2,mm2 /*61 Set to zero*/ |
| 171 | punpcklwd mm1,mm6 /*57 Unpack dx*dx lows*/ |
| 172 | movq mm0,mm4 /*62 Copy dx*dy*/ |
| 173 | punpckhwd mm5,mm6 /*58 Unpack dx*dx highs*/ |
| 174 | pcmpgtw mm2,mm4 /*63 Create unpack mask for dx*dy*/ |
| 175 | movq [edx+16],mm1 /*59 Store dx*dx lows*/ |
| 176 | punpcklwd mm4,mm2 /*64 Unpack dx*dy lows*/ |
| 177 | movq [edx+24],mm5 /*60 Store dx*dx highs*/ |
| 178 | punpckhwd mm0,mm2 /*65 Unpack dx*dy highs*/ |
| 179 | movq [edx+528],mm4 /*66 Store dx*dy lows*/ |
| 180 | pxor mm6,mm6 /*68 Set to zero*/ |
| 181 | movq [edx+536],mm0 /*67 Store dx*dy highs*/ |
| 182 | movq mm5,mm3 /*69 Copy dy*dy*/ |
| 183 | pcmpgtw mm6,mm3 /*70 Create unpack mask for dy*dy*/ |
| 184 | add eax,8 /*75*/ |
| 185 | punpcklwd mm3,mm6 /*71 Unpack dy*dy lows*/ |
| 186 | add ebx,8 /*76*/ |
| 187 | punpckhwd mm5,mm6 /*72 Unpack dy*dy highs*/ |
| 188 | add ecx,8 /*77*/ |
| 189 | movq [edx+1040],mm3 /*73 Store dy*dy lows*/ |
| 190 | /*Stall*/ |
| 191 | movq [edx+1048],mm5 /*74 Store dy*dy highs*/ |
| 192 | /*Stall*/ |
| 193 | add edx,32 /*78*/ |
| 194 | dec esi /*79*/ |
| 195 | jnz loopstart |
| 196 | |
| 197 | emms |
| 198 | } |
| 199 | |
| 200 | #else |
| 201 | int c; |
| 202 | int Ix,Iy; |
| 203 | |
| 204 | for(c=0;c<nc;c++) |
| 205 | { |
| 206 | Ix=(img[i][j+c-1]-img[i][j+c+1])>>1; |
| 207 | Iy=(img[i-1][j+c]-img[i+1][j+c])>>1; |
| 208 | dxx[c]=Ix*Ix; |
| 209 | dxx[c+128]=Ix*Iy; |
| 210 | dxx[c+256]=Iy*Iy; |
| 211 | } |
| 212 | #endif /*DB_USE_MMX*/ |
| 213 | } |
| 214 | |
| 215 | /*Filter vertically five rows of derivatives of length chunk_width into gxx,gxy,gyy*/ |
| 216 | inline void db_gxx_gxy_gyy_row_f(float *gxx,float *gxy,float *gyy,int chunk_width, |
| 217 | float *Ix0,float *Ix1,float *Ix2,float *Ix3,float *Ix4, |
| 218 | float *Iy0,float *Iy1,float *Iy2,float *Iy3,float *Iy4) |
| 219 | { |
| 220 | int c; |
| 221 | float dx,dy; |
| 222 | float Ixx0,Ixy0,Iyy0,Ixx1,Ixy1,Iyy1,Ixx2,Ixy2,Iyy2,Ixx3,Ixy3,Iyy3,Ixx4,Ixy4,Iyy4; |
| 223 | |
| 224 | for(c=0;c<chunk_width;c++) |
| 225 | { |
| 226 | dx=Ix0[c]; |
| 227 | dy=Iy0[c]; |
| 228 | Ixx0=dx*dx; |
| 229 | Ixy0=dx*dy; |
| 230 | Iyy0=dy*dy; |
| 231 | |
| 232 | dx=Ix1[c]; |
| 233 | dy=Iy1[c]; |
| 234 | Ixx1=dx*dx; |
| 235 | Ixy1=dx*dy; |
| 236 | Iyy1=dy*dy; |
| 237 | |
| 238 | dx=Ix2[c]; |
| 239 | dy=Iy2[c]; |
| 240 | Ixx2=dx*dx; |
| 241 | Ixy2=dx*dy; |
| 242 | Iyy2=dy*dy; |
| 243 | |
| 244 | dx=Ix3[c]; |
| 245 | dy=Iy3[c]; |
| 246 | Ixx3=dx*dx; |
| 247 | Ixy3=dx*dy; |
| 248 | Iyy3=dy*dy; |
| 249 | |
| 250 | dx=Ix4[c]; |
| 251 | dy=Iy4[c]; |
| 252 | Ixx4=dx*dx; |
| 253 | Ixy4=dx*dy; |
| 254 | Iyy4=dy*dy; |
| 255 | |
| 256 | /*Filter vertically*/ |
| 257 | gxx[c]=Ixx0+Ixx1*4.0f+Ixx2*6.0f+Ixx3*4.0f+Ixx4; |
| 258 | gxy[c]=Ixy0+Ixy1*4.0f+Ixy2*6.0f+Ixy3*4.0f+Ixy4; |
| 259 | gyy[c]=Iyy0+Iyy1*4.0f+Iyy2*6.0f+Iyy3*4.0f+Iyy4; |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | /*Filter vertically five rows of derivatives of length 128 into gxx,gxy,gyy*/ |
| 264 | inline void db_gxx_gxy_gyy_row_s(int *g,int *d0,int *d1,int *d2,int *d3,int *d4,int nc) |
| 265 | { |
| 266 | #ifdef DB_USE_MMX |
| 267 | int c; |
| 268 | |
| 269 | _asm |
| 270 | { |
| 271 | mov c,64 |
| 272 | mov eax,d0 |
| 273 | mov ebx,d1 |
| 274 | mov ecx,d2 |
| 275 | mov edx,d3 |
| 276 | mov edi,d4 |
| 277 | mov esi,g |
| 278 | |
| 279 | loopstart: |
| 280 | /***************dxx part 1-14*********************************/ |
| 281 | movq mm0,[eax] /*1 Get dxx0*/ |
| 282 | /*Stall*/ |
| 283 | movq mm1,[ebx] /*2 Get dxx1*/ |
| 284 | /*Stall*/ |
| 285 | movq mm2,[ecx] /*5 Get dxx2*/ |
| 286 | pslld mm1,2 /*3 Shift dxx1*/ |
| 287 | movq mm3,[edx] /*10 Get dxx3*/ |
| 288 | paddd mm0,mm1 /*4 Accumulate dxx1*/ |
| 289 | movq mm4,[eax+512] /*15 Get dxy0*/ |
| 290 | pslld mm2,1 /*6 Shift dxx2 1*/ |
| 291 | paddd mm0,mm2 /*7 Accumulate dxx2 1*/ |
| 292 | pslld mm2,1 /*8 Shift dxx2 2*/ |
| 293 | movq mm5,[ebx+512] /*16 Get dxy1*/ |
| 294 | paddd mm0,mm2 /*9 Accumulate dxx2 2*/ |
| 295 | pslld mm3,2 /*11 Shift dxx3*/ |
| 296 | /*Stall*/ |
| 297 | paddd mm0,mm3 /*12 Accumulate dxx3*/ |
| 298 | pslld mm5,2 /*17 Shift dxy1*/ |
| 299 | paddd mm0,[edi] /*13 Accumulate dxx4*/ |
| 300 | paddd mm4,mm5 /*18 Accumulate dxy1*/ |
| 301 | movq mm6,[ecx+512] /*19 Get dxy2*/ |
| 302 | /*Stall*/ |
| 303 | movq [esi],mm0 /*14 Store dxx sums*/ |
| 304 | /***************dxy part 15-28*********************************/ |
| 305 | pslld mm6,1 /*20 Shift dxy2 1*/ |
| 306 | paddd mm4,mm6 /*21 Accumulate dxy2 1*/ |
| 307 | pslld mm6,1 /*22 Shift dxy2 2*/ |
| 308 | movq mm0,[eax+1024] /*29 Get dyy0*/ |
| 309 | paddd mm4,mm6 /*23 Accumulate dxy2 2*/ |
| 310 | movq mm7,[edx+512] /*24 Get dxy3*/ |
| 311 | pslld mm7,2 /*25 Shift dxy3*/ |
| 312 | movq mm1,[ebx+1024] /*30 Get dyy1*/ |
| 313 | paddd mm4,mm7 /*26 Accumulate dxy3*/ |
| 314 | paddd mm4,[edi+512] /*27 Accumulate dxy4*/ |
| 315 | pslld mm1,2 /*31 Shift dyy1*/ |
| 316 | movq mm2,[ecx+1024] /*33 Get dyy2*/ |
| 317 | paddd mm0,mm1 /*32 Accumulate dyy1*/ |
| 318 | movq [esi+512],mm4 /*28 Store dxy sums*/ |
| 319 | pslld mm2,1 /*34 Shift dyy2 1*/ |
| 320 | /***************dyy part 29-49*********************************/ |
| 321 | |
| 322 | |
| 323 | movq mm3,[edx+1024] /*38 Get dyy3*/ |
| 324 | paddd mm0,mm2 /*35 Accumulate dyy2 1*/ |
| 325 | paddd mm0,[edi+1024] /*41 Accumulate dyy4*/ |
| 326 | pslld mm2,1 /*36 Shift dyy2 2*/ |
| 327 | paddd mm0,mm2 /*37 Accumulate dyy2 2*/ |
| 328 | pslld mm3,2 /*39 Shift dyy3*/ |
| 329 | paddd mm0,mm3 /*40 Accumulate dyy3*/ |
| 330 | add eax,8 /*43*/ |
| 331 | add ebx,8 /*44*/ |
| 332 | add ecx,8 /*45*/ |
| 333 | movq [esi+1024],mm0 /*42 Store dyy sums*/ |
| 334 | /*Stall*/ |
| 335 | add edx,8 /*46*/ |
| 336 | add edi,8 /*47*/ |
| 337 | add esi,8 /*48*/ |
| 338 | dec c /*49*/ |
| 339 | jnz loopstart |
| 340 | |
| 341 | emms |
| 342 | } |
| 343 | |
| 344 | #else |
| 345 | int c,dd; |
| 346 | |
| 347 | for(c=0;c<nc;c++) |
| 348 | { |
| 349 | /*Filter vertically*/ |
| 350 | dd=d2[c]; |
| 351 | g[c]=d0[c]+(d1[c]<<2)+(dd<<2)+(dd<<1)+(d3[c]<<2)+d4[c]; |
| 352 | |
| 353 | dd=d2[c+128]; |
| 354 | g[c+128]=d0[c+128]+(d1[c+128]<<2)+(dd<<2)+(dd<<1)+(d3[c+128]<<2)+d4[c+128]; |
| 355 | |
| 356 | dd=d2[c+256]; |
| 357 | g[c+256]=d0[c+256]+(d1[c+256]<<2)+(dd<<2)+(dd<<1)+(d3[c+256]<<2)+d4[c+256]; |
| 358 | } |
| 359 | #endif /*DB_USE_MMX*/ |
| 360 | } |
| 361 | |
| 362 | /*Filter horizontally the three rows gxx,gxy,gyy into the strength subrow starting at i,j |
| 363 | and with width chunk_width. gxx,gxy and gyy are assumed to be four pixels wider than chunk_width |
| 364 | and starting at (i,j-2)*/ |
| 365 | inline void db_HarrisStrength_row_f(float **s,float *gxx,float *gxy,float *gyy,int i,int j,int chunk_width) |
| 366 | { |
| 367 | float Gxx,Gxy,Gyy,det,trc; |
| 368 | int c; |
| 369 | |
| 370 | for(c=0;c<chunk_width;c++) |
| 371 | { |
| 372 | Gxx=gxx[c]+gxx[c+1]*4.0f+gxx[c+2]*6.0f+gxx[c+3]*4.0f+gxx[c+4]; |
| 373 | Gxy=gxy[c]+gxy[c+1]*4.0f+gxy[c+2]*6.0f+gxy[c+3]*4.0f+gxy[c+4]; |
| 374 | Gyy=gyy[c]+gyy[c+1]*4.0f+gyy[c+2]*6.0f+gyy[c+3]*4.0f+gyy[c+4]; |
| 375 | |
| 376 | det=Gxx*Gyy-Gxy*Gxy; |
| 377 | trc=Gxx+Gyy; |
| 378 | s[i][j+c]=det-0.06f*trc*trc; |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | /*Filter g of length 128 in place with 14641. Output is shifted two steps |
| 383 | and of length 124*/ |
| 384 | inline void db_Filter14641_128_i(int *g,int nc) |
| 385 | { |
| 386 | #ifdef DB_USE_MMX |
| 387 | int mask; |
| 388 | |
| 389 | mask=0xFFFFFFFF; |
| 390 | _asm |
| 391 | { |
| 392 | mov esi,31 |
| 393 | mov eax,g |
| 394 | |
| 395 | /*Get bitmask 00000000FFFFFFFF into mm7*/ |
| 396 | movd mm7,mask |
| 397 | |
| 398 | /*Warming iteration one 1-16********************/ |
| 399 | movq mm6,[eax] /*1 Load new data*/ |
| 400 | paddd mm0,mm6 /*2 Add 1* behind two steps*/ |
| 401 | movq mm2,mm6 /*3 Start with 1* in front two steps*/ |
| 402 | pslld mm6,1 /*4*/ |
| 403 | paddd mm1,mm6 /*5 Add 2* same place*/ |
| 404 | pslld mm6,1 /*6*/ |
| 405 | paddd mm1,mm6 /*7 Add 4* same place*/ |
| 406 | pshufw mm6,mm6,4Eh /*8 Swap the two double-words using bitmask 01001110=4Eh*/ |
| 407 | paddd mm1,mm6 /*9 Add 4* swapped*/ |
| 408 | movq mm5,mm6 /*10 Copy*/ |
| 409 | pand mm6,mm7 /*11 Get low double-word only*/ |
| 410 | paddd mm2,mm6 /*12 Add 4* in front one step*/ |
| 411 | pxor mm6,mm5 /*13 Get high double-word only*/ |
| 412 | paddd mm0,mm6 /*14 Add 4* behind one step*/ |
| 413 | movq mm0,mm1 /*15 Shift along*/ |
| 414 | movq mm1,mm2 /*16 Shift along*/ |
| 415 | /*Warming iteration two 17-32********************/ |
| 416 | movq mm4,[eax+8] /*17 Load new data*/ |
| 417 | paddd mm0,mm4 /*18 Add 1* behind two steps*/ |
| 418 | movq mm2,mm4 /*19 Start with 1* in front two steps*/ |
| 419 | pslld mm4,1 /*20*/ |
| 420 | paddd mm1,mm4 /*21 Add 2* same place*/ |
| 421 | pslld mm4,1 /*22*/ |
| 422 | paddd mm1,mm4 /*23 Add 4* same place*/ |
| 423 | pshufw mm4,mm4,4Eh /*24 Swap the two double-words using bitmask 01001110=4Eh*/ |
| 424 | paddd mm1,mm4 /*25 Add 4* swapped*/ |
| 425 | movq mm3,mm4 /*26 Copy*/ |
| 426 | pand mm4,mm7 /*27 Get low double-word only*/ |
| 427 | paddd mm2,mm4 /*28 Add 4* in front one step*/ |
| 428 | pxor mm4,mm3 /*29 Get high double-word only*/ |
| 429 | paddd mm0,mm4 /*30 Add 4* behind one step*/ |
| 430 | movq mm0,mm1 /*31 Shift along*/ |
| 431 | movq mm1,mm2 /*32 Shift along*/ |
| 432 | |
| 433 | /*Loop********************/ |
| 434 | loopstart: |
| 435 | /*First part of loop 33-47********/ |
| 436 | movq mm6,[eax+16] /*33 Load new data*/ |
| 437 | /*Stall*/ |
| 438 | paddd mm0,mm6 /*34 Add 1* behind two steps*/ |
| 439 | movq mm2,mm6 /*35 Start with 1* in front two steps*/ |
| 440 | movq mm4,[eax+24] /*48 Load new data*/ |
| 441 | pslld mm6,1 /*36*/ |
| 442 | paddd mm1,mm6 /*37 Add 2* same place*/ |
| 443 | pslld mm6,1 /*38*/ |
| 444 | paddd mm1,mm6 /*39 Add 4* same place*/ |
| 445 | pshufw mm6,mm6,4Eh /*40 Swap the two double-words using bitmask 01001110=4Eh*/ |
| 446 | paddd mm1,mm4 /*49 Add 1* behind two steps*/ |
| 447 | movq mm5,mm6 /*41 Copy*/ |
| 448 | paddd mm1,mm6 /*42 Add 4* swapped*/ |
| 449 | pand mm6,mm7 /*43 Get low double-word only*/ |
| 450 | paddd mm2,mm6 /*44 Add 4* in front one step*/ |
| 451 | pxor mm6,mm5 /*45 Get high double-word only*/ |
| 452 | paddd mm0,mm6 /*46 Add 4* behind one step*/ |
| 453 | movq mm6,mm4 /*50a Copy*/ |
| 454 | pslld mm4,1 /*51*/ |
| 455 | /*Stall*/ |
| 456 | movq [eax],mm0 /*47 Store result two steps behind*/ |
| 457 | /*Second part of loop 48-66********/ |
| 458 | movq mm0,mm6 /*50b Start with 1* in front two steps*/ |
| 459 | paddd mm2,mm4 /*52 Add 2* same place*/ |
| 460 | pslld mm4,1 /*53*/ |
| 461 | paddd mm2,mm4 /*54 Add 4* same place*/ |
| 462 | pshufw mm4,mm4,4Eh /*55 Swap the two double-words using bitmask 01001110=4Eh*/ |
| 463 | paddd mm2,mm4 /*56 Add 4* swapped*/ |
| 464 | movq mm3,mm4 /*57 Copy*/ |
| 465 | pand mm4,mm7 /*58 Get low double-word only*/ |
| 466 | /*Stall*/ |
| 467 | paddd mm0,mm4 /*59 Add 4* in front one step*/ |
| 468 | pxor mm4,mm3 /*60 Get high double-word only*/ |
| 469 | paddd mm1,mm4 /*61 Add 4* behind one step*/ |
| 470 | add eax,16 /*65*/ |
| 471 | dec esi /*66*/ |
| 472 | /*Stall*/ |
| 473 | movq [eax-8],mm1 /*62 Store result two steps behind*/ |
| 474 | movq mm1,mm0 /*63 Shift along*/ |
| 475 | movq mm0,mm2 /*64 Shift along*/ |
| 476 | jnz loopstart |
| 477 | |
| 478 | emms |
| 479 | } |
| 480 | |
| 481 | #else |
| 482 | int c; |
| 483 | |
| 484 | for(c=0;c<nc-4;c++) |
| 485 | { |
| 486 | g[c]=g[c]+(g[c+1]<<2)+(g[c+2]<<2)+(g[c+2]<<1)+(g[c+3]<<2)+g[c+4]; |
| 487 | } |
| 488 | #endif /*DB_USE_MMX*/ |
| 489 | } |
| 490 | |
| 491 | /*Filter horizontally the three rows gxx,gxy,gyy of length 128 into the strength subrow s |
| 492 | of length 124. gxx,gxy and gyy are assumed to be starting at (i,j-2) if s[i][j] is sought. |
| 493 | s should be 16 byte aligned*/ |
| 494 | inline void db_HarrisStrength_row_s(float *s,int *gxx,int *gxy,int *gyy,int nc) |
| 495 | { |
| 496 | float k; |
| 497 | |
| 498 | k=0.06f; |
| 499 | |
| 500 | db_Filter14641_128_i(gxx,nc); |
| 501 | db_Filter14641_128_i(gxy,nc); |
| 502 | db_Filter14641_128_i(gyy,nc); |
| 503 | |
| 504 | #ifdef DB_USE_SIMD |
| 505 | |
| 506 | |
| 507 | _asm |
| 508 | { |
| 509 | mov esi,15 |
| 510 | mov eax,gxx |
| 511 | mov ebx,gxy |
| 512 | mov ecx,gyy |
| 513 | mov edx,s |
| 514 | |
| 515 | /*broadcast k to all positions of xmm7*/ |
| 516 | movss xmm7,k |
| 517 | shufps xmm7,xmm7,0 |
| 518 | |
| 519 | /*****Warm up 1-10**************************************/ |
| 520 | cvtpi2ps xmm0,[eax+8] /*1 Convert two integers into floating point of low double-word*/ |
| 521 | /*Stall*/ |
| 522 | cvtpi2ps xmm1,[ebx+8] /*4 Convert two integers into floating point of low double-word*/ |
| 523 | movlhps xmm0,xmm0 /*2 Move them to the high double-word*/ |
| 524 | cvtpi2ps xmm2,[ecx+8] /*7 Convert two integers into floating point of low double-word*/ |
| 525 | movlhps xmm1,xmm1 /*5 Move them to the high double-word*/ |
| 526 | cvtpi2ps xmm0,[eax] /*3 Convert two integers into floating point of low double-word*/ |
| 527 | movlhps xmm2,xmm2 /*8 Move them to the high double-word*/ |
| 528 | cvtpi2ps xmm1,[ebx] /*6 Convert two integers into floating point of low double-word*/ |
| 529 | movaps xmm3,xmm0 /*10 Copy Cxx*/ |
| 530 | cvtpi2ps xmm2,[ecx] /*9 Convert two integers into floating point of low double-word*/ |
| 531 | /*Stall*/ |
| 532 | loopstart: |
| 533 | /*****First part of loop 11-18***********************/ |
| 534 | mulps xmm0,xmm2 /*11 Multiply to get Gxx*Gyy*/ |
| 535 | addps xmm2,xmm3 /*12 Add to get Gxx+Gyy*/ |
| 536 | cvtpi2ps xmm4,[eax+24] /*19 Convert two integers into floating point of low double-word*/ |
| 537 | mulps xmm1,xmm1 /*13 Multiply to get Gxy*Gxy*/ |
| 538 | mulps xmm2,xmm2 /*14 Multiply to get (Gxx+Gyy)*(Gxx+Gyy)*/ |
| 539 | movlhps xmm4,xmm4 /*20 Move them to the high double-word*/ |
| 540 | cvtpi2ps xmm4,[eax+16] /*21 Convert two integers into floating point of low double-word*/ |
| 541 | /*Stall*/ |
| 542 | subps xmm0,xmm1 /*15 Subtract to get Gxx*Gyy-Gxy*Gxy*/ |
| 543 | mulps xmm2,xmm7 /*16 Multiply to get k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 544 | cvtpi2ps xmm5,[ebx+24] /*22 Convert two integers into floating point of low double-word*/ |
| 545 | /*Stall*/ |
| 546 | movlhps xmm5,xmm5 /*23 Move them to the high double-word*/ |
| 547 | /*Stall*/ |
| 548 | cvtpi2ps xmm5,[ebx+16] /*24 Convert two integers into floating point of low double-word*/ |
| 549 | subps xmm0,xmm2 /*17 Subtract to get Gxx*Gyy-Gxy*Gxy-k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 550 | cvtpi2ps xmm6,[ecx+24] /*25 Convert two integers into floating point of low double-word*/ |
| 551 | /*Stall*/ |
| 552 | movaps [edx],xmm0 /*18 Store*/ |
| 553 | /*****Second part of loop 26-40***********************/ |
| 554 | movlhps xmm6,xmm6 /*26 Move them to the high double-word*/ |
| 555 | cvtpi2ps xmm6,[ecx+16] /*27 Convert two integers into floating point of low double-word*/ |
| 556 | movaps xmm3,xmm4 /*28 Copy Cxx*/ |
| 557 | mulps xmm4,xmm6 /*29 Multiply to get Gxx*Gyy*/ |
| 558 | addps xmm6,xmm3 /*30 Add to get Gxx+Gyy*/ |
| 559 | cvtpi2ps xmm0,[eax+40] /*(1 Next) Convert two integers into floating point of low double-word*/ |
| 560 | mulps xmm5,xmm5 /*31 Multiply to get Gxy*Gxy*/ |
| 561 | cvtpi2ps xmm1,[ebx+40] /*(4 Next) Convert two integers into floating point of low double-word*/ |
| 562 | mulps xmm6,xmm6 /*32 Multiply to get (Gxx+Gyy)*(Gxx+Gyy)*/ |
| 563 | cvtpi2ps xmm2,[ecx+40] /*(7 Next) Convert two integers into floating point of low double-word*/ |
| 564 | movlhps xmm0,xmm0 /*(2 Next) Move them to the high double-word*/ |
| 565 | subps xmm4,xmm5 /*33 Subtract to get Gxx*Gyy-Gxy*Gxy*/ |
| 566 | movlhps xmm1,xmm1 /*(5 Next) Move them to the high double-word*/ |
| 567 | cvtpi2ps xmm0,[eax+32] /*(3 Next)Convert two integers into floating point of low double-word*/ |
| 568 | mulps xmm6,xmm7 /*34 Multiply to get k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 569 | cvtpi2ps xmm1,[ebx+32] /*(6 Next) Convert two integers into floating point of low double-word*/ |
| 570 | movlhps xmm2,xmm2 /*(8 Next) Move them to the high double-word*/ |
| 571 | movaps xmm3,xmm0 /*(10 Next) Copy Cxx*/ |
| 572 | add eax,32 /*37*/ |
| 573 | subps xmm4,xmm6 /*35 Subtract to get Gxx*Gyy-Gxy*Gxy-k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 574 | add ebx,32 /*38*/ |
| 575 | cvtpi2ps xmm2,[ecx+32] /*(9 Next) Convert two integers into floating point of low double-word*/ |
| 576 | /*Stall*/ |
| 577 | movaps [edx+16],xmm4 /*36 Store*/ |
| 578 | /*Stall*/ |
| 579 | add ecx,32 /*39*/ |
| 580 | add edx,32 /*40*/ |
| 581 | dec esi /*41*/ |
| 582 | jnz loopstart |
| 583 | |
| 584 | /****Cool down***************/ |
| 585 | mulps xmm0,xmm2 /*Multiply to get Gxx*Gyy*/ |
| 586 | addps xmm2,xmm3 /*Add to get Gxx+Gyy*/ |
| 587 | mulps xmm1,xmm1 /*Multiply to get Gxy*Gxy*/ |
| 588 | mulps xmm2,xmm2 /*Multiply to get (Gxx+Gyy)*(Gxx+Gyy)*/ |
| 589 | subps xmm0,xmm1 /*Subtract to get Gxx*Gyy-Gxy*Gxy*/ |
| 590 | mulps xmm2,xmm7 /*Multiply to get k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 591 | subps xmm0,xmm2 /*Subtract to get Gxx*Gyy-Gxy*Gxy-k*(Gxx+Gyy)*(Gxx+Gyy)*/ |
| 592 | movaps [edx],xmm0 /*Store*/ |
| 593 | } |
| 594 | |
| 595 | #else |
| 596 | float Gxx,Gxy,Gyy,det,trc; |
| 597 | int c; |
| 598 | |
| 599 | //for(c=0;c<124;c++) |
| 600 | for(c=0;c<nc-4;c++) |
| 601 | { |
| 602 | Gxx=(float)gxx[c]; |
| 603 | Gxy=(float)gxy[c]; |
| 604 | Gyy=(float)gyy[c]; |
| 605 | |
| 606 | det=Gxx*Gyy-Gxy*Gxy; |
| 607 | trc=Gxx+Gyy; |
| 608 | s[c]=det-k*trc*trc; |
| 609 | } |
| 610 | #endif /*DB_USE_SIMD*/ |
| 611 | } |
| 612 | |
| 613 | /*Compute the Harris corner strength of the chunk [left,top,right,bottom] of img and |
| 614 | store it into the corresponding region of s. left and top have to be at least 3 and |
| 615 | right and bottom have to be at most width-4,height-4*/ |
| 616 | inline void db_HarrisStrengthChunk_f(float **s,const float * const *img,int left,int top,int right,int bottom, |
| 617 | /*temp should point to at least |
| 618 | 13*(right-left+5) of allocated memory*/ |
| 619 | float *temp) |
| 620 | { |
| 621 | float *Ix[5],*Iy[5]; |
| 622 | float *gxx,*gxy,*gyy; |
| 623 | int i,chunk_width,chunk_width_p4; |
| 624 | |
| 625 | chunk_width=right-left+1; |
| 626 | chunk_width_p4=chunk_width+4; |
| 627 | gxx=temp; |
| 628 | gxy=gxx+chunk_width_p4; |
| 629 | gyy=gxy+chunk_width_p4; |
| 630 | for(i=0;i<5;i++) |
| 631 | { |
| 632 | Ix[i]=gyy+chunk_width_p4+(2*i*chunk_width_p4); |
| 633 | Iy[i]=Ix[i]+chunk_width_p4; |
| 634 | } |
| 635 | |
| 636 | /*Fill four rows of the wrap-around derivative buffers*/ |
| 637 | for(i=top-2;i<top+2;i++) db_IxIyRow_f(Ix[i%5],Iy[i%5],img,i,left-2,chunk_width_p4); |
| 638 | |
| 639 | /*For each output row*/ |
| 640 | for(i=top;i<=bottom;i++) |
| 641 | { |
| 642 | /*Step the derivative buffers*/ |
| 643 | db_IxIyRow_f(Ix[(i+2)%5],Iy[(i+2)%5],img,(i+2),left-2,chunk_width_p4); |
| 644 | |
| 645 | /*Filter Ix2,IxIy,Iy2 vertically into gxx,gxy,gyy*/ |
| 646 | db_gxx_gxy_gyy_row_f(gxx,gxy,gyy,chunk_width_p4, |
| 647 | Ix[(i-2)%5],Ix[(i-1)%5],Ix[i%5],Ix[(i+1)%5],Ix[(i+2)%5], |
| 648 | Iy[(i-2)%5],Iy[(i-1)%5],Iy[i%5],Iy[(i+1)%5],Iy[(i+2)%5]); |
| 649 | |
| 650 | /*Filter gxx,gxy,gyy horizontally and compute corner response s*/ |
| 651 | db_HarrisStrength_row_f(s,gxx,gxy,gyy,i,left,chunk_width); |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | /*Compute the Harris corner strength of the chunk [left,top,left+123,bottom] of img and |
| 656 | store it into the corresponding region of s. left and top have to be at least 3 and |
| 657 | right and bottom have to be at most width-4,height-4. The left of the region in s should |
| 658 | be 16 byte aligned*/ |
| 659 | inline void db_HarrisStrengthChunk_u(float **s,const unsigned char * const *img,int left,int top,int bottom, |
| 660 | /*temp should point to at least |
| 661 | 18*128 of allocated memory*/ |
| 662 | int *temp, int nc) |
| 663 | { |
| 664 | int *Ixx[5],*Ixy[5],*Iyy[5]; |
| 665 | int *gxx,*gxy,*gyy; |
| 666 | int i; |
| 667 | |
| 668 | gxx=temp; |
| 669 | gxy=gxx+128; |
| 670 | gyy=gxy+128; |
| 671 | for(i=0;i<5;i++) |
| 672 | { |
| 673 | Ixx[i]=gyy+(3*i+1)*128; |
| 674 | Ixy[i]=gyy+(3*i+2)*128; |
| 675 | Iyy[i]=gyy+(3*i+3)*128; |
| 676 | } |
| 677 | |
| 678 | /*Fill four rows of the wrap-around derivative buffers*/ |
| 679 | for(i=top-2;i<top+2;i++) db_IxIyRow_u(Ixx[i%5],img,i,left-2,nc); |
| 680 | |
| 681 | /*For each output row*/ |
| 682 | for(i=top;i<=bottom;i++) |
| 683 | { |
| 684 | /*Step the derivative buffers*/ |
| 685 | db_IxIyRow_u(Ixx[(i+2)%5],img,(i+2),left-2,nc); |
| 686 | |
| 687 | /*Filter Ix2,IxIy,Iy2 vertically into gxx,gxy,gyy*/ |
| 688 | db_gxx_gxy_gyy_row_s(gxx,Ixx[(i-2)%5],Ixx[(i-1)%5],Ixx[i%5],Ixx[(i+1)%5],Ixx[(i+2)%5],nc); |
| 689 | |
| 690 | /*Filter gxx,gxy,gyy horizontally and compute corner response s*/ |
| 691 | db_HarrisStrength_row_s(s[i]+left,gxx,gxy,gyy,nc); |
| 692 | } |
| 693 | |
| 694 | } |
| 695 | |
| 696 | /*Compute Harris corner strength of img. Strength is returned for the region |
| 697 | with (3,3) as upper left and (w-4,h-4) as lower right, positioned in the |
| 698 | same place in s. In other words,image should be at least 7 pixels wide and 7 pixels high |
| 699 | for a meaningful result*/ |
| 700 | void db_HarrisStrength_f(float **s,const float * const *img,int w,int h, |
| 701 | /*temp should point to at least |
| 702 | 13*(chunk_width+4) of allocated memory*/ |
| 703 | float *temp, |
| 704 | int chunk_width) |
| 705 | { |
| 706 | int x,next_x,last,right; |
| 707 | |
| 708 | last=w-4; |
| 709 | for(x=3;x<=last;x=next_x) |
| 710 | { |
| 711 | next_x=x+chunk_width; |
| 712 | right=next_x-1; |
| 713 | if(right>last) right=last; |
| 714 | /*Compute the Harris strength of a chunk*/ |
| 715 | db_HarrisStrengthChunk_f(s,img,x,3,right,h-4,temp); |
| 716 | } |
| 717 | } |
| 718 | |
| 719 | /*Compute Harris corner strength of img. Strength is returned for the region |
| 720 | with (3,3) as upper left and (w-4,h-4) as lower right, positioned in the |
| 721 | same place in s. In other words,image should be at least 7 pixels wide and 7 pixels high |
| 722 | for a meaningful result.Moreover, the image should be overallocated by 256 bytes. |
| 723 | s[i][3] should by 16 byte aligned for any i*/ |
| 724 | void db_HarrisStrength_u(float **s, const unsigned char * const *img,int w,int h, |
| 725 | /*temp should point to at least |
| 726 | 18*128 of allocated memory*/ |
| 727 | int *temp) |
| 728 | { |
| 729 | int x,next_x,last; |
| 730 | int nc; |
| 731 | |
| 732 | last=w-4; |
| 733 | for(x=3;x<=last;x=next_x) |
| 734 | { |
| 735 | next_x=x+124; |
| 736 | |
| 737 | // mayban: to revert to the original full chunks state, change the line below to: nc = 128; |
| 738 | nc = db_mini(128,last-x+1); |
| 739 | //nc = 128; |
| 740 | |
| 741 | /*Compute the Harris strength of a chunk*/ |
| 742 | db_HarrisStrengthChunk_u(s,img,x,3,h-4,temp,nc); |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | inline float db_Max_128Aligned16_f(float *v) |
| 747 | { |
| 748 | #ifdef DB_USE_SIMD |
| 749 | float back; |
| 750 | |
| 751 | _asm |
| 752 | { |
| 753 | mov eax,v |
| 754 | |
| 755 | /*Chunk1*/ |
| 756 | movaps xmm0,[eax] |
| 757 | movaps xmm1,[eax+16] |
| 758 | movaps xmm2,[eax+32] |
| 759 | movaps xmm3,[eax+48] |
| 760 | movaps xmm4,[eax+64] |
| 761 | movaps xmm5,[eax+80] |
| 762 | movaps xmm6,[eax+96] |
| 763 | movaps xmm7,[eax+112] |
| 764 | |
| 765 | /*Chunk2*/ |
| 766 | maxps xmm0,[eax+128] |
| 767 | maxps xmm1,[eax+144] |
| 768 | maxps xmm2,[eax+160] |
| 769 | maxps xmm3,[eax+176] |
| 770 | maxps xmm4,[eax+192] |
| 771 | maxps xmm5,[eax+208] |
| 772 | maxps xmm6,[eax+224] |
| 773 | maxps xmm7,[eax+240] |
| 774 | |
| 775 | /*Chunk3*/ |
| 776 | maxps xmm0,[eax+256] |
| 777 | maxps xmm1,[eax+272] |
| 778 | maxps xmm2,[eax+288] |
| 779 | maxps xmm3,[eax+304] |
| 780 | maxps xmm4,[eax+320] |
| 781 | maxps xmm5,[eax+336] |
| 782 | maxps xmm6,[eax+352] |
| 783 | maxps xmm7,[eax+368] |
| 784 | |
| 785 | /*Chunk4*/ |
| 786 | maxps xmm0,[eax+384] |
| 787 | maxps xmm1,[eax+400] |
| 788 | maxps xmm2,[eax+416] |
| 789 | maxps xmm3,[eax+432] |
| 790 | maxps xmm4,[eax+448] |
| 791 | maxps xmm5,[eax+464] |
| 792 | maxps xmm6,[eax+480] |
| 793 | maxps xmm7,[eax+496] |
| 794 | |
| 795 | /*Collect*/ |
| 796 | maxps xmm0,xmm1 |
| 797 | maxps xmm2,xmm3 |
| 798 | maxps xmm4,xmm5 |
| 799 | maxps xmm6,xmm7 |
| 800 | maxps xmm0,xmm2 |
| 801 | maxps xmm4,xmm6 |
| 802 | maxps xmm0,xmm4 |
| 803 | movhlps xmm1,xmm0 |
| 804 | maxps xmm0,xmm1 |
| 805 | shufps xmm1,xmm0,1 |
| 806 | maxps xmm0,xmm1 |
| 807 | movss back,xmm0 |
| 808 | } |
| 809 | |
| 810 | return(back); |
| 811 | #else |
| 812 | float val,max_val; |
| 813 | float *p,*stop_p; |
| 814 | max_val=v[0]; |
| 815 | for(p=v+1,stop_p=v+128;p!=stop_p;) |
| 816 | { |
| 817 | val= *p++; |
| 818 | if(val>max_val) max_val=val; |
| 819 | } |
| 820 | return(max_val); |
| 821 | #endif /*DB_USE_SIMD*/ |
| 822 | } |
| 823 | |
| 824 | inline float db_Max_64Aligned16_f(float *v) |
| 825 | { |
| 826 | #ifdef DB_USE_SIMD |
| 827 | float back; |
| 828 | |
| 829 | _asm |
| 830 | { |
| 831 | mov eax,v |
| 832 | |
| 833 | /*Chunk1*/ |
| 834 | movaps xmm0,[eax] |
| 835 | movaps xmm1,[eax+16] |
| 836 | movaps xmm2,[eax+32] |
| 837 | movaps xmm3,[eax+48] |
| 838 | movaps xmm4,[eax+64] |
| 839 | movaps xmm5,[eax+80] |
| 840 | movaps xmm6,[eax+96] |
| 841 | movaps xmm7,[eax+112] |
| 842 | |
| 843 | /*Chunk2*/ |
| 844 | maxps xmm0,[eax+128] |
| 845 | maxps xmm1,[eax+144] |
| 846 | maxps xmm2,[eax+160] |
| 847 | maxps xmm3,[eax+176] |
| 848 | maxps xmm4,[eax+192] |
| 849 | maxps xmm5,[eax+208] |
| 850 | maxps xmm6,[eax+224] |
| 851 | maxps xmm7,[eax+240] |
| 852 | |
| 853 | /*Collect*/ |
| 854 | maxps xmm0,xmm1 |
| 855 | maxps xmm2,xmm3 |
| 856 | maxps xmm4,xmm5 |
| 857 | maxps xmm6,xmm7 |
| 858 | maxps xmm0,xmm2 |
| 859 | maxps xmm4,xmm6 |
| 860 | maxps xmm0,xmm4 |
| 861 | movhlps xmm1,xmm0 |
| 862 | maxps xmm0,xmm1 |
| 863 | shufps xmm1,xmm0,1 |
| 864 | maxps xmm0,xmm1 |
| 865 | movss back,xmm0 |
| 866 | } |
| 867 | |
| 868 | return(back); |
| 869 | #else |
| 870 | float val,max_val; |
| 871 | float *p,*stop_p; |
| 872 | max_val=v[0]; |
| 873 | for(p=v+1,stop_p=v+64;p!=stop_p;) |
| 874 | { |
| 875 | val= *p++; |
| 876 | if(val>max_val) max_val=val; |
| 877 | } |
| 878 | return(max_val); |
| 879 | #endif /*DB_USE_SIMD*/ |
| 880 | } |
| 881 | |
| 882 | inline float db_Max_32Aligned16_f(float *v) |
| 883 | { |
| 884 | #ifdef DB_USE_SIMD |
| 885 | float back; |
| 886 | |
| 887 | _asm |
| 888 | { |
| 889 | mov eax,v |
| 890 | |
| 891 | /*Chunk1*/ |
| 892 | movaps xmm0,[eax] |
| 893 | movaps xmm1,[eax+16] |
| 894 | movaps xmm2,[eax+32] |
| 895 | movaps xmm3,[eax+48] |
| 896 | movaps xmm4,[eax+64] |
| 897 | movaps xmm5,[eax+80] |
| 898 | movaps xmm6,[eax+96] |
| 899 | movaps xmm7,[eax+112] |
| 900 | |
| 901 | /*Collect*/ |
| 902 | maxps xmm0,xmm1 |
| 903 | maxps xmm2,xmm3 |
| 904 | maxps xmm4,xmm5 |
| 905 | maxps xmm6,xmm7 |
| 906 | maxps xmm0,xmm2 |
| 907 | maxps xmm4,xmm6 |
| 908 | maxps xmm0,xmm4 |
| 909 | movhlps xmm1,xmm0 |
| 910 | maxps xmm0,xmm1 |
| 911 | shufps xmm1,xmm0,1 |
| 912 | maxps xmm0,xmm1 |
| 913 | movss back,xmm0 |
| 914 | } |
| 915 | |
| 916 | return(back); |
| 917 | #else |
| 918 | float val,max_val; |
| 919 | float *p,*stop_p; |
| 920 | max_val=v[0]; |
| 921 | for(p=v+1,stop_p=v+32;p!=stop_p;) |
| 922 | { |
| 923 | val= *p++; |
| 924 | if(val>max_val) max_val=val; |
| 925 | } |
| 926 | return(max_val); |
| 927 | #endif /*DB_USE_SIMD*/ |
| 928 | } |
| 929 | |
| 930 | inline float db_Max_16Aligned16_f(float *v) |
| 931 | { |
| 932 | #ifdef DB_USE_SIMD |
| 933 | float back; |
| 934 | |
| 935 | _asm |
| 936 | { |
| 937 | mov eax,v |
| 938 | |
| 939 | /*Chunk1*/ |
| 940 | movaps xmm0,[eax] |
| 941 | movaps xmm1,[eax+16] |
| 942 | movaps xmm2,[eax+32] |
| 943 | movaps xmm3,[eax+48] |
| 944 | |
| 945 | /*Collect*/ |
| 946 | maxps xmm0,xmm1 |
| 947 | maxps xmm2,xmm3 |
| 948 | maxps xmm0,xmm2 |
| 949 | movhlps xmm1,xmm0 |
| 950 | maxps xmm0,xmm1 |
| 951 | shufps xmm1,xmm0,1 |
| 952 | maxps xmm0,xmm1 |
| 953 | movss back,xmm0 |
| 954 | } |
| 955 | |
| 956 | return(back); |
| 957 | #else |
| 958 | float val,max_val; |
| 959 | float *p,*stop_p; |
| 960 | max_val=v[0]; |
| 961 | for(p=v+1,stop_p=v+16;p!=stop_p;) |
| 962 | { |
| 963 | val= *p++; |
| 964 | if(val>max_val) max_val=val; |
| 965 | } |
| 966 | return(max_val); |
| 967 | #endif /*DB_USE_SIMD*/ |
| 968 | } |
| 969 | |
| 970 | inline float db_Max_8Aligned16_f(float *v) |
| 971 | { |
| 972 | #ifdef DB_USE_SIMD |
| 973 | float back; |
| 974 | |
| 975 | _asm |
| 976 | { |
| 977 | mov eax,v |
| 978 | |
| 979 | /*Chunk1*/ |
| 980 | movaps xmm0,[eax] |
| 981 | movaps xmm1,[eax+16] |
| 982 | |
| 983 | /*Collect*/ |
| 984 | maxps xmm0,xmm1 |
| 985 | movhlps xmm1,xmm0 |
| 986 | maxps xmm0,xmm1 |
| 987 | shufps xmm1,xmm0,1 |
| 988 | maxps xmm0,xmm1 |
| 989 | movss back,xmm0 |
| 990 | } |
| 991 | |
| 992 | return(back); |
| 993 | #else |
| 994 | float val,max_val; |
| 995 | float *p,*stop_p; |
| 996 | max_val=v[0]; |
| 997 | for(p=v+1,stop_p=v+8;p!=stop_p;) |
| 998 | { |
| 999 | val= *p++; |
| 1000 | if(val>max_val) max_val=val; |
| 1001 | } |
| 1002 | return(max_val); |
| 1003 | #endif /*DB_USE_SIMD*/ |
| 1004 | } |
| 1005 | |
| 1006 | inline float db_Max_Aligned16_f(float *v,int size) |
| 1007 | { |
| 1008 | float val,max_val; |
| 1009 | float *stop_v; |
| 1010 | |
| 1011 | max_val=v[0]; |
| 1012 | for(;size>=128;size-=128) |
| 1013 | { |
| 1014 | val=db_Max_128Aligned16_f(v); |
| 1015 | v+=128; |
| 1016 | if(val>max_val) max_val=val; |
| 1017 | } |
| 1018 | if(size&64) |
| 1019 | { |
| 1020 | val=db_Max_64Aligned16_f(v); |
| 1021 | v+=64; |
| 1022 | if(val>max_val) max_val=val; |
| 1023 | } |
| 1024 | if(size&32) |
| 1025 | { |
| 1026 | val=db_Max_32Aligned16_f(v); |
| 1027 | v+=32; |
| 1028 | if(val>max_val) max_val=val; |
| 1029 | } |
| 1030 | if(size&16) |
| 1031 | { |
| 1032 | val=db_Max_16Aligned16_f(v); |
| 1033 | v+=16; |
| 1034 | if(val>max_val) max_val=val; |
| 1035 | } |
| 1036 | if(size&8) |
| 1037 | { |
| 1038 | val=db_Max_8Aligned16_f(v); |
| 1039 | v+=8; |
| 1040 | if(val>max_val) max_val=val; |
| 1041 | } |
| 1042 | if(size&7) |
| 1043 | { |
| 1044 | for(stop_v=v+(size&7);v!=stop_v;) |
| 1045 | { |
| 1046 | val= *v++; |
| 1047 | if(val>max_val) max_val=val; |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | return(max_val); |
| 1052 | } |
| 1053 | |
| 1054 | /*Find maximum value of img in the region starting at (left,top) |
| 1055 | and with width w and height h. img[left] should be 16 byte aligned*/ |
| 1056 | float db_MaxImage_Aligned16_f(float **img,int left,int top,int w,int h) |
| 1057 | { |
| 1058 | float val,max_val; |
| 1059 | int i,stop_i; |
| 1060 | |
| 1061 | if(w && h) |
| 1062 | { |
| 1063 | stop_i=top+h; |
| 1064 | max_val=img[top][left]; |
| 1065 | |
| 1066 | for(i=top;i<stop_i;i++) |
| 1067 | { |
| 1068 | val=db_Max_Aligned16_f(img[i]+left,w); |
| 1069 | if(val>max_val) max_val=val; |
| 1070 | } |
| 1071 | return(max_val); |
| 1072 | } |
| 1073 | return(0.0); |
| 1074 | } |
| 1075 | |
| 1076 | inline void db_MaxVector_128_Aligned16_f(float *m,float *v1,float *v2) |
| 1077 | { |
| 1078 | #ifdef DB_USE_SIMD |
| 1079 | _asm |
| 1080 | { |
| 1081 | mov eax,v1 |
| 1082 | mov ebx,v2 |
| 1083 | mov ecx,m |
| 1084 | |
| 1085 | /*Chunk1*/ |
| 1086 | movaps xmm0,[eax] |
| 1087 | movaps xmm1,[eax+16] |
| 1088 | movaps xmm2,[eax+32] |
| 1089 | movaps xmm3,[eax+48] |
| 1090 | movaps xmm4,[eax+64] |
| 1091 | movaps xmm5,[eax+80] |
| 1092 | movaps xmm6,[eax+96] |
| 1093 | movaps xmm7,[eax+112] |
| 1094 | maxps xmm0,[ebx] |
| 1095 | maxps xmm1,[ebx+16] |
| 1096 | maxps xmm2,[ebx+32] |
| 1097 | maxps xmm3,[ebx+48] |
| 1098 | maxps xmm4,[ebx+64] |
| 1099 | maxps xmm5,[ebx+80] |
| 1100 | maxps xmm6,[ebx+96] |
| 1101 | maxps xmm7,[ebx+112] |
| 1102 | movaps [ecx],xmm0 |
| 1103 | movaps [ecx+16],xmm1 |
| 1104 | movaps [ecx+32],xmm2 |
| 1105 | movaps [ecx+48],xmm3 |
| 1106 | movaps [ecx+64],xmm4 |
| 1107 | movaps [ecx+80],xmm5 |
| 1108 | movaps [ecx+96],xmm6 |
| 1109 | movaps [ecx+112],xmm7 |
| 1110 | |
| 1111 | /*Chunk2*/ |
| 1112 | movaps xmm0,[eax+128] |
| 1113 | movaps xmm1,[eax+144] |
| 1114 | movaps xmm2,[eax+160] |
| 1115 | movaps xmm3,[eax+176] |
| 1116 | movaps xmm4,[eax+192] |
| 1117 | movaps xmm5,[eax+208] |
| 1118 | movaps xmm6,[eax+224] |
| 1119 | movaps xmm7,[eax+240] |
| 1120 | maxps xmm0,[ebx+128] |
| 1121 | maxps xmm1,[ebx+144] |
| 1122 | maxps xmm2,[ebx+160] |
| 1123 | maxps xmm3,[ebx+176] |
| 1124 | maxps xmm4,[ebx+192] |
| 1125 | maxps xmm5,[ebx+208] |
| 1126 | maxps xmm6,[ebx+224] |
| 1127 | maxps xmm7,[ebx+240] |
| 1128 | movaps [ecx+128],xmm0 |
| 1129 | movaps [ecx+144],xmm1 |
| 1130 | movaps [ecx+160],xmm2 |
| 1131 | movaps [ecx+176],xmm3 |
| 1132 | movaps [ecx+192],xmm4 |
| 1133 | movaps [ecx+208],xmm5 |
| 1134 | movaps [ecx+224],xmm6 |
| 1135 | movaps [ecx+240],xmm7 |
| 1136 | |
| 1137 | /*Chunk3*/ |
| 1138 | movaps xmm0,[eax+256] |
| 1139 | movaps xmm1,[eax+272] |
| 1140 | movaps xmm2,[eax+288] |
| 1141 | movaps xmm3,[eax+304] |
| 1142 | movaps xmm4,[eax+320] |
| 1143 | movaps xmm5,[eax+336] |
| 1144 | movaps xmm6,[eax+352] |
| 1145 | movaps xmm7,[eax+368] |
| 1146 | maxps xmm0,[ebx+256] |
| 1147 | maxps xmm1,[ebx+272] |
| 1148 | maxps xmm2,[ebx+288] |
| 1149 | maxps xmm3,[ebx+304] |
| 1150 | maxps xmm4,[ebx+320] |
| 1151 | maxps xmm5,[ebx+336] |
| 1152 | maxps xmm6,[ebx+352] |
| 1153 | maxps xmm7,[ebx+368] |
| 1154 | movaps [ecx+256],xmm0 |
| 1155 | movaps [ecx+272],xmm1 |
| 1156 | movaps [ecx+288],xmm2 |
| 1157 | movaps [ecx+304],xmm3 |
| 1158 | movaps [ecx+320],xmm4 |
| 1159 | movaps [ecx+336],xmm5 |
| 1160 | movaps [ecx+352],xmm6 |
| 1161 | movaps [ecx+368],xmm7 |
| 1162 | |
| 1163 | /*Chunk4*/ |
| 1164 | movaps xmm0,[eax+384] |
| 1165 | movaps xmm1,[eax+400] |
| 1166 | movaps xmm2,[eax+416] |
| 1167 | movaps xmm3,[eax+432] |
| 1168 | movaps xmm4,[eax+448] |
| 1169 | movaps xmm5,[eax+464] |
| 1170 | movaps xmm6,[eax+480] |
| 1171 | movaps xmm7,[eax+496] |
| 1172 | maxps xmm0,[ebx+384] |
| 1173 | maxps xmm1,[ebx+400] |
| 1174 | maxps xmm2,[ebx+416] |
| 1175 | maxps xmm3,[ebx+432] |
| 1176 | maxps xmm4,[ebx+448] |
| 1177 | maxps xmm5,[ebx+464] |
| 1178 | maxps xmm6,[ebx+480] |
| 1179 | maxps xmm7,[ebx+496] |
| 1180 | movaps [ecx+384],xmm0 |
| 1181 | movaps [ecx+400],xmm1 |
| 1182 | movaps [ecx+416],xmm2 |
| 1183 | movaps [ecx+432],xmm3 |
| 1184 | movaps [ecx+448],xmm4 |
| 1185 | movaps [ecx+464],xmm5 |
| 1186 | movaps [ecx+480],xmm6 |
| 1187 | movaps [ecx+496],xmm7 |
| 1188 | } |
| 1189 | #else |
| 1190 | int i; |
| 1191 | float a,b; |
| 1192 | for(i=0;i<128;i++) |
| 1193 | { |
| 1194 | a=v1[i]; |
| 1195 | b=v2[i]; |
| 1196 | if(a>=b) m[i]=a; |
| 1197 | else m[i]=b; |
| 1198 | } |
| 1199 | #endif /*DB_USE_SIMD*/ |
| 1200 | } |
| 1201 | |
| 1202 | inline void db_MaxVector_128_SecondSourceDestAligned16_f(float *m,float *v1,float *v2) |
| 1203 | { |
| 1204 | #ifdef DB_USE_SIMD |
| 1205 | _asm |
| 1206 | { |
| 1207 | mov eax,v1 |
| 1208 | mov ebx,v2 |
| 1209 | mov ecx,m |
| 1210 | |
| 1211 | /*Chunk1*/ |
| 1212 | movups xmm0,[eax] |
| 1213 | movups xmm1,[eax+16] |
| 1214 | movups xmm2,[eax+32] |
| 1215 | movups xmm3,[eax+48] |
| 1216 | movups xmm4,[eax+64] |
| 1217 | movups xmm5,[eax+80] |
| 1218 | movups xmm6,[eax+96] |
| 1219 | movups xmm7,[eax+112] |
| 1220 | maxps xmm0,[ebx] |
| 1221 | maxps xmm1,[ebx+16] |
| 1222 | maxps xmm2,[ebx+32] |
| 1223 | maxps xmm3,[ebx+48] |
| 1224 | maxps xmm4,[ebx+64] |
| 1225 | maxps xmm5,[ebx+80] |
| 1226 | maxps xmm6,[ebx+96] |
| 1227 | maxps xmm7,[ebx+112] |
| 1228 | movaps [ecx],xmm0 |
| 1229 | movaps [ecx+16],xmm1 |
| 1230 | movaps [ecx+32],xmm2 |
| 1231 | movaps [ecx+48],xmm3 |
| 1232 | movaps [ecx+64],xmm4 |
| 1233 | movaps [ecx+80],xmm5 |
| 1234 | movaps [ecx+96],xmm6 |
| 1235 | movaps [ecx+112],xmm7 |
| 1236 | |
| 1237 | /*Chunk2*/ |
| 1238 | movups xmm0,[eax+128] |
| 1239 | movups xmm1,[eax+144] |
| 1240 | movups xmm2,[eax+160] |
| 1241 | movups xmm3,[eax+176] |
| 1242 | movups xmm4,[eax+192] |
| 1243 | movups xmm5,[eax+208] |
| 1244 | movups xmm6,[eax+224] |
| 1245 | movups xmm7,[eax+240] |
| 1246 | maxps xmm0,[ebx+128] |
| 1247 | maxps xmm1,[ebx+144] |
| 1248 | maxps xmm2,[ebx+160] |
| 1249 | maxps xmm3,[ebx+176] |
| 1250 | maxps xmm4,[ebx+192] |
| 1251 | maxps xmm5,[ebx+208] |
| 1252 | maxps xmm6,[ebx+224] |
| 1253 | maxps xmm7,[ebx+240] |
| 1254 | movaps [ecx+128],xmm0 |
| 1255 | movaps [ecx+144],xmm1 |
| 1256 | movaps [ecx+160],xmm2 |
| 1257 | movaps [ecx+176],xmm3 |
| 1258 | movaps [ecx+192],xmm4 |
| 1259 | movaps [ecx+208],xmm5 |
| 1260 | movaps [ecx+224],xmm6 |
| 1261 | movaps [ecx+240],xmm7 |
| 1262 | |
| 1263 | /*Chunk3*/ |
| 1264 | movups xmm0,[eax+256] |
| 1265 | movups xmm1,[eax+272] |
| 1266 | movups xmm2,[eax+288] |
| 1267 | movups xmm3,[eax+304] |
| 1268 | movups xmm4,[eax+320] |
| 1269 | movups xmm5,[eax+336] |
| 1270 | movups xmm6,[eax+352] |
| 1271 | movups xmm7,[eax+368] |
| 1272 | maxps xmm0,[ebx+256] |
| 1273 | maxps xmm1,[ebx+272] |
| 1274 | maxps xmm2,[ebx+288] |
| 1275 | maxps xmm3,[ebx+304] |
| 1276 | maxps xmm4,[ebx+320] |
| 1277 | maxps xmm5,[ebx+336] |
| 1278 | maxps xmm6,[ebx+352] |
| 1279 | maxps xmm7,[ebx+368] |
| 1280 | movaps [ecx+256],xmm0 |
| 1281 | movaps [ecx+272],xmm1 |
| 1282 | movaps [ecx+288],xmm2 |
| 1283 | movaps [ecx+304],xmm3 |
| 1284 | movaps [ecx+320],xmm4 |
| 1285 | movaps [ecx+336],xmm5 |
| 1286 | movaps [ecx+352],xmm6 |
| 1287 | movaps [ecx+368],xmm7 |
| 1288 | |
| 1289 | /*Chunk4*/ |
| 1290 | movups xmm0,[eax+384] |
| 1291 | movups xmm1,[eax+400] |
| 1292 | movups xmm2,[eax+416] |
| 1293 | movups xmm3,[eax+432] |
| 1294 | movups xmm4,[eax+448] |
| 1295 | movups xmm5,[eax+464] |
| 1296 | movups xmm6,[eax+480] |
| 1297 | movups xmm7,[eax+496] |
| 1298 | maxps xmm0,[ebx+384] |
| 1299 | maxps xmm1,[ebx+400] |
| 1300 | maxps xmm2,[ebx+416] |
| 1301 | maxps xmm3,[ebx+432] |
| 1302 | maxps xmm4,[ebx+448] |
| 1303 | maxps xmm5,[ebx+464] |
| 1304 | maxps xmm6,[ebx+480] |
| 1305 | maxps xmm7,[ebx+496] |
| 1306 | movaps [ecx+384],xmm0 |
| 1307 | movaps [ecx+400],xmm1 |
| 1308 | movaps [ecx+416],xmm2 |
| 1309 | movaps [ecx+432],xmm3 |
| 1310 | movaps [ecx+448],xmm4 |
| 1311 | movaps [ecx+464],xmm5 |
| 1312 | movaps [ecx+480],xmm6 |
| 1313 | movaps [ecx+496],xmm7 |
| 1314 | } |
| 1315 | #else |
| 1316 | int i; |
| 1317 | float a,b; |
| 1318 | for(i=0;i<128;i++) |
| 1319 | { |
| 1320 | a=v1[i]; |
| 1321 | b=v2[i]; |
| 1322 | if(a>=b) m[i]=a; |
| 1323 | else m[i]=b; |
| 1324 | } |
| 1325 | #endif /*DB_USE_SIMD*/ |
| 1326 | } |
| 1327 | |
| 1328 | /*Compute Max-suppression-filtered image for a chunk of sf starting at (left,top), of width 124 and |
| 1329 | stopping at bottom. The output is shifted two steps left and overwrites 128 elements for each row. |
| 1330 | The input s should be of width at least 128, and exist for 2 pixels outside the specified region. |
| 1331 | s[i][left-2] and sf[i][left-2] should be 16 byte aligned. Top must be at least 3*/ |
| 1332 | inline void db_MaxSuppressFilterChunk_5x5_Aligned16_f(float **sf,float **s,int left,int top,int bottom, |
| 1333 | /*temp should point to at least |
| 1334 | 6*132 floats of 16-byte-aligned allocated memory*/ |
| 1335 | float *temp) |
| 1336 | { |
| 1337 | #ifdef DB_USE_SIMD |
| 1338 | int i,lm2; |
| 1339 | float *two[4]; |
| 1340 | float *four,*five; |
| 1341 | |
| 1342 | lm2=left-2; |
| 1343 | |
| 1344 | /*Set pointers to pre-allocated memory*/ |
| 1345 | four=temp; |
| 1346 | five=four+132; |
| 1347 | for(i=0;i<4;i++) |
| 1348 | { |
| 1349 | two[i]=five+(i+1)*132; |
| 1350 | } |
| 1351 | |
| 1352 | /*Set rests of four and five to zero to avoid |
| 1353 | floating point exceptions*/ |
| 1354 | for(i=129;i<132;i++) |
| 1355 | { |
| 1356 | four[i]=0.0; |
| 1357 | five[i]=0.0; |
| 1358 | } |
| 1359 | |
| 1360 | /*Fill three rows of the wrap-around max buffers*/ |
| 1361 | for(i=top-3;i<top;i++) db_MaxVector_128_Aligned16_f(two[i&3],s[i+1]+lm2,s[i+2]+lm2); |
| 1362 | |
| 1363 | /*For each output row*/ |
| 1364 | for(;i<=bottom;i++) |
| 1365 | { |
| 1366 | /*Compute max of the lowest pair of rows in the five row window*/ |
| 1367 | db_MaxVector_128_Aligned16_f(two[i&3],s[i+1]+lm2,s[i+2]+lm2); |
| 1368 | /*Compute max of the lowest and highest pair of rows in the five row window*/ |
| 1369 | db_MaxVector_128_Aligned16_f(four,two[i&3],two[(i-3)&3]); |
| 1370 | /*Compute max of all rows*/ |
| 1371 | db_MaxVector_128_Aligned16_f(five,four,two[(i-1)&3]); |
| 1372 | /*Compute max of 2x5 chunks*/ |
| 1373 | db_MaxVector_128_SecondSourceDestAligned16_f(five,five+1,five); |
| 1374 | /*Compute max of pairs of 2x5 chunks*/ |
| 1375 | db_MaxVector_128_SecondSourceDestAligned16_f(five,five+3,five); |
| 1376 | /*Compute max of pairs of 5x5 except middle*/ |
| 1377 | db_MaxVector_128_SecondSourceDestAligned16_f(sf[i]+lm2,four+2,five); |
| 1378 | } |
| 1379 | |
| 1380 | #else |
| 1381 | int i,j,right; |
| 1382 | float sv; |
| 1383 | |
| 1384 | right=left+128; |
| 1385 | for(i=top;i<=bottom;i++) for(j=left;j<right;j++) |
| 1386 | { |
| 1387 | sv=s[i][j]; |
| 1388 | |
| 1389 | if( sv>s[i-2][j-2] && sv>s[i-2][j-1] && sv>s[i-2][j] && sv>s[i-2][j+1] && sv>s[i-2][j+2] && |
| 1390 | sv>s[i-1][j-2] && sv>s[i-1][j-1] && sv>s[i-1][j] && sv>s[i-1][j+1] && sv>s[i-1][j+2] && |
| 1391 | sv>s[ i][j-2] && sv>s[ i][j-1] && sv>s[ i][j+1] && sv>s[ i][j+2] && |
| 1392 | sv>s[i+1][j-2] && sv>s[i+1][j-1] && sv>s[i+1][j] && sv>s[i+1][j+1] && sv>s[i+1][j+2] && |
| 1393 | sv>s[i+2][j-2] && sv>s[i+2][j-1] && sv>s[i+2][j] && sv>s[i+2][j+1] && sv>s[i+2][j+2]) |
| 1394 | { |
| 1395 | sf[i][j-2]=0.0; |
| 1396 | } |
| 1397 | else sf[i][j-2]=sv; |
| 1398 | } |
| 1399 | #endif /*DB_USE_SIMD*/ |
| 1400 | } |
| 1401 | |
| 1402 | /*Compute Max-suppression-filtered image for a chunk of sf starting at (left,top) and |
| 1403 | stopping at bottom. The output is shifted two steps left. The input s should exist for 2 pixels |
| 1404 | outside the specified region. s[i][left-2] and sf[i][left-2] should be 16 byte aligned. |
| 1405 | Top must be at least 3. Reading and writing from and to the input and output images is done |
| 1406 | as if the region had a width equal to a multiple of 124. If this is not the case, the images |
| 1407 | should be over-allocated and the input cleared for a sufficient region*/ |
| 1408 | void db_MaxSuppressFilter_5x5_Aligned16_f(float **sf,float **s,int left,int top,int right,int bottom, |
| 1409 | /*temp should point to at least |
| 1410 | 6*132 floats of 16-byte-aligned allocated memory*/ |
| 1411 | float *temp) |
| 1412 | { |
| 1413 | int x,next_x; |
| 1414 | |
| 1415 | for(x=left;x<=right;x=next_x) |
| 1416 | { |
| 1417 | next_x=x+124; |
| 1418 | db_MaxSuppressFilterChunk_5x5_Aligned16_f(sf,s,x,top,bottom,temp); |
| 1419 | } |
| 1420 | } |
| 1421 | |
| 1422 | /*Extract corners from the chunk (left,top) to (right,bottom). Store in x_temp,y_temp and s_temp |
| 1423 | which should point to space of at least as many positions as there are pixels in the chunk*/ |
| 1424 | inline int db_CornersFromChunk(float **strength,int left,int top,int right,int bottom,float threshold,double *x_temp,double *y_temp,double *s_temp) |
| 1425 | { |
| 1426 | int i,j,nr; |
| 1427 | float s; |
| 1428 | |
| 1429 | nr=0; |
| 1430 | for(i=top;i<=bottom;i++) for(j=left;j<=right;j++) |
| 1431 | { |
| 1432 | s=strength[i][j]; |
| 1433 | |
| 1434 | if(s>=threshold && |
| 1435 | s>strength[i-2][j-2] && s>strength[i-2][j-1] && s>strength[i-2][j] && s>strength[i-2][j+1] && s>strength[i-2][j+2] && |
| 1436 | s>strength[i-1][j-2] && s>strength[i-1][j-1] && s>strength[i-1][j] && s>strength[i-1][j+1] && s>strength[i-1][j+2] && |
| 1437 | s>strength[ i][j-2] && s>strength[ i][j-1] && s>strength[ i][j+1] && s>strength[ i][j+2] && |
| 1438 | s>strength[i+1][j-2] && s>strength[i+1][j-1] && s>strength[i+1][j] && s>strength[i+1][j+1] && s>strength[i+1][j+2] && |
| 1439 | s>strength[i+2][j-2] && s>strength[i+2][j-1] && s>strength[i+2][j] && s>strength[i+2][j+1] && s>strength[i+2][j+2]) |
| 1440 | { |
| 1441 | x_temp[nr]=(double) j; |
| 1442 | y_temp[nr]=(double) i; |
| 1443 | s_temp[nr]=(double) s; |
| 1444 | nr++; |
| 1445 | } |
| 1446 | } |
| 1447 | return(nr); |
| 1448 | } |
| 1449 | |
| 1450 | |
| 1451 | //Sub-pixel accuracy using 2D quadratic interpolation.(YCJ) |
| 1452 | inline void db_SubPixel(float **strength, const double xd, const double yd, double &xs, double &ys) |
| 1453 | { |
| 1454 | int x = (int) xd; |
| 1455 | int y = (int) yd; |
| 1456 | |
| 1457 | float fxx = strength[y][x-1] - strength[y][x] - strength[y][x] + strength[y][x+1]; |
| 1458 | float fyy = strength[y-1][x] - strength[y][x] - strength[y][x] + strength[y+1][x]; |
| 1459 | float fxy = (strength[y-1][x-1] - strength[y-1][x+1] - strength[y+1][x-1] + strength[y+1][x+1])/(float)4.0; |
| 1460 | |
| 1461 | float denom = (fxx * fyy - fxy * fxy) * (float) 2.0; |
| 1462 | |
| 1463 | xs = xd; |
| 1464 | ys = yd; |
| 1465 | |
| 1466 | if ( db_absf(denom) <= FLT_EPSILON ) |
| 1467 | { |
| 1468 | return; |
| 1469 | } |
| 1470 | else |
| 1471 | { |
| 1472 | float fx = strength[y][x+1] - strength[y][x-1]; |
| 1473 | float fy = strength[y+1][x] - strength[y-1][x]; |
| 1474 | |
| 1475 | float dx = (fyy * fx - fxy * fy) / denom; |
| 1476 | float dy = (fxx * fy - fxy * fx) / denom; |
| 1477 | |
| 1478 | if ( db_absf(dx) > 1.0 || db_absf(dy) > 1.0 ) |
| 1479 | { |
| 1480 | return; |
| 1481 | } |
| 1482 | else |
| 1483 | { |
| 1484 | xs -= dx; |
| 1485 | ys -= dy; |
| 1486 | } |
| 1487 | } |
| 1488 | |
| 1489 | return; |
| 1490 | } |
| 1491 | |
| 1492 | /*Extract corners from the image part from (left,top) to (right,bottom). |
| 1493 | Store in x and y, extracting at most satnr corners in each block of size (bw,bh). |
| 1494 | The pointer temp_d should point to at least 5*bw*bh positions. |
| 1495 | area_factor holds how many corners max to extract per 10000 pixels*/ |
| 1496 | void db_ExtractCornersSaturated(float **strength,int left,int top,int right,int bottom, |
| 1497 | int bw,int bh,unsigned long area_factor, |
| 1498 | float threshold,double *temp_d, |
| 1499 | double *x_coord,double *y_coord,int *nr_corners) |
| 1500 | { |
| 1501 | double *x_temp,*y_temp,*s_temp,*select_temp; |
| 1502 | double loc_thresh; |
| 1503 | unsigned long bwbh,area,saturation; |
| 1504 | int x,next_x,last_x; |
| 1505 | int y,next_y,last_y; |
| 1506 | int nr,nr_points,i,stop; |
| 1507 | |
| 1508 | bwbh=bw*bh; |
| 1509 | x_temp=temp_d; |
| 1510 | y_temp=x_temp+bwbh; |
| 1511 | s_temp=y_temp+bwbh; |
| 1512 | select_temp=s_temp+bwbh; |
| 1513 | |
| 1514 | #ifdef DB_SUB_PIXEL |
| 1515 | // subpixel processing may sometimes push the corner ourside the real border |
| 1516 | // increasing border size: |
| 1517 | left++; |
| 1518 | top++; |
| 1519 | bottom--; |
| 1520 | right--; |
| 1521 | #endif /*DB_SUB_PIXEL*/ |
| 1522 | |
| 1523 | nr_points=0; |
| 1524 | for(y=top;y<=bottom;y=next_y) |
| 1525 | { |
| 1526 | next_y=y+bh; |
| 1527 | last_y=next_y-1; |
| 1528 | if(last_y>bottom) last_y=bottom; |
| 1529 | for(x=left;x<=right;x=next_x) |
| 1530 | { |
| 1531 | next_x=x+bw; |
| 1532 | last_x=next_x-1; |
| 1533 | if(last_x>right) last_x=right; |
| 1534 | |
| 1535 | area=(last_x-x+1)*(last_y-y+1); |
| 1536 | saturation=(area*area_factor)/10000; |
| 1537 | nr=db_CornersFromChunk(strength,x,y,last_x,last_y,threshold,x_temp,y_temp,s_temp); |
| 1538 | if(nr) |
| 1539 | { |
| 1540 | if(((unsigned long)nr)>saturation) loc_thresh=db_LeanQuickSelect(s_temp,nr,nr-saturation,select_temp); |
| 1541 | else loc_thresh=threshold; |
| 1542 | |
| 1543 | stop=nr_points+saturation; |
| 1544 | for(i=0;(i<nr)&&(nr_points<stop);i++) |
| 1545 | { |
| 1546 | if(s_temp[i]>=loc_thresh) |
| 1547 | { |
| 1548 | #ifdef DB_SUB_PIXEL |
| 1549 | db_SubPixel(strength, x_temp[i], y_temp[i], x_coord[nr_points], y_coord[nr_points]); |
| 1550 | #else |
| 1551 | x_coord[nr_points]=x_temp[i]; |
| 1552 | y_coord[nr_points]=y_temp[i]; |
| 1553 | #endif |
| 1554 | |
| 1555 | nr_points++; |
| 1556 | } |
| 1557 | } |
| 1558 | } |
| 1559 | } |
| 1560 | } |
| 1561 | *nr_corners=nr_points; |
| 1562 | } |
| 1563 | |
| 1564 | db_CornerDetector_f::db_CornerDetector_f() |
| 1565 | { |
| 1566 | m_w=0; m_h=0; |
| 1567 | } |
| 1568 | |
| 1569 | db_CornerDetector_f::~db_CornerDetector_f() |
| 1570 | { |
| 1571 | Clean(); |
| 1572 | } |
| 1573 | |
| 1574 | void db_CornerDetector_f::Clean() |
| 1575 | { |
| 1576 | if(m_w!=0) |
| 1577 | { |
| 1578 | delete [] m_temp_f; |
| 1579 | delete [] m_temp_d; |
| 1580 | db_FreeStrengthImage_f(m_strength_mem,m_strength,m_h); |
| 1581 | } |
| 1582 | m_w=0; m_h=0; |
| 1583 | } |
| 1584 | |
| 1585 | unsigned long db_CornerDetector_f::Init(int im_width,int im_height,int target_nr_corners, |
| 1586 | int nr_horizontal_blocks,int nr_vertical_blocks, |
| 1587 | double absolute_threshold,double relative_threshold) |
| 1588 | { |
| 1589 | int chunkwidth=208; |
| 1590 | int block_width,block_height; |
| 1591 | unsigned long area_factor; |
| 1592 | int active_width,active_height; |
| 1593 | |
| 1594 | active_width=db_maxi(1,im_width-10); |
| 1595 | active_height=db_maxi(1,im_height-10); |
| 1596 | block_width=db_maxi(1,active_width/nr_horizontal_blocks); |
| 1597 | block_height=db_maxi(1,active_height/nr_vertical_blocks); |
| 1598 | |
| 1599 | area_factor=db_minl(1000,db_maxl(1,(long)(10000.0*((double)target_nr_corners)/ |
| 1600 | (((double)active_width)*((double)active_height))))); |
| 1601 | |
| 1602 | return(Start(im_width,im_height,block_width,block_height,area_factor, |
| 1603 | absolute_threshold,relative_threshold,chunkwidth)); |
| 1604 | } |
| 1605 | |
| 1606 | unsigned long db_CornerDetector_f::Start(int im_width,int im_height, |
| 1607 | int block_width,int block_height,unsigned long area_factor, |
| 1608 | double absolute_threshold,double relative_threshold,int chunkwidth) |
| 1609 | { |
| 1610 | Clean(); |
| 1611 | |
| 1612 | m_w=im_width; |
| 1613 | m_h=im_height; |
| 1614 | m_cw=chunkwidth; |
| 1615 | m_bw=block_width; |
| 1616 | m_bh=block_height; |
| 1617 | m_area_factor=area_factor; |
| 1618 | m_r_thresh=relative_threshold; |
| 1619 | m_a_thresh=absolute_threshold; |
| 1620 | m_max_nr=db_maxl(1,1+(m_w*m_h*m_area_factor)/10000); |
| 1621 | |
| 1622 | m_temp_f=new float[13*(m_cw+4)]; |
| 1623 | m_temp_d=new double[5*m_bw*m_bh]; |
| 1624 | m_strength=db_AllocStrengthImage_f(&m_strength_mem,m_w,m_h); |
| 1625 | |
| 1626 | return(m_max_nr); |
| 1627 | } |
| 1628 | |
| 1629 | void db_CornerDetector_f::DetectCorners(const float * const *img,double *x_coord,double *y_coord,int *nr_corners) const |
| 1630 | { |
| 1631 | float max_val,threshold; |
| 1632 | |
| 1633 | db_HarrisStrength_f(m_strength,img,m_w,m_h,m_temp_f,m_cw); |
| 1634 | |
| 1635 | if(m_r_thresh) |
| 1636 | { |
| 1637 | max_val=db_MaxImage_Aligned16_f(m_strength,3,3,m_w-6,m_h-6); |
| 1638 | threshold= (float) db_maxd(m_a_thresh,max_val*m_r_thresh); |
| 1639 | } |
| 1640 | else threshold= (float) m_a_thresh; |
| 1641 | |
| 1642 | db_ExtractCornersSaturated(m_strength,BORDER,BORDER,m_w-BORDER-1,m_h-BORDER-1,m_bw,m_bh,m_area_factor,threshold, |
| 1643 | m_temp_d,x_coord,y_coord,nr_corners); |
| 1644 | } |
| 1645 | |
| 1646 | db_CornerDetector_u::db_CornerDetector_u() |
| 1647 | { |
| 1648 | m_w=0; m_h=0; |
| 1649 | } |
| 1650 | |
| 1651 | db_CornerDetector_u::~db_CornerDetector_u() |
| 1652 | { |
| 1653 | Clean(); |
| 1654 | } |
| 1655 | |
| 1656 | db_CornerDetector_u::db_CornerDetector_u(const db_CornerDetector_u& cd) |
| 1657 | { |
| 1658 | Start(cd.m_w, cd.m_h, cd.m_bw, cd.m_bh, cd.m_area_factor, |
| 1659 | cd.m_a_thresh, cd.m_r_thresh); |
| 1660 | } |
| 1661 | |
| 1662 | db_CornerDetector_u& db_CornerDetector_u::operator=(const db_CornerDetector_u& cd) |
| 1663 | { |
| 1664 | if ( this == &cd ) return *this; |
| 1665 | |
| 1666 | Clean(); |
| 1667 | |
| 1668 | Start(cd.m_w, cd.m_h, cd.m_bw, cd.m_bh, cd.m_area_factor, |
| 1669 | cd.m_a_thresh, cd.m_r_thresh); |
| 1670 | |
| 1671 | return *this; |
| 1672 | } |
| 1673 | |
| 1674 | void db_CornerDetector_u::Clean() |
| 1675 | { |
| 1676 | if(m_w!=0) |
| 1677 | { |
| 1678 | delete [] m_temp_i; |
| 1679 | delete [] m_temp_d; |
| 1680 | db_FreeStrengthImage_f(m_strength_mem,m_strength,m_h); |
| 1681 | } |
| 1682 | m_w=0; m_h=0; |
| 1683 | } |
| 1684 | |
| 1685 | unsigned long db_CornerDetector_u::Init(int im_width,int im_height,int target_nr_corners, |
| 1686 | int nr_horizontal_blocks,int nr_vertical_blocks, |
| 1687 | double absolute_threshold,double relative_threshold) |
| 1688 | { |
| 1689 | int block_width,block_height; |
| 1690 | unsigned long area_factor; |
| 1691 | int active_width,active_height; |
| 1692 | |
| 1693 | active_width=db_maxi(1,im_width-10); |
| 1694 | active_height=db_maxi(1,im_height-10); |
| 1695 | block_width=db_maxi(1,active_width/nr_horizontal_blocks); |
| 1696 | block_height=db_maxi(1,active_height/nr_vertical_blocks); |
| 1697 | |
| 1698 | area_factor=db_minl(1000,db_maxl(1,(long)(10000.0*((double)target_nr_corners)/ |
| 1699 | (((double)active_width)*((double)active_height))))); |
| 1700 | |
| 1701 | return(Start(im_width,im_height,block_width,block_height,area_factor, |
| 1702 | 16.0*absolute_threshold,relative_threshold)); |
| 1703 | } |
| 1704 | |
| 1705 | unsigned long db_CornerDetector_u::Start(int im_width,int im_height, |
| 1706 | int block_width,int block_height,unsigned long area_factor, |
| 1707 | double absolute_threshold,double relative_threshold) |
| 1708 | { |
| 1709 | Clean(); |
| 1710 | |
| 1711 | m_w=im_width; |
| 1712 | m_h=im_height; |
| 1713 | m_bw=block_width; |
| 1714 | m_bh=block_height; |
| 1715 | m_area_factor=area_factor; |
| 1716 | m_r_thresh=relative_threshold; |
| 1717 | m_a_thresh=absolute_threshold; |
| 1718 | m_max_nr=db_maxl(1,1+(m_w*m_h*m_area_factor)/10000); |
| 1719 | |
| 1720 | m_temp_i=new int[18*128]; |
| 1721 | m_temp_d=new double[5*m_bw*m_bh]; |
| 1722 | m_strength=db_AllocStrengthImage_f(&m_strength_mem,m_w,m_h); |
| 1723 | |
| 1724 | return(m_max_nr); |
| 1725 | } |
| 1726 | |
| 1727 | void db_CornerDetector_u::DetectCorners(const unsigned char * const *img,double *x_coord,double *y_coord,int *nr_corners, |
| 1728 | const unsigned char * const *msk, unsigned char fgnd) const |
| 1729 | { |
| 1730 | float max_val,threshold; |
| 1731 | |
| 1732 | db_HarrisStrength_u(m_strength,img,m_w,m_h,m_temp_i); |
| 1733 | |
| 1734 | |
| 1735 | if(m_r_thresh) |
| 1736 | { |
| 1737 | max_val=db_MaxImage_Aligned16_f(m_strength,3,3,m_w-6,m_h-6); |
| 1738 | threshold= (float) db_maxd(m_a_thresh,max_val*m_r_thresh); |
| 1739 | } |
| 1740 | else threshold= (float) m_a_thresh; |
| 1741 | |
| 1742 | db_ExtractCornersSaturated(m_strength,BORDER,BORDER,m_w-BORDER-1,m_h-BORDER-1,m_bw,m_bh,m_area_factor,threshold, |
| 1743 | m_temp_d,x_coord,y_coord,nr_corners); |
| 1744 | |
| 1745 | |
| 1746 | if ( msk ) |
| 1747 | { |
| 1748 | int nr_corners_mask=0; |
| 1749 | |
| 1750 | for ( int i = 0; i < *nr_corners; ++i) |
| 1751 | { |
| 1752 | int cor_x = db_roundi(*(x_coord+i)); |
| 1753 | int cor_y = db_roundi(*(y_coord+i)); |
| 1754 | if ( msk[cor_y][cor_x] == fgnd ) |
| 1755 | { |
| 1756 | x_coord[nr_corners_mask] = x_coord[i]; |
| 1757 | y_coord[nr_corners_mask] = y_coord[i]; |
| 1758 | nr_corners_mask++; |
| 1759 | } |
| 1760 | } |
| 1761 | *nr_corners = nr_corners_mask; |
| 1762 | } |
| 1763 | } |
| 1764 | |
| 1765 | void db_CornerDetector_u::ExtractCorners(float ** strength, double *x_coord, double *y_coord, int *nr_corners) { |
| 1766 | if ( m_w!=0 ) |
| 1767 | db_ExtractCornersSaturated(strength,BORDER,BORDER,m_w-BORDER-1,m_h-BORDER-1,m_bw,m_bh,m_area_factor,float(m_a_thresh), |
| 1768 | m_temp_d,x_coord,y_coord,nr_corners); |
| 1769 | } |
| 1770 | |