Fulguro: /home/christophe/src/fulguro/trunk/src/core/flgrCoreCompareFast.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 #ifdef __SSE2__
00022 #include <emmintrin.h>
00023 #endif
00024 
00025 #if defined(__SSE__) && defined(__MMX__)
00026 #include <xmmintrin.h>
00027 #include <mmintrin.h>
00028 #endif
00029 
00030 #include "flgrCoreData.h"
00031 #include "flgrCoreMalloc.h"
00032 #include "flgrCoreDispatch.h"
00033 
00034 
00035 int flgr1d_compare_eq_fast_fgUINT8(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00036 #ifdef __SSE2__
00037   int i, nbvector = dat1->length/16;
00038   fgUINT8 array_cmp[16] __attribute__ ((aligned (16)));
00039   fgUINT8 *psrc1 = (fgUINT8*) dat1->array;
00040   fgUINT8 *psrc2 = (fgUINT8*) dat2->array;
00041   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00042   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00043   __m128i vectest,vectmp;
00044   fgUINT8 cmp;
00045   
00046   
00047 
00048   i=dat1->length;
00049   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00050   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00051   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00052   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00053   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00054   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00055   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00056   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00057  
00058   vectest = _mm_cmpeq_epi8(vec_psrc1[0], vec_psrc2[0]);
00059   for(i=1 ; i<=nbvector ; i++) {
00060     vectmp = _mm_cmpeq_epi8(vec_psrc1[i], vec_psrc2[i]);
00061     vectest = _mm_and_si128(vectest,vectmp);
00062   }
00063   
00064   *((__m128i*) array_cmp) = vectest;
00065 
00066   cmp = (array_cmp[0] != 0);
00067 
00068   for(i=1 ; i<16 ; i++) {
00069     cmp &= (array_cmp[i] != 0);
00070   }
00071 
00072   if(cmp != 0) 
00073     return FLGR_TRUE;
00074   else 
00075     return FLGR_FALSE;
00076 
00077 #else
00078 #if defined(__MMX__) && (__SSE__)
00079   int i, nbvector = dat1->length/8;
00080   fgUINT8 array_cmp[8] __attribute__ ((aligned (16)));
00081   fgUINT8 *psrc1 = (fgUINT8*) dat1->array;
00082   fgUINT8 *psrc2 = (fgUINT8*) dat2->array;
00083   __m64 *vec_psrc1 = (__m64*) dat1->array;
00084   __m64 *vec_psrc2 = (__m64*) dat2->array;
00085   __m64 vectest,vectmp;
00086   fgUINT8 cmp;
00087   
00088   
00089 
00090   i=dat1->length;
00091   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00092   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00093   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00094   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00095  
00096   vectest = _mm_cmpeq_pi8(vec_psrc1[0], vec_psrc2[0]);
00097   for(i=1 ; i<=nbvector ; i++) {
00098     vectmp = _mm_cmpeq_pi8(vec_psrc1[i], vec_psrc2[i]);
00099     vectest = _mm_and_si64(vectest,vectmp);
00100   }
00101   
00102   *((__m64*) array_cmp) = vectest;
00103 
00104   cmp = (array_cmp[0] != 0);
00105 
00106   for(i=1 ; i<8 ; i++) {
00107     cmp &= (array_cmp[i] != 0);
00108   }
00109 
00110   if(cmp != 0) 
00111     return FLGR_TRUE;
00112   else 
00113     return FLGR_FALSE;
00114 
00115 #else
00116   
00117   POST_ERROR("SSE2 not activated\n");
00118   return FLGR_RET_NOT_IMPLEMENTED;
00119 #endif
00120 #endif
00121 }
00122 
00123 
00124 int flgr1d_compare_eq_fast_fgUINT16(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00125 #ifdef __SSE2__
00126   int i, nbvector = dat1->length/8;
00127   fgUINT16 array_cmp[8] __attribute__ ((aligned (16)));
00128   fgUINT16 *psrc1 = (fgUINT16*) dat1->array;
00129   fgUINT16 *psrc2 = (fgUINT16*) dat2->array;
00130   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00131   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00132   __m128i vectest,vectmp;
00133   fgUINT16 cmp;
00134   
00135   
00136 
00137   i=dat1->length;
00138   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00139   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00140   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00141   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00142  
00143   vectest = _mm_cmpeq_epi16(vec_psrc1[0], vec_psrc2[0]);
00144   for(i=1 ; i<=nbvector ; i++) {
00145     vectmp = _mm_cmpeq_epi16(vec_psrc1[i], vec_psrc2[i]);
00146     vectest = _mm_and_si128(vectest,vectmp);
00147   }
00148   
00149   *((__m128i*) array_cmp) = vectest;
00150 
00151   cmp = (array_cmp[0] != 0);
00152 
00153   for(i=1 ; i<8 ; i++) {
00154     cmp &= (array_cmp[i] != 0);
00155   }
00156 
00157   if(cmp != 0) 
00158     return FLGR_TRUE;
00159   else 
00160     return FLGR_FALSE;
00161 
00162 #else
00163 #if defined(__MMX__) && (__SSE__)
00164   int i, nbvector = dat1->length/4;
00165   fgUINT16 array_cmp[8] __attribute__ ((aligned (16)));
00166   fgUINT16 *psrc1 = (fgUINT16*) dat1->array;
00167   fgUINT16 *psrc2 = (fgUINT16*) dat2->array;
00168   __m64 *vec_psrc1 = (__m64*) dat1->array;
00169   __m64 *vec_psrc2 = (__m64*) dat2->array;
00170   __m64 vectest,vectmp;
00171   fgUINT16 cmp;
00172   
00173   
00174 
00175   i=dat1->length;
00176   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00177   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00178  
00179   vectest = _mm_cmpeq_pi16(vec_psrc1[0], vec_psrc2[0]);
00180   for(i=1 ; i<=nbvector ; i++) {
00181     vectmp = _mm_cmpeq_pi16(vec_psrc1[i], vec_psrc2[i]);
00182     vectest = _mm_and_si64(vectest,vectmp);
00183   }
00184   
00185   *((__m64*) array_cmp) = vectest;
00186 
00187   cmp = (array_cmp[0] != 0);
00188 
00189   for(i=1 ; i<4 ; i++) {
00190     cmp &= (array_cmp[i] != 0);
00191   }
00192 
00193   if(cmp != 0) 
00194     return FLGR_TRUE;
00195   else 
00196     return FLGR_FALSE;
00197 
00198 #else
00199   
00200   POST_ERROR("SSE/SSE2 not activated\n");
00201   return FLGR_RET_NOT_IMPLEMENTED;
00202 #endif
00203 #endif
00204 }
00205 
00206  
00207 int flgr1d_compare_eq_fast_fgUINT32(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00208 #ifdef __SSE2__
00209   int i, nbvector = dat1->length/4;
00210   fgUINT32 array_cmp[4] __attribute__ ((aligned (16)));
00211   fgUINT32 *psrc1 = (fgUINT32*) dat1->array;
00212   fgUINT32 *psrc2 = (fgUINT32*) dat2->array;
00213   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00214   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00215   __m128i vectest,vectmp;
00216   fgUINT32 cmp;
00217   
00218   
00219 
00220   i=dat1->length;
00221   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00222   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00223  
00224   vectest = _mm_cmpeq_epi32(vec_psrc1[0], vec_psrc2[0]);
00225   for(i=1 ; i<=nbvector ; i++) {
00226     vectmp = _mm_cmpeq_epi32(vec_psrc1[i], vec_psrc2[i]);
00227     vectest = _mm_and_si128(vectest,vectmp);
00228   }
00229   
00230   *((__m128i*) array_cmp) = vectest;
00231 
00232   cmp = (array_cmp[0] != 0);
00233 
00234   for(i=1 ; i<4 ; i++) {
00235     cmp &= (array_cmp[i] != 0);
00236   }
00237 
00238   if(cmp != 0) 
00239     return FLGR_TRUE;
00240   else 
00241     return FLGR_FALSE;
00242 
00243 #else
00244 #if defined(__MMX__) && (__SSE__)
00245   int i, nbvector = dat1->length/2;
00246   fgUINT32 array_cmp[8] __attribute__ ((aligned (16)));
00247   fgUINT32 *psrc1 = (fgUINT32*) dat1->array;
00248   fgUINT32 *psrc2 = (fgUINT32*) dat2->array;
00249   __m64 *vec_psrc1 = (__m64*) dat1->array;
00250   __m64 *vec_psrc2 = (__m64*) dat2->array;
00251   __m64 vectest,vectmp;
00252   fgUINT32 cmp;
00253   
00254   
00255 
00256   i=dat1->length;
00257   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00258  
00259   vectest = _mm_cmpeq_pi32(vec_psrc1[0], vec_psrc2[0]);
00260   for(i=1 ; i<=nbvector ; i++) {
00261     vectmp = _mm_cmpeq_pi32(vec_psrc1[i], vec_psrc2[i]);
00262     vectest = _mm_and_si64(vectest,vectmp);
00263   }
00264   
00265   *((__m64*) array_cmp) = vectest;
00266 
00267   cmp = (array_cmp[0] != 0);
00268 
00269   for(i=1 ; i<2; i++) {
00270     cmp &= (array_cmp[i] != 0);
00271   }
00272 
00273   if(cmp != 0) 
00274     return FLGR_TRUE;
00275   else 
00276     return FLGR_FALSE;
00277 
00278 #else
00279   
00280   POST_ERROR("SSE/SSE2 not activated\n");
00281   return FLGR_RET_NOT_IMPLEMENTED;
00282 #endif
00283 #endif
00284 }
00285 
00286 
00287 
00288 int flgr1d_compare_eq_fast_fgINT8(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00289 #ifdef __SSE2__
00290   int i, nbvector = dat1->length/16;
00291   fgINT8 array_cmp[16] __attribute__ ((aligned (16)));
00292   fgINT8 *psrc1 = (fgINT8*) dat1->array;
00293   fgINT8 *psrc2 = (fgINT8*) dat2->array;
00294   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00295   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00296   __m128i vectest,vectmp;
00297   fgINT8 cmp;
00298   
00299   
00300 
00301   i=dat1->length;
00302   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00303   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00304   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00305   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00306   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00307   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00308   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00309   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00310  
00311   vectest = _mm_cmpeq_epi8(vec_psrc1[0], vec_psrc2[0]);
00312   for(i=1 ; i<=nbvector ; i++) {
00313     vectmp = _mm_cmpeq_epi8(vec_psrc1[i], vec_psrc2[i]);
00314     vectest = _mm_and_si128(vectest,vectmp);
00315   }
00316   
00317   *((__m128i*) array_cmp) = vectest;
00318 
00319   cmp = (array_cmp[0] != 0);
00320 
00321   for(i=1 ; i<16 ; i++) {
00322     cmp &= (array_cmp[i] != 0);
00323   }
00324 
00325   if(cmp != 0) 
00326     return FLGR_TRUE;
00327   else 
00328     return FLGR_FALSE;
00329 
00330 #else
00331 #if defined(__MMX__) && (__SSE__)
00332   int i, nbvector = dat1->length/8;
00333   fgINT8 array_cmp[8] __attribute__ ((aligned (16)));
00334   fgINT8 *psrc1 = (fgINT8*) dat1->array;
00335   fgINT8 *psrc2 = (fgINT8*) dat2->array;
00336   __m64 *vec_psrc1 = (__m64*) dat1->array;
00337   __m64 *vec_psrc2 = (__m64*) dat2->array;
00338   __m64 vectest,vectmp;
00339   fgINT8 cmp;
00340   
00341   
00342 
00343   i=dat1->length;
00344   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00345   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00346   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00347   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00348  
00349   vectest = _mm_cmpeq_pi8(vec_psrc1[0], vec_psrc2[0]);
00350   for(i=1 ; i<=nbvector ; i++) {
00351     vectmp = _mm_cmpeq_pi8(vec_psrc1[i], vec_psrc2[i]);
00352     vectest = _mm_and_si64(vectest,vectmp);
00353   }
00354   
00355   *((__m64*) array_cmp) = vectest;
00356 
00357   cmp = (array_cmp[0] != 0);
00358 
00359   for(i=1 ; i<8 ; i++) {
00360     cmp &= (array_cmp[i] != 0);
00361   }
00362 
00363   if(cmp != 0) 
00364     return FLGR_TRUE;
00365   else 
00366     return FLGR_FALSE;
00367 
00368 #else
00369   
00370   POST_ERROR("SSE2 not activated\n");
00371   return FLGR_RET_NOT_IMPLEMENTED;
00372 #endif
00373 #endif 
00374 }
00375 
00376 
00377 int flgr1d_compare_eq_fast_fgINT16(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00378 #ifdef __SSE2__
00379   int i, nbvector = dat1->length/8;
00380   fgINT16 array_cmp[8] __attribute__ ((aligned (16)));
00381   fgINT16 *psrc1 = (fgINT16*) dat1->array;
00382   fgINT16 *psrc2 = (fgINT16*) dat2->array;
00383   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00384   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00385   __m128i vectest,vectmp;
00386   fgINT16 cmp;
00387   
00388   
00389 
00390   i=dat1->length;
00391   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00392   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00393   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00394   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00395  
00396   vectest = _mm_cmpeq_epi16(vec_psrc1[0], vec_psrc2[0]);
00397   for(i=1 ; i<=nbvector ; i++) {
00398     vectmp = _mm_cmpeq_epi16(vec_psrc1[i], vec_psrc2[i]);
00399     vectest = _mm_and_si128(vectest,vectmp);
00400   }
00401   
00402   *((__m128i*) array_cmp) = vectest;
00403 
00404   cmp = (array_cmp[0] != 0);
00405 
00406   for(i=1 ; i<8 ; i++) {
00407     cmp &= (array_cmp[i] != 0);
00408   }
00409 
00410   if(cmp != 0) 
00411     return FLGR_TRUE;
00412   else 
00413     return FLGR_FALSE;
00414 
00415 #else
00416 #if defined(__MMX__) && (__SSE__)
00417   int i, nbvector = dat1->length/4;
00418   fgINT16 array_cmp[8] __attribute__ ((aligned (16)));
00419   fgINT16 *psrc1 = (fgINT16*) dat1->array;
00420   fgINT16 *psrc2 = (fgINT16*) dat2->array;
00421   __m64 *vec_psrc1 = (__m64*) dat1->array;
00422   __m64 *vec_psrc2 = (__m64*) dat2->array;
00423   __m64 vectest,vectmp;
00424   fgINT16 cmp;
00425   
00426   
00427 
00428   i=dat1->length;
00429   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00430   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00431  
00432   vectest = _mm_cmpeq_pi16(vec_psrc1[0], vec_psrc2[0]);
00433   for(i=1 ; i<=nbvector ; i++) {
00434     vectmp = _mm_cmpeq_pi16(vec_psrc1[i], vec_psrc2[i]);
00435     vectest = _mm_and_si64(vectest,vectmp);
00436   }
00437   
00438   *((__m64*) array_cmp) = vectest;
00439 
00440   cmp = (array_cmp[0] != 0);
00441 
00442   for(i=1 ; i<4 ; i++) {
00443     cmp &= (array_cmp[i] != 0);
00444   }
00445 
00446   if(cmp != 0) 
00447     return FLGR_TRUE;
00448   else 
00449     return FLGR_FALSE;
00450 #else
00451   
00452   POST_ERROR("SSE/SSE2 not activated\n");
00453   return FLGR_RET_NOT_IMPLEMENTED;
00454 #endif
00455 #endif
00456 }
00457 
00458  
00459 int flgr1d_compare_eq_fast_fgINT32(FLGR_Data1D *dat1, FLGR_Data1D *dat2) {
00460 #ifdef __SSE2__
00461   int i, nbvector = dat1->length/4;
00462   fgINT32 array_cmp[4] __attribute__ ((aligned (16)));
00463   fgINT32 *psrc1 = (fgINT32*) dat1->array;
00464   fgINT32 *psrc2 = (fgINT32*) dat2->array;
00465   __m128i *vec_psrc1 = (__m128i*) dat1->array;
00466   __m128i *vec_psrc2 = (__m128i*) dat2->array;
00467   __m128i vectest,vectmp;
00468   fgINT32 cmp;
00469   
00470   
00471 
00472   i=dat1->length;
00473   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00474   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00475  
00476   vectest = _mm_cmpeq_epi32(vec_psrc1[0], vec_psrc2[0]);
00477   for(i=1 ; i<=nbvector ; i++) {
00478     vectmp = _mm_cmpeq_epi32(vec_psrc1[i], vec_psrc2[i]);
00479     vectest = _mm_and_si128(vectest,vectmp);
00480   }
00481   
00482   *((__m128i*) array_cmp) = vectest;
00483 
00484   cmp = (array_cmp[0] != 0);
00485 
00486   for(i=1 ; i<4 ; i++) {
00487     cmp &= (array_cmp[i] != 0);
00488   }
00489 
00490   if(cmp != 0) 
00491     return FLGR_TRUE;
00492   else 
00493     return FLGR_FALSE;
00494 
00495 #else
00496 #if defined(__MMX__) && (__SSE__)
00497   int i, nbvector = dat1->length/2;
00498   fgINT32 array_cmp[8] __attribute__ ((aligned (16)));
00499   fgINT32 *psrc1 = (fgINT32*) dat1->array;
00500   fgINT32 *psrc2 = (fgINT32*) dat2->array;
00501   __m64 *vec_psrc1 = (__m64*) dat1->array;
00502   __m64 *vec_psrc2 = (__m64*) dat2->array;
00503   __m64 vectest,vectmp;
00504   fgINT32 cmp;
00505   
00506   
00507 
00508   i=dat1->length;
00509   psrc1[i]=0;psrc2[i++]=0;psrc1[i]=0;psrc2[i++]=0;
00510  
00511   vectest = _mm_cmpeq_pi32(vec_psrc1[0], vec_psrc2[0]);
00512   for(i=1 ; i<=nbvector ; i++) {
00513     vectmp = _mm_cmpeq_pi32(vec_psrc1[i], vec_psrc2[i]);
00514     vectest = _mm_and_si64(vectest,vectmp);
00515   }
00516   
00517   *((__m64*) array_cmp) = vectest;
00518 
00519   cmp = (array_cmp[0] != 0);
00520 
00521   for(i=1 ; i<2; i++) {
00522     cmp &= (array_cmp[i] != 0);
00523   }
00524 
00525   if(cmp != 0) 
00526     return FLGR_TRUE;
00527   else 
00528     return FLGR_FALSE;
00529 
00530 #else
00531   
00532   POST_ERROR("SSE/SSE2 not activated\n");
00533   return FLGR_RET_NOT_IMPLEMENTED;
00534 #endif
00535 #endif
00536 }
00537 
00538 
00539