Fulguro: /home/christophe/src/fulguro/trunk/src/morpho/flgrMorphoBaseSegmentFast.c Source File

00001 /****************************************************************
00002  * Fulguro
00003  * Copyright (C) 2004 Christophe Clienti
00004  *
00005  * This library is free software; you can redistribute it and/or
00006  * modify it under the terms of the GNU Lesser General Public
00007  * License as published by the Free Software Foundation; either
00008  * version 2 of the License, or (at your option) any later version.
00009  *
00010  * This library is distributed in the hope that it will be useful,
00011  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00012  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
00013  * Lesser General Public License for more details.
00014  *
00015  * You should have received a copy of the GNU Lesser General Public
00016  * License along with this library; if not, write to the
00017  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
00018  * Boston, MA 02111-1307, USA.
00019  ***************************************************************/
00020 
00021 
00022 #include <stdio.h>
00023 #include <stdlib.h>
00024 #include <string.h>
00025 #include <flgrCoreMalloc.h>
00026 #include <flgrCoreDispatch.h>
00027 
00028 #ifdef __SSE2__
00029 #include <emmintrin.h>
00030 #else
00031 #if defined(__SSE__) && defined(__MMX__)
00032 #include <mmintrin.h>
00033 #include <xmmintrin.h>
00034 #endif
00035 #endif
00036 
00037 #include <flgrCoreTranspose.h>
00038 #include "flgrMorphoBaseSegmentFast.h"
00039 
00040 
00041 void flgr2d_erode_segment_vertical_fast_fgUINT8(FLGR_Data2D *imgdest, FLGR_Data2D *imgsrc, int sesize_y) {
00042 #ifdef __SSE2__
00043   int i,j,k,m=0,n=0,p=0;
00044   int se2=sesize_y/2;
00045   int kpad;
00046   int rowsize_y=imgdest->size_y;
00047   int finalPass=rowsize_y%sesize_y;
00048   int nbbloc = rowsize_y/sesize_y;
00049   fgUINT8 **rowin;
00050   fgUINT8 **rowout;
00051   __m128i *bufg;
00052   __m128i *bufh;
00053   __m128i tmpg,tmph;
00054 
00055   
00056 
00057   bufg  = (__m128i *) flgr_malloc_align(sizeof(__m128i)*(rowsize_y+16),16);
00058   bufh  = (__m128i *) flgr_malloc_align(sizeof(__m128i)*(rowsize_y+16),16);
00059 
00060   rowin = (fgUINT8**) malloc(sizeof(fgUINT8*)*(imgsrc->size_y+16));
00061   rowout = (fgUINT8**) malloc(sizeof(fgUINT8*)*(rowsize_y+16));
00062 
00063   for(i=0 ; i<(rowsize_y+16) ; i++) {
00064     rowin[i] = (fgUINT8*) imgsrc->array[i];
00065     rowout[i] = (fgUINT8*) imgdest->array[i];
00066   }
00067 
00068 
00069   for(j=0 ; j<imgdest->size_x ; j+=16) {
00070 
00071     m=0;
00072     n=nbbloc*sesize_y-1;
00073 
00074     for(k=0;k<nbbloc;k++) {
00075 
00076       tmpg = bufg[m]=*((__m128i*) (rowin[m]+j));
00077       tmph = bufh[n]=*((__m128i*) (rowin[n]+j));
00078       m++;n--;
00079 
00080       for(i=1;i<sesize_y;i++,m++,n--) {
00081     bufg[m] = tmpg = _mm_min_epu8(*((__m128i*) (rowin[m]+j)),tmpg);
00082     bufh[n] = tmph = _mm_min_epu8(*((__m128i*) (rowin[n]+j)),tmph);
00083       }
00084 
00085     }
00086 
00087     n=rowsize_y-1;
00088     tmpg = bufg[m] = *((__m128i*) (rowin[m]+j));
00089     tmph = bufh[n] = *((__m128i*) (rowin[n]+j));
00090     m++;n--;
00091 
00092     for(k=1 ; k<finalPass ; k++,m++,n--) {
00093     bufg[m] = tmpg = _mm_min_epu8(*((__m128i*) (rowin[m]+j)),tmpg);
00094     bufh[n] = tmph = _mm_min_epu8(*((__m128i*) (rowin[n]+j)),tmph);
00095     }
00096 
00097     n=se2;p=0;m=0;
00098     for( i=0; i<se2 ; i++) {
00099       *((__m128i*) (rowout[m++]+j)) = bufg[n++];
00100     }
00101     for( i=se2; i<rowsize_y-se2 ;i++) {
00102       *((__m128i*) (rowout[m]+j)) = _mm_min_epu8(bufg[n], bufh[p]);
00103       n++;p++;m++;
00104     }
00105     n--;
00106     kpad = finalPass;
00107     kpad = (kpad==0) ? 0 : sesize_y-kpad;
00108     for( i=0 ; (i<kpad) && (m<rowsize_y) ; i++) {
00109       *((__m128i*) (rowout[m]+j)) = _mm_min_epu8(bufg[n], bufh[p]);
00110       m++;p++;
00111     }
00112     for( i=m ; i<rowsize_y ; i++) *((__m128i*) (rowout[i]+j)) = bufh[p++];
00113 
00114   }
00115 
00116   flgr_free_align(bufg);
00117   flgr_free_align(bufh);
00118   free(rowin);
00119   free(rowout);
00120 
00121   return;
00122 
00123 
00124 #else
00125 #if defined(__SSE__) && defined(__MMX__)
00126   int i,j,k,m=0,n=0,p=0;
00127   int kpad;
00128   int se2=sesize_y/2;
00129   int rowsize_y=imgdest->size_y;
00130   int finalPass=rowsize_y%sesize_y;
00131   int nbbloc = rowsize_y/sesize_y;
00132   fgUINT8 **rowin;
00133   fgUINT8 **rowout;
00134   __m64 *bufg;
00135   __m64 *bufh;
00136   __m64 tmpg,tmph;
00137 
00138   
00139 
00140   bufg  = (__m64 *) flgr_malloc_align(sizeof(__m64)*(rowsize_y+16),16);
00141   bufh  = (__m64 *) flgr_malloc_align(sizeof(__m64)*(rowsize_y+16),16);
00142   
00143   rowin = (fgUINT8**) malloc(sizeof(fgUINT8*)*(imgsrc->size_y+16));
00144   rowout = (fgUINT8**) malloc(sizeof(fgUINT8*)*(rowsize_y+16));
00145 
00146   for(i=0 ; i<(rowsize_y+16) ; i++) {
00147     rowin[i] = (fgUINT8*) imgsrc->array[i];
00148     rowout[i] = (fgUINT8*) imgdest->array[i];
00149   }
00150 
00151   _mm_empty();
00152 
00153 
00154   for(j=0 ; j<imgdest->size_x ; j+=8) {
00155 
00156 
00157     m=0;
00158     n=nbbloc*sesize_y-1;
00159 
00160     for(k=0;k<nbbloc;k++) {
00161 
00162       tmpg = bufg[m]=*((__m64*) (rowin[m]+j));
00163       tmph = bufh[n]=*((__m64*) (rowin[n]+j));
00164       m++;n--;
00165 
00166       for(i=1;i<sesize_y;i++,m++,n--) {
00167     bufg[m] = tmpg = _mm_min_pu8(*((__m64*) (rowin[m]+j)),tmpg);
00168     bufh[n] = tmph = _mm_min_pu8(*((__m64*) (rowin[n]+j)),tmph);    
00169       }
00170       
00171     }
00172 
00173     n=rowsize_y-1;
00174     tmpg = bufg[m] = *((__m64*) (rowin[m]+j));
00175     tmph = bufh[n] = *((__m64*) (rowin[n]+j));
00176     m++;n--;
00177 
00178     for(k=1 ; k<finalPass ; k++,m++,n--) {
00179     bufg[m] = tmpg = _mm_min_pu8(*((__m64*) (rowin[m]+j)),tmpg);
00180     bufh[n] = tmph = _mm_min_pu8(*((__m64*) (rowin[n]+j)),tmph);      
00181     }
00182 
00183     n=se2;p=0;m=0;
00184     for( i=0; i<se2 ; i++) {
00185       *((__m64*) (rowout[m++]+j)) = bufg[n++];
00186     }
00187     for( i=se2; i<rowsize_y-se2 ;i++) {
00188       *((__m64*) (rowout[m]+j)) = _mm_min_pu8(bufg[n], bufh[p]);
00189       n++;p++;m++;
00190     }
00191     n--;
00192     kpad = finalPass;
00193     kpad = (kpad==0) ? 0 : sesize_y-kpad;
00194     for( i=0 ; (i<kpad) && (m<rowsize_y) ; i++) {
00195       *((__m64*) (rowout[m]+j)) = _mm_min_pu8(bufg[n], bufh[p]);
00196       m++;p++;
00197     }
00198     for( i=m ; i<rowsize_y ; i++) *((__m64*) (rowout[i]+j)) = bufh[p++]; 
00199 
00200   }  
00201 
00202   flgr_free_align(bufg);
00203   flgr_free_align(bufh);
00204   free(rowin);
00205   free(rowout);
00206 
00207   _mm_empty();
00208   return;
00209 
00210  #else
00211    POST_ERROR("MMX/SSE/SSE2 Instruction not supported!\n");
00212  #endif
00213 #endif
00214 }
00215 
00216 
00217 
00218 void flgr2d_dilate_segment_vertical_fast_fgUINT8(FLGR_Data2D *imgdest, FLGR_Data2D *imgsrc, int sesize_y) {
00219 #ifdef __SSE2__
00220   int i,j,k,m=0,n=0,p=0;
00221   int se2=sesize_y/2;
00222   int kpad;
00223   int rowsize_y=imgdest->size_y;
00224   int finalPass=rowsize_y%sesize_y;
00225   int nbbloc = rowsize_y/sesize_y;
00226   fgUINT8 **rowin;
00227   fgUINT8 **rowout;
00228   __m128i *bufg;
00229   __m128i *bufh;
00230   __m128i tmpg,tmph;
00231 
00232   
00233 
00234   bufg  = (__m128i *) flgr_malloc_align(sizeof(__m128i)*(rowsize_y+16),16);
00235   bufh  = (__m128i *) flgr_malloc_align(sizeof(__m128i)*(rowsize_y+16),16);
00236 
00237   rowin = (fgUINT8**) malloc(sizeof(fgUINT8*)*(imgsrc->size_y+16));
00238   rowout = (fgUINT8**) malloc(sizeof(fgUINT8*)*(rowsize_y+16));
00239 
00240   for(i=0 ; i<(rowsize_y+16) ; i++) {
00241     rowin[i] = (fgUINT8*) imgsrc->array[i];
00242     rowout[i] = (fgUINT8*) imgdest->array[i];
00243   }
00244 
00245 
00246 
00247   for(j=0 ; j<imgdest->size_x ; j+=16) {
00248 
00249     m=0;
00250     n=nbbloc*sesize_y-1;
00251 
00252     for(k=0;k<nbbloc;k++) {
00253 
00254       tmpg = bufg[m]=*((__m128i*) (rowin[m]+j));
00255       tmph = bufh[n]=*((__m128i*) (rowin[n]+j));
00256       m++;n--;
00257 
00258       for(i=1;i<sesize_y;i++,m++,n--) {
00259     bufg[m] = tmpg = _mm_max_epu8(*((__m128i*) (rowin[m]+j)),tmpg);
00260     bufh[n] = tmph = _mm_max_epu8(*((__m128i*) (rowin[n]+j)),tmph);
00261       }
00262 
00263     }
00264 
00265     n=rowsize_y-1;
00266     tmpg = bufg[m] = *((__m128i*) (rowin[m]+j));
00267     tmph = bufh[n] = *((__m128i*) (rowin[n]+j));
00268     m++;n--;
00269 
00270     for(k=1 ; k<finalPass ; k++,m++,n--) {
00271     bufg[m] = tmpg = _mm_max_epu8(*((__m128i*) (rowin[m]+j)),tmpg);
00272     bufh[n] = tmph = _mm_max_epu8(*((__m128i*) (rowin[n]+j)),tmph);
00273     }
00274 
00275     n=se2;p=0;m=0;
00276     for( i=0; i<se2 ; i++) {
00277       *((__m128i*) (rowout[m++]+j)) = bufg[n++];
00278     }
00279     for( i=se2; i<rowsize_y-se2 ;i++) {
00280       *((__m128i*) (rowout[m]+j)) = _mm_max_epu8(bufg[n], bufh[p]);
00281       n++;p++;m++;
00282     }
00283     n--;
00284     kpad = finalPass;
00285     kpad = (kpad==0) ? 0 : sesize_y-kpad;
00286     for( i=0 ; (i<kpad) && (m<rowsize_y) ; i++) {
00287       *((__m128i*) (rowout[m]+j)) = _mm_max_epu8(bufg[n], bufh[p]);
00288       m++;p++;
00289     }
00290     for( i=m ; i<rowsize_y ; i++) *((__m128i*) (rowout[i]+j)) = bufh[p++];
00291 
00292   }
00293 
00294   flgr_free_align(bufg);
00295   flgr_free_align(bufh);
00296   free(rowin);
00297   free(rowout);
00298 
00299   return;
00300 
00301 #else
00302 #if defined(__SSE__) && defined(__MMX__)
00303   int i,j,k,m=0,n=0,p=0;
00304   int kpad;
00305   int se2=sesize_y/2;
00306   int rowsize_y=imgdest->size_y;
00307   int finalPass=rowsize_y%sesize_y;
00308   int nbbloc = rowsize_y/sesize_y;
00309   fgUINT8 **rowin;
00310   fgUINT8 **rowout;
00311   __m64 *bufg;
00312   __m64 *bufh;
00313   __m64 tmpg,tmph;
00314 
00315   
00316 
00317   bufg  = (__m64 *) flgr_malloc_align(sizeof(__m64)*(rowsize_y+16),16);
00318   bufh  = (__m64 *) flgr_malloc_align(sizeof(__m64)*(rowsize_y+16),16);
00319   
00320   rowin = (fgUINT8**) malloc(sizeof(fgUINT8*)*(imgsrc->size_y+16));
00321   rowout = (fgUINT8**) malloc(sizeof(fgUINT8*)*(rowsize_y+16));
00322 
00323   for(i=0 ; i<(rowsize_y+16) ; i++) {
00324     rowin[i] = (fgUINT8*) imgsrc->array[i];
00325     rowout[i] = (fgUINT8*) imgdest->array[i];
00326   }
00327 
00328   _mm_empty();
00329 
00330 
00331   for(j=0 ; j<imgdest->size_x ; j+=8) {
00332 
00333 
00334     m=0;
00335     n=nbbloc*sesize_y-1;
00336 
00337     for(k=0;k<nbbloc;k++) {
00338 
00339       tmpg = bufg[m]=*((__m64*) (rowin[m]+j));
00340       tmph = bufh[n]=*((__m64*) (rowin[n]+j));
00341       m++;n--;
00342 
00343       for(i=1;i<sesize_y;i++,m++,n--) {
00344     bufg[m] = tmpg = _mm_max_pu8(*((__m64*) (rowin[m]+j)),tmpg);
00345     bufh[n] = tmph = _mm_max_pu8(*((__m64*) (rowin[n]+j)),tmph);    
00346       }
00347       
00348     }
00349 
00350     n=rowsize_y-1;
00351     tmpg = bufg[m] = *((__m64*) (rowin[m]+j));
00352     tmph = bufh[n] = *((__m64*) (rowin[n]+j));
00353     m++;n--;
00354 
00355     for(k=1 ; k<finalPass ; k++,m++,n--) {
00356     bufg[m] = tmpg = _mm_max_pu8(*((__m64*) (rowin[m]+j)),tmpg);
00357     bufh[n] = tmph = _mm_max_pu8(*((__m64*) (rowin[n]+j)),tmph);      
00358     }
00359 
00360     n=se2;p=0;m=0;
00361     for( i=0; i<se2 ; i++) {
00362       *((__m64*) (rowout[m++]+j)) = bufg[n++];
00363     }
00364     for( i=se2; i<rowsize_y-se2 ;i++) {
00365       *((__m64*) (rowout[m]+j)) = _mm_max_pu8(bufg[n], bufh[p]);
00366       n++;p++;m++;
00367     }
00368     n--;
00369     kpad = finalPass;
00370     kpad = (kpad==0) ? 0 : sesize_y-kpad;
00371     for( i=0 ; (i<kpad) && (m<rowsize_y) ; i++) {
00372       *((__m64*) (rowout[m]+j)) = _mm_max_pu8(bufg[n], bufh[p]);
00373       m++;p++;
00374     }
00375     for( i=m ; i<rowsize_y ; i++) *((__m64*) (rowout[i]+j)) = bufh[p++]; 
00376 
00377   }  
00378 
00379   flgr_free_align(bufg);
00380   flgr_free_align(bufh);
00381   free(rowin);
00382   free(rowout);
00383 
00384   _mm_empty();
00385   return ;
00386 
00387 #else
00388   POST_ERROR("MMX/SSE/SSE2 Instruction not supported!\n");
00389 #endif
00390 #endif
00391 }
00392 
00393 
00394 
00395 void flgr2d_erode_segment_horizontal_fast_fgUINT8(FLGR_Data2D *imgdest, FLGR_Data2D *imgsrc, int sesize_x) {
00396   FLGR_Data2D *tmpdest,*tmpsrc;
00397 
00398   
00399 
00400   tmpdest = flgr2d_create_pixmap(imgdest->size_x,imgdest->size_y,imgsrc->spp,imgdest->type);
00401   tmpsrc = flgr2d_create_pixmap(imgdest->size_x,imgdest->size_y,imgsrc->spp,imgdest->type);
00402 
00403   flgr2d_transpose_fgUINT8(tmpsrc,imgsrc);
00404   flgr2d_erode_segment_vertical_fast_fgUINT8(tmpdest,tmpsrc,sesize_x);
00405   flgr2d_transpose_fgUINT8(imgdest,tmpdest);
00406 
00407   flgr2d_destroy(tmpdest);
00408   flgr2d_destroy(tmpsrc);
00409 
00410   return ;
00411 }
00412 
00413 
00414 void flgr2d_dilate_segment_horizontal_fast_fgUINT8(FLGR_Data2D *imgdest, FLGR_Data2D *imgsrc, int sesize_x) {
00415   FLGR_Data2D *tmpdest,*tmpsrc;
00416 
00417   
00418 
00419   tmpdest = flgr2d_create_pixmap(imgdest->size_x,imgdest->size_y,imgsrc->spp,imgdest->type);
00420   tmpsrc = flgr2d_create_pixmap(imgdest->size_x,imgdest->size_y,imgsrc->spp,imgdest->type);
00421 
00422   flgr2d_transpose_fgUINT8(tmpsrc,imgsrc);
00423   flgr2d_dilate_segment_vertical_fast_fgUINT8(tmpdest,tmpsrc,sesize_x);
00424   flgr2d_transpose_fgUINT8(imgdest,tmpdest);
00425 
00426   flgr2d_destroy(tmpdest);
00427   flgr2d_destroy(tmpsrc);
00428 
00429   return ;
00430 
00431 }