libcm is a C development framework with an emphasis on audio signal processing applications.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

cmVectOpsRICode.h 29KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248
  1. #ifdef cmVectOpsRICode_h
  2. VECT_OP_TYPE* VECT_OP_FUNC(Col)( VECT_OP_TYPE* m, unsigned ci, unsigned rn, unsigned cn )
  3. {
  4. assert(ci<cn);
  5. return m + (ci*rn);
  6. }
  7. VECT_OP_TYPE* VECT_OP_FUNC(Row)( VECT_OP_TYPE* m, unsigned ri, unsigned rn, unsigned cn )
  8. {
  9. assert(ri<rn);
  10. return m + ri;
  11. }
  12. VECT_OP_TYPE* VECT_OP_FUNC(ElePtr)( VECT_OP_TYPE* m, unsigned ri, unsigned ci, unsigned rn, unsigned cn )
  13. {
  14. assert(ri<rn && ci<cn);
  15. return m + (ci*rn) + ri;
  16. }
  17. VECT_OP_TYPE VECT_OP_FUNC(Ele)( VECT_OP_TYPE* m, unsigned ri, unsigned ci, unsigned rn, unsigned cn )
  18. { return *VECT_OP_FUNC(ElePtr)(m,ri,ci,rn,cn); }
  19. void VECT_OP_FUNC(Set)( VECT_OP_TYPE* m, unsigned ri, unsigned ci, unsigned rn, unsigned cn, VECT_OP_TYPE v )
  20. { *(VECT_OP_FUNC(ElePtr)(m,ri,ci,rn,cn)) = v; }
  21. const VECT_OP_TYPE* VECT_OP_FUNC(CCol)( const VECT_OP_TYPE* m, unsigned ci, unsigned rn, unsigned cn )
  22. {
  23. assert(ci<cn);
  24. return m + (ci*rn);
  25. }
  26. const VECT_OP_TYPE* VECT_OP_FUNC(CRow)( const VECT_OP_TYPE* m, unsigned ri, unsigned rn, unsigned cn )
  27. {
  28. assert(ri<rn);
  29. return m + ri;
  30. }
  31. const VECT_OP_TYPE* VECT_OP_FUNC(CElePtr)( const VECT_OP_TYPE* m, unsigned ri, unsigned ci, unsigned rn, unsigned cn )
  32. {
  33. assert(ri<rn && ci<cn);
  34. return m + (ci*rn) + ri;
  35. }
  36. VECT_OP_TYPE VECT_OP_FUNC(CEle)( const VECT_OP_TYPE* m, unsigned ri, unsigned ci, unsigned rn, unsigned cn )
  37. { return *VECT_OP_FUNC(CElePtr)(m,ri,ci,rn,cn); }
  38. VECT_OP_TYPE* VECT_OP_FUNC(Fill)( VECT_OP_TYPE* dbp, unsigned dn, VECT_OP_TYPE value )
  39. {
  40. const VECT_OP_TYPE* dep = dbp + dn;
  41. VECT_OP_TYPE* dp = dbp;
  42. if( value == 0 )
  43. memset(dbp,0,(dep-dbp)*sizeof(VECT_OP_TYPE));
  44. else
  45. {
  46. while( dbp < dep )
  47. *dbp++ = value;
  48. }
  49. return dp;
  50. }
  51. VECT_OP_TYPE* VECT_OP_FUNC(Zero)( VECT_OP_TYPE* dbp, unsigned dn )
  52. {
  53. memset( dbp, 0, sizeof(VECT_OP_TYPE)*dn);
  54. return dbp;
  55. }
  56. VECT_OP_TYPE* VECT_OP_FUNC(Move)( VECT_OP_TYPE* bp, unsigned bn, const VECT_OP_TYPE* sp )
  57. {
  58. memmove(bp,sp,sizeof(VECT_OP_TYPE)*bn);
  59. return bp;
  60. }
  61. VECT_OP_TYPE* VECT_OP_FUNC(Copy)( VECT_OP_TYPE* bp, unsigned bn, const VECT_OP_TYPE* sp )
  62. {
  63. memcpy(bp,sp,sizeof(VECT_OP_TYPE)*bn);
  64. return bp;
  65. }
  66. VECT_OP_TYPE* VECT_OP_FUNC(CopyN)( VECT_OP_TYPE* bp, unsigned bn, unsigned d_stride, const VECT_OP_TYPE* sp, unsigned s_stride )
  67. {
  68. VECT_OP_TYPE* dbp = bp;
  69. const VECT_OP_TYPE* ep = bp + (bn*d_stride);
  70. for(; bp < ep; bp += d_stride, sp += s_stride )
  71. *bp = *sp;
  72. return dbp;
  73. }
  74. VECT_OP_TYPE* VECT_OP_FUNC(CopyU)( VECT_OP_TYPE* bp, unsigned bn, const unsigned* sp )
  75. {
  76. VECT_OP_TYPE* dbp = bp;
  77. const VECT_OP_TYPE* ep = bp + bn;
  78. VECT_OP_TYPE* dp = bp;
  79. while( dp < ep )
  80. *dp++ = (VECT_OP_TYPE)*sp++;
  81. return dbp;
  82. }
  83. VECT_OP_TYPE* VECT_OP_FUNC(CopyI)( VECT_OP_TYPE* dbp, unsigned dn, const int* sp )
  84. {
  85. const VECT_OP_TYPE* dep = dbp + dn;
  86. VECT_OP_TYPE* dp = dbp;
  87. while( dp < dep )
  88. *dp++ = (VECT_OP_TYPE)*sp++;
  89. return dbp;
  90. }
  91. VECT_OP_TYPE* VECT_OP_FUNC(CopyF)( VECT_OP_TYPE* dbp, unsigned dn, const float* sp )
  92. {
  93. const VECT_OP_TYPE* dep = dbp + dn;
  94. VECT_OP_TYPE* dp = dbp;
  95. while( dp < dep )
  96. *dp++ = (VECT_OP_TYPE)*sp++;
  97. return dbp;
  98. }
  99. VECT_OP_TYPE* VECT_OP_FUNC(CopyD)( VECT_OP_TYPE* dbp, unsigned dn, const double* sp )
  100. {
  101. const VECT_OP_TYPE* dep = dbp + dn;
  102. VECT_OP_TYPE* dp = dbp;
  103. while( dp < dep )
  104. *dp++ = (VECT_OP_TYPE)*sp++;
  105. return dbp;
  106. }
  107. VECT_OP_TYPE* VECT_OP_FUNC(CopyS)( VECT_OP_TYPE* dbp, unsigned dn, const cmSample_t* sp )
  108. {
  109. const VECT_OP_TYPE* dep = dbp + dn;
  110. VECT_OP_TYPE* dp = dbp;
  111. while( dp < dep )
  112. *dp++ = (VECT_OP_TYPE)*sp++;
  113. return dbp;
  114. }
  115. VECT_OP_TYPE* VECT_OP_FUNC(CopyR)( VECT_OP_TYPE* dbp, unsigned dn, const cmReal_t* sp )
  116. {
  117. const VECT_OP_TYPE* dep = dbp + dn;
  118. VECT_OP_TYPE* dp = dbp;
  119. while( dp < dep )
  120. *dp++ = (VECT_OP_TYPE)*sp++;
  121. return dbp;
  122. }
  123. VECT_OP_TYPE* VECT_OP_FUNC(CopyStride)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sp, unsigned srcStride )
  124. {
  125. const VECT_OP_TYPE* dep = dbp + dn;
  126. VECT_OP_TYPE* dp = dbp;
  127. for(; dp < dep; sp += srcStride )
  128. *dp++ = *sp;
  129. return dbp;
  130. }
  131. VECT_OP_TYPE* VECT_OP_FUNC(Shrink)( VECT_OP_TYPE* s, unsigned sn, const VECT_OP_TYPE* t, unsigned tn )
  132. {
  133. assert( s <= t && t <= (s+sn) );
  134. assert( s <= (t+tn) && (t+tn) <= (s+sn));
  135. //VECT_OP_FUNC(Move)(s,sn - ((t - s) + tn),t+tn);
  136. VECT_OP_FUNC(Move)((VECT_OP_TYPE*)t,(sn - ((t+tn)-s)) + 1,t+tn);
  137. return s;
  138. }
  139. VECT_OP_TYPE* VECT_OP_FUNC(Expand)( VECT_OP_TYPE* s, unsigned sn, const VECT_OP_TYPE* t, unsigned tn )
  140. {
  141. assert( s <= t && t <= s+sn );
  142. unsigned i = t - s;
  143. s = cmMemResizeP(VECT_OP_TYPE,s,sn+tn);
  144. t = s + i;
  145. assert( t + tn + sn - i == s + sn + tn );
  146. VECT_OP_FUNC(Move)(((VECT_OP_TYPE*)t)+tn,sn-i,t);
  147. return s;
  148. }
  149. VECT_OP_TYPE* VECT_OP_FUNC(Replace)(VECT_OP_TYPE* s, unsigned* sn, const VECT_OP_TYPE* t, unsigned tn, const VECT_OP_TYPE* u, unsigned un )
  150. {
  151. // if s is empty and t[tn] is empty
  152. if( s == NULL && tn == 0 )
  153. {
  154. if( un == 0 )
  155. return s;
  156. s = cmMemAllocZ(VECT_OP_TYPE,un);
  157. VECT_OP_FUNC(Copy)(s,un,u);
  158. if( sn != NULL )
  159. *sn = un;
  160. return s;
  161. }
  162. assert( s!=NULL && t != NULL );
  163. assert( (u!=NULL && un>0) || (u==NULL && un==0) );
  164. if( (tn==0 && un==0) || (t==NULL && u==NULL))
  165. return s;
  166. // if the area to replace is greater than the area to insert ...
  167. if( tn > un )
  168. {
  169. VECT_OP_FUNC(Shrink)(s,*sn,t+un,tn-un); // ... then shrink the buffer
  170. *sn -= tn-un;
  171. }
  172. else
  173. // if the area to insert is greater than the area to replace ...
  174. if( un > tn )
  175. {
  176. unsigned offs = t - s;
  177. s = VECT_OP_FUNC(Expand)(s,*sn,t+tn,un-tn); // ... then expand the buffer
  178. t = s + offs;
  179. *sn += un-tn;
  180. }
  181. assert(t+un <= s+(*sn));
  182. if( u!=NULL )
  183. VECT_OP_FUNC(Copy)((VECT_OP_TYPE*)t,un,u);
  184. return s;
  185. }
  186. VECT_OP_TYPE* VECT_OP_FUNC(Rotate)( VECT_OP_TYPE* dbp, unsigned dn, int shiftCnt )
  187. {
  188. VECT_OP_TYPE* dep = dbp + dn;
  189. int i = 0;
  190. unsigned k = 0;
  191. int n = dep - dbp;
  192. VECT_OP_TYPE t1 = dbp[i];
  193. for(k=0; k<n; ++k)
  194. {
  195. int j;
  196. j = (i + shiftCnt) % n;
  197. if( j<0 )
  198. j += n;
  199. VECT_OP_TYPE t2 = dbp[j];
  200. dbp[j] = t1;
  201. t1 = t2;
  202. i = j;
  203. }
  204. return dbp;
  205. }
  206. VECT_OP_TYPE* VECT_OP_FUNC(RotateM)( VECT_OP_TYPE* dbp, unsigned drn, unsigned dcn, const VECT_OP_TYPE* sbp, int rShiftCnt, int cShiftCnt )
  207. {
  208. int j;
  209. while( rShiftCnt < 0 )
  210. rShiftCnt += drn;
  211. while( cShiftCnt < 0 )
  212. cShiftCnt += dcn;
  213. int m = rShiftCnt % drn;
  214. int n = cShiftCnt % dcn;
  215. for(j=0; j<dcn; ++j,++n)
  216. {
  217. if(n==dcn)
  218. n = 0;
  219. // cnt from dst position to end of column
  220. unsigned cn = drn - m;
  221. // copy from top of src col to bottom of dst column
  222. VECT_OP_FUNC(Copy)(dbp + (n*drn) + m, cn, sbp );
  223. sbp+=cn;
  224. if( cn < drn )
  225. {
  226. // copy from bottom of src col to top of dst column
  227. VECT_OP_FUNC(Copy)(dbp + (n*drn), drn-cn, sbp );
  228. sbp += drn-cn;
  229. }
  230. }
  231. return dbp;
  232. }
  233. VECT_OP_TYPE* VECT_OP_FUNC(Shift)( VECT_OP_TYPE* dbp, unsigned dn, int shiftCnt, VECT_OP_TYPE fillValue )
  234. {
  235. VECT_OP_TYPE* dep = dbp + dn;
  236. VECT_OP_TYPE* rp = dbp;
  237. unsigned n = dep - dbp;
  238. if( shiftCnt == 0 )
  239. return dbp;
  240. if( abs(shiftCnt) >= n )
  241. return VECT_OP_FUNC(Fill)(dbp,dn,fillValue);
  242. if( shiftCnt > 0 )
  243. {
  244. const VECT_OP_TYPE* sbp = dep - (shiftCnt+1);
  245. const VECT_OP_TYPE* sep = dbp;
  246. VECT_OP_TYPE* dp = dbp + (n-1);
  247. while( sbp >= sep )
  248. *dp-- = *sbp--;
  249. while(dbp <= dp )
  250. *dbp++ = fillValue;
  251. }
  252. else
  253. {
  254. const VECT_OP_TYPE* sbp = dbp + abs(shiftCnt);
  255. while( sbp < dep )
  256. *dbp++ = *sbp++;
  257. while(dbp<dep)
  258. *dbp++ = fillValue;
  259. }
  260. return rp;
  261. }
  262. VECT_OP_TYPE* VECT_OP_FUNC(Flip)( VECT_OP_TYPE* dbp, unsigned dn)
  263. {
  264. VECT_OP_TYPE* p0 = dbp;
  265. VECT_OP_TYPE* p1 = dbp + dn - 1;
  266. while( p0 < p1 )
  267. {
  268. VECT_OP_TYPE t = *p0;
  269. *p0++ = *p1;
  270. *p1-- = t;
  271. }
  272. return dbp;
  273. }
  274. VECT_OP_TYPE* VECT_OP_FUNC(SubVS)( VECT_OP_TYPE* bp, unsigned n, VECT_OP_TYPE v )
  275. {
  276. const VECT_OP_TYPE* ep = bp + n;
  277. VECT_OP_TYPE* dp = bp;
  278. while( dp < ep )
  279. *dp++ -= v;
  280. return bp;
  281. }
  282. VECT_OP_TYPE* VECT_OP_FUNC(SubVV)( VECT_OP_TYPE* bp, unsigned n, const VECT_OP_TYPE* v )
  283. {
  284. const VECT_OP_TYPE* ep = bp + n;
  285. VECT_OP_TYPE* dp = bp;
  286. while( dp < ep )
  287. *dp++ -= *v++;
  288. return bp;
  289. }
  290. VECT_OP_TYPE* VECT_OP_FUNC(SubVVS)( VECT_OP_TYPE* bp, unsigned n, const VECT_OP_TYPE* v, VECT_OP_TYPE s )
  291. {
  292. const VECT_OP_TYPE* ep = bp + n;
  293. VECT_OP_TYPE* dp = bp;
  294. while( dp < ep )
  295. *dp++ = *v++ - s;
  296. return bp;
  297. }
  298. VECT_OP_TYPE* VECT_OP_FUNC(SubVVNN)(VECT_OP_TYPE* dp, unsigned dn, unsigned dnn, const VECT_OP_TYPE* v, unsigned n )
  299. {
  300. const VECT_OP_TYPE* ep = dp + (dn*dnn);
  301. VECT_OP_TYPE* dbp = dp;
  302. for(; dp < ep; dp+=dnn, v+=n )
  303. *dp -= *v;
  304. return dbp;
  305. }
  306. VECT_OP_TYPE* VECT_OP_FUNC(SubVVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, const VECT_OP_TYPE* sb1p )
  307. {
  308. const VECT_OP_TYPE* dep = dbp + dn;
  309. VECT_OP_TYPE* dp = dbp;
  310. while( dbp < dep )
  311. *dbp++ = *sb0p++ - *sb1p++;
  312. return dp;
  313. }
  314. VECT_OP_TYPE* VECT_OP_FUNC(SubVSV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE s0, const VECT_OP_TYPE* sb1p )
  315. {
  316. const VECT_OP_TYPE* dep = dbp + dn;
  317. VECT_OP_TYPE* dp = dbp;
  318. while( dbp < dep )
  319. *dbp++ = s0 - *sb1p++;
  320. return dp;
  321. }
  322. VECT_OP_TYPE* VECT_OP_FUNC(AddVS)( VECT_OP_TYPE* bp, unsigned n, VECT_OP_TYPE v )
  323. {
  324. const VECT_OP_TYPE* ep = bp + n;
  325. VECT_OP_TYPE* dp = bp;
  326. while( dp < ep )
  327. *dp++ += v;
  328. return bp;
  329. }
  330. VECT_OP_TYPE* VECT_OP_FUNC(AddVV)( VECT_OP_TYPE* bp, unsigned bn, const VECT_OP_TYPE* v )
  331. {
  332. const VECT_OP_TYPE* ep = bp + bn;
  333. VECT_OP_TYPE* dp = bp;
  334. while( dp < ep )
  335. *dp++ += *v++;
  336. return bp;
  337. }
  338. VECT_OP_TYPE* VECT_OP_FUNC(AddVVS)( VECT_OP_TYPE* bp, unsigned bn, const VECT_OP_TYPE* v, VECT_OP_TYPE s )
  339. {
  340. const VECT_OP_TYPE* ep = bp + bn;
  341. VECT_OP_TYPE* dp = bp;
  342. while( dp < ep )
  343. *dp++ = *v++ + s;
  344. return bp;
  345. }
  346. VECT_OP_TYPE* VECT_OP_FUNC(AddVVNN)(VECT_OP_TYPE* dp, unsigned dn, unsigned dnn, const VECT_OP_TYPE* v, unsigned n )
  347. {
  348. const VECT_OP_TYPE* ep = dp + (dn*dnn);
  349. VECT_OP_TYPE* dbp = dp;
  350. for(; dp < ep; v+=n, dp+=dnn )
  351. *dp += *v;
  352. return dbp;
  353. }
  354. VECT_OP_TYPE* VECT_OP_FUNC(AddVVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, const VECT_OP_TYPE* sb1p )
  355. {
  356. const VECT_OP_TYPE* dep = dbp + dn;
  357. VECT_OP_TYPE* dp = dbp;
  358. while( dbp < dep )
  359. *dbp++ = *sb0p++ + *sb1p++;
  360. return dp;
  361. }
  362. VECT_OP_TYPE* VECT_OP_FUNC(MultVVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, const VECT_OP_TYPE* sb1p )
  363. {
  364. const VECT_OP_TYPE* dep = dbp + dn;
  365. VECT_OP_TYPE* dp = dbp;
  366. while( dbp < dep )
  367. *dbp++ = *sb0p++ * *sb1p++;
  368. return dp;
  369. }
  370. VECT_OP_TYPE* VECT_OP_FUNC(MultVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sbp )
  371. {
  372. const VECT_OP_TYPE* dep = dbp + dn;
  373. VECT_OP_TYPE* dp = dbp;
  374. while( dbp < dep )
  375. *dbp++ *= *sbp++;
  376. return dp;
  377. }
  378. VECT_OP_TYPE* VECT_OP_FUNC(MultVVNN)(VECT_OP_TYPE* dp, unsigned dn, unsigned dnn, const VECT_OP_TYPE* v, unsigned n )
  379. {
  380. const VECT_OP_TYPE* ep = dp + (dn*dnn);
  381. VECT_OP_TYPE* dbp = dp;
  382. for(; dp < ep; v+=n, dp+=dnn )
  383. *dp *= *v;
  384. return dbp;
  385. }
  386. VECT_OP_TYPE* VECT_OP_FUNC(MultVS)( VECT_OP_TYPE* dbp, unsigned dn, VECT_OP_TYPE s )
  387. {
  388. const VECT_OP_TYPE* dep = dbp + dn;
  389. VECT_OP_TYPE* dp = dbp;
  390. while( dbp < dep )
  391. *dbp++ *= s;
  392. return dp;
  393. }
  394. VECT_OP_TYPE* VECT_OP_FUNC(MultVVS)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sbp, VECT_OP_TYPE s )
  395. {
  396. const VECT_OP_TYPE* dep = dbp + dn;
  397. VECT_OP_TYPE* dp = dbp;
  398. while( dbp < dep )
  399. *dbp++ = *sbp++ * s;
  400. return dp;
  401. }
  402. VECT_OP_TYPE* VECT_OP_FUNC(MultVaVS)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sbp, VECT_OP_TYPE s )
  403. {
  404. const VECT_OP_TYPE* dep = dbp + dn;
  405. VECT_OP_TYPE* dp = dbp;
  406. while( dbp < dep )
  407. *dbp++ += *sbp++ * s;
  408. return dp;
  409. }
  410. VECT_OP_TYPE* VECT_OP_FUNC(MultSumVVS)(VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sbp, VECT_OP_TYPE s )
  411. {
  412. const VECT_OP_TYPE* dep = dbp + dn;
  413. VECT_OP_TYPE* dp = dbp;
  414. while( dbp < dep )
  415. *dbp++ += *sbp++ * s;
  416. return dp;
  417. }
  418. VECT_OP_TYPE* VECT_OP_FUNC(DivVVS)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, VECT_OP_TYPE s1 )
  419. {
  420. const VECT_OP_TYPE* dep = dbp + dn;
  421. VECT_OP_TYPE* dp = dbp;
  422. while( dbp < dep )
  423. *dbp++ = *sb0p++ / s1;
  424. return dp;
  425. }
  426. VECT_OP_TYPE* VECT_OP_FUNC(DivVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p )
  427. {
  428. const VECT_OP_TYPE* dep = dbp + dn;
  429. VECT_OP_TYPE* dp = dbp;
  430. while( dbp < dep )
  431. *dbp++ /= *sb0p++;
  432. return dp;
  433. }
  434. VECT_OP_TYPE* VECT_OP_FUNC(DivVVV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, const VECT_OP_TYPE* sb1p )
  435. {
  436. const VECT_OP_TYPE* dep = dbp + dn;
  437. VECT_OP_TYPE* dp = dbp;
  438. while( dbp < dep )
  439. *dbp++ = *sb0p++ / *sb1p++;
  440. return dp;
  441. }
  442. VECT_OP_TYPE* VECT_OP_FUNC(DivVVNN)(VECT_OP_TYPE* dp, unsigned dn, unsigned dnn, const VECT_OP_TYPE* v, unsigned n )
  443. {
  444. const VECT_OP_TYPE* ep = dp + (dn*dnn);
  445. VECT_OP_TYPE* dbp = dp;
  446. for(; dp < ep; v+=n, dp+=dnn )
  447. *dp /= *v;
  448. return dbp;
  449. }
  450. VECT_OP_TYPE* VECT_OP_FUNC(DivVS)( VECT_OP_TYPE* dbp, unsigned dn, VECT_OP_TYPE s )
  451. {
  452. const VECT_OP_TYPE* dep = dbp + dn;
  453. VECT_OP_TYPE* dp = dbp;
  454. while( dbp < dep )
  455. *dbp++ /= s;
  456. return dp;
  457. }
  458. VECT_OP_TYPE* VECT_OP_FUNC(DivVSV)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE s0, const VECT_OP_TYPE* sb1p )
  459. {
  460. const VECT_OP_TYPE* dep = dbp + dn;
  461. VECT_OP_TYPE* dp = dbp;
  462. while( dbp < dep )
  463. *dbp++ = s0 / *sb1p++;
  464. return dp;
  465. }
  466. VECT_OP_TYPE* VECT_OP_FUNC(DivVVZ)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p )
  467. {
  468. const VECT_OP_TYPE* dep = dbp + dn;
  469. VECT_OP_TYPE* dp = dbp;
  470. for(; dbp < dep; ++sb0p )
  471. if( *sb0p == 0 )
  472. *dbp++ = 0;
  473. else
  474. *dbp++ /= *sb0p;
  475. return dp;
  476. }
  477. VECT_OP_TYPE* VECT_OP_FUNC(DivVVVZ)( VECT_OP_TYPE* dbp, unsigned dn, const VECT_OP_TYPE* sb0p, const VECT_OP_TYPE* sb1p )
  478. {
  479. const VECT_OP_TYPE* dep = dbp + dn;
  480. VECT_OP_TYPE* dp = dbp;
  481. for(; dbp < dep; ++sb0p,++sb1p )
  482. if( *sb1p == 0 )
  483. *dbp++ = 0;
  484. else
  485. *dbp++ = *sb0p / *sb1p;
  486. return dp;
  487. }
  488. VECT_OP_TYPE* VECT_OP_FUNC(DivMS)( VECT_OP_TYPE* dp, unsigned drn, unsigned dcn, const VECT_OP_TYPE* sp )
  489. {
  490. unsigned i;
  491. for(i=0; i<dcn; ++i)
  492. VECT_OP_FUNC(DivVS)( dp + i*drn, drn, sp[i] );
  493. return dp;
  494. }
  495. VECT_OP_TYPE VECT_OP_FUNC(Sum)( const VECT_OP_TYPE* bp, unsigned n )
  496. {
  497. const VECT_OP_TYPE* ep = bp + n;
  498. VECT_OP_TYPE s = 0;
  499. while( bp < ep )
  500. s += *bp++;
  501. return s;
  502. }
  503. VECT_OP_TYPE VECT_OP_FUNC(SumN)( const VECT_OP_TYPE* bp, unsigned n, unsigned stride )
  504. {
  505. const VECT_OP_TYPE* ep = bp + (n*stride);
  506. VECT_OP_TYPE s = 0;
  507. for(; bp < ep; bp += stride )
  508. s += *bp;
  509. return s;
  510. }
  511. VECT_OP_TYPE* VECT_OP_FUNC(SumM)(const VECT_OP_TYPE* sp, unsigned srn, unsigned scn, VECT_OP_TYPE* dp )
  512. {
  513. unsigned i;
  514. for(i=0; i<scn; ++i)
  515. dp[i] = VECT_OP_FUNC(Sum)(sp + (i*srn), srn );
  516. return dp;
  517. }
  518. VECT_OP_TYPE* VECT_OP_FUNC(SumMN)(const VECT_OP_TYPE* sp, unsigned srn, unsigned scn, VECT_OP_TYPE* dp )
  519. {
  520. unsigned i;
  521. for(i=0; i<srn; ++i)
  522. dp[i] = VECT_OP_FUNC(SumN)(sp + i, scn, srn );
  523. return dp;
  524. }
  525. VECT_OP_TYPE* VECT_OP_FUNC(Abs)( VECT_OP_TYPE* dbp, unsigned dn )
  526. {
  527. unsigned i;
  528. for(i=0; i<dn; ++i)
  529. if( dbp[i]<0 )
  530. dbp[i] = -dbp[i];
  531. return dbp;
  532. }
  533. // mi is a target value - it holds the number of elements in ap[an] which must be be less than the median value.
  534. // If the initial array contains an even number of values then the median value is formed by averaging the two center values.
  535. // In this case *evenFlPtr is set and used to indicate that the center-upper value must be found during undwinding.
  536. VECT_OP_TYPE VECT_OP_FUNC(MedianSearch)( unsigned mi, const VECT_OP_TYPE* ap, unsigned an, bool* evenFlPtr )
  537. {
  538. VECT_OP_TYPE x = ap[0]; // pick a random value as a potential median value
  539. VECT_OP_TYPE a1[ an ]; // values below x
  540. VECT_OP_TYPE a3[ an ]; // values above x
  541. unsigned a1n = 0;
  542. unsigned a2n = 0; // values equal to x
  543. unsigned a3n = 0;
  544. const VECT_OP_TYPE* abp = ap;
  545. const VECT_OP_TYPE* aep = abp + an;
  546. for(; abp < aep; ++abp )
  547. {
  548. if( *abp < x )
  549. a1[a1n++] = *abp;
  550. else
  551. {
  552. if( *abp > x )
  553. a3[a3n++] = *abp;
  554. else
  555. ++a2n;
  556. }
  557. }
  558. //printf("%i : %i %i %i\n",mi,a1n,a2n,a3n);
  559. // there are more values below x (mi remains the target split point)
  560. if( a1n > mi )
  561. {
  562. x = VECT_OP_FUNC(MedianSearch)(mi,a1,a1n,evenFlPtr);
  563. }
  564. else
  565. {
  566. // the target was located
  567. if( a1n+a2n >= mi )
  568. {
  569. // if a1n alone matches mi then the max value in a1[] holds the median value otherwise x is the median
  570. if(a1n>=1 && a1n==mi)
  571. {
  572. VECT_OP_TYPE mv = VECT_OP_FUNC(Max)(a1,a1n,1);
  573. x = *evenFlPtr ? (mv+x)/2 : mv;
  574. *evenFlPtr = false;
  575. }
  576. // if the evenFl is set then the closest value above the median (x) must be located
  577. if( *evenFlPtr )
  578. {
  579. // if the next greater value is in a2[]
  580. if( a2n > 1 && (a1n+a2n) > mi )
  581. *evenFlPtr = false;
  582. else
  583. // if the next greater value is in a3[]
  584. if( a3n > 1 )
  585. {
  586. x = (x + VECT_OP_FUNC(Min)(a3,a3n,1))/2;
  587. *evenFlPtr = false;
  588. }
  589. }
  590. // no need for unwind processing - all the possibilities at this level have been exhausted
  591. return x;
  592. }
  593. else
  594. {
  595. // There are more values above x - the median must therefore be in a3[].
  596. // Reset mi cmcounting for the fact that we know that there are
  597. // a1n+a2n values below the lowest value in a3.
  598. x = VECT_OP_FUNC(MedianSearch)(mi - (a1n+a2n), a3, a3n, evenFlPtr );
  599. }
  600. }
  601. if( *evenFlPtr )
  602. {
  603. // find the first value greater than x
  604. while( ap < aep && *ap <= x )
  605. ++ap;
  606. if( ap < aep )
  607. {
  608. VECT_OP_TYPE v = *ap++;
  609. // find the nearest value greater than x
  610. for(; ap < aep; ++ap )
  611. if( *ap > x && ((*ap - x) < (v-x)))
  612. v = *ap;
  613. x = (v + x)/2;
  614. *evenFlPtr = false;
  615. }
  616. }
  617. return x;
  618. }
  619. VECT_OP_TYPE VECT_OP_FUNC(Median)( const VECT_OP_TYPE* bp, unsigned n )
  620. {
  621. bool evenFl = cmIsEvenU(n);
  622. unsigned medIdx = evenFl ? n/2 : (n+1)/2;
  623. return VECT_OP_FUNC(MedianSearch)( medIdx, bp, n, &evenFl );
  624. }
  625. unsigned VECT_OP_FUNC(MinIndex)( const VECT_OP_TYPE* bp, unsigned n, unsigned stride )
  626. {
  627. const VECT_OP_TYPE* ep = bp + (n*stride);
  628. if( bp >= ep )
  629. return cmInvalidIdx;
  630. const VECT_OP_TYPE* p = bp;
  631. const VECT_OP_TYPE* mp = bp;
  632. bp+=stride;
  633. for(; bp < ep; bp+=stride )
  634. if( *bp < *mp )
  635. mp = bp;
  636. return (mp - p)/stride;
  637. }
  638. unsigned VECT_OP_FUNC(MaxIndex)( const VECT_OP_TYPE* bp, unsigned n, unsigned stride )
  639. {
  640. const VECT_OP_TYPE* ep = bp + (n*stride);
  641. if( bp >= ep )
  642. return cmInvalidIdx;
  643. const VECT_OP_TYPE* p = bp;
  644. const VECT_OP_TYPE* mp = bp;
  645. bp+=stride;
  646. for(; bp < ep; bp+=stride )
  647. if( *bp > *mp )
  648. mp = bp;
  649. return (mp - p)/stride;
  650. }
  651. VECT_OP_TYPE VECT_OP_FUNC(Min)( const VECT_OP_TYPE* bp, unsigned n, unsigned stride )
  652. {
  653. unsigned i;
  654. if((i = VECT_OP_FUNC(MinIndex)(bp,n,stride)) == cmInvalidIdx )
  655. {
  656. assert(0);
  657. return 0;
  658. }
  659. return bp[i*stride];
  660. }
  661. VECT_OP_TYPE VECT_OP_FUNC(Max)( const VECT_OP_TYPE* bp, unsigned n, unsigned stride )
  662. {
  663. unsigned i;
  664. if((i = VECT_OP_FUNC(MaxIndex)(bp,n,stride)) == cmInvalidIdx )
  665. {
  666. assert(0);
  667. return 0;
  668. }
  669. return bp[i*stride];
  670. }
  671. VECT_OP_TYPE* VECT_OP_FUNC(MinVV)( VECT_OP_TYPE* dp, unsigned dn, const VECT_OP_TYPE* sp )
  672. {
  673. unsigned i;
  674. for(i=0; i<dn; ++i)
  675. if( sp[i] < dp[i] )
  676. dp[i] = sp[i];
  677. return dp;
  678. }
  679. VECT_OP_TYPE* VECT_OP_FUNC(MaxVV)( VECT_OP_TYPE* dp, unsigned dn, const VECT_OP_TYPE* sp )
  680. {
  681. unsigned i;
  682. for(i=0; i<dn; ++i)
  683. if( sp[i] > dp[i] )
  684. dp[i] = sp[i];
  685. return dp;
  686. }
  687. unsigned* VECT_OP_FUNC(MinIndexM)( unsigned* dp, const VECT_OP_TYPE* sp, unsigned srn, unsigned scn )
  688. {
  689. unsigned i = 0;
  690. for(i=0; i<scn; ++i)
  691. dp[i] = VECT_OP_FUNC(MinIndex)(sp + (i*srn), srn, 1 );
  692. return dp;
  693. }
  694. unsigned* VECT_OP_FUNC(MaxIndexM)( unsigned* dp, const VECT_OP_TYPE* sp, unsigned srn, unsigned scn )
  695. {
  696. unsigned i = 0;
  697. for(i=0; i<scn; ++i)
  698. dp[i] = VECT_OP_FUNC(MaxIndex)(sp + (i*srn), srn, 1 );
  699. return dp;
  700. }
  701. bool VECT_OP_FUNC(IsEqual)( const VECT_OP_TYPE* s0p, const VECT_OP_TYPE* s1p, unsigned sn )
  702. {
  703. const VECT_OP_TYPE* ep = s0p + sn;
  704. for(; s0p < ep; ++s0p,++s1p )
  705. if( *s0p != *s1p )
  706. return false;
  707. return true;
  708. }
  709. bool VECT_OP_FUNC(IsClose)( const VECT_OP_TYPE* s0p, const VECT_OP_TYPE* s1p, unsigned sn, double pct )
  710. {
  711. const VECT_OP_TYPE* ep = s0p + sn;
  712. for(; s0p < ep; ++s0p,++s1p )
  713. {
  714. double d = *s1p - *s0p;
  715. double s = cmMin(*s1p,*s0p);
  716. // take abs value of d and s
  717. if( d < 0 )
  718. d *= -1;
  719. if( s < 0 )
  720. s *= -1;
  721. if( d*100.0/s > pct )
  722. return false;
  723. }
  724. return true;
  725. }
  726. VECT_OP_TYPE VECT_OP_FUNC(Mode)( const VECT_OP_TYPE* sp, unsigned sn )
  727. {
  728. unsigned n[sn];
  729. VECT_OP_TYPE v[sn];
  730. unsigned i,j,k = 0;
  731. unsigned n0 = 0; // idx of most freq occurring ele
  732. unsigned n1 = -1; // idx of 2nd most freq occurring ele
  733. for(i=0; i<sn; ++i)
  734. {
  735. // find sp[i] in v[]
  736. for(j=0; j<k; ++j)
  737. if( sp[i] == v[j] )
  738. {
  739. ++n[j];
  740. break;
  741. }
  742. // sp[i] was not found in v[]
  743. if( k == j )
  744. {
  745. v[j] = sp[i];
  746. n[j] = 1;
  747. ++k;
  748. }
  749. // n[j] holds frq of sp[i]
  750. // do nothing if j is already most freq
  751. if( j != n0 )
  752. {
  753. // if j is new most freq
  754. if( n[j] > n[n0] )
  755. {
  756. n1 = n0;
  757. n0 = j;
  758. }
  759. else
  760. // if j is 2nd most freq
  761. if( (n1==-1) || (n[j] > n[n1]) )
  762. n1 = j;
  763. }
  764. // if diff between two most freq is greater than remaining ele's
  765. if( (n1!=-1) && (n[n0]-n[n1]) >= (sn-i) )
  766. break;
  767. }
  768. // if there are no ele's with same count
  769. if( n[n0] > n[n1] )
  770. return v[n0];
  771. // break tie between ele's with same count be returning min value
  772. // (this is the same as Matlab tie break criteria)
  773. j = 0;
  774. for(i=1; i<k; ++i)
  775. if( (n[i] > n[j]) || (n[i] == n[j] && v[i] < v[j]) )
  776. j=i;
  777. return v[j];
  778. }
  779. unsigned VECT_OP_FUNC(Find)( const VECT_OP_TYPE* sp, unsigned sn, VECT_OP_TYPE key )
  780. {
  781. const VECT_OP_TYPE* sbp = sp;
  782. const VECT_OP_TYPE* ep = sp + sn;
  783. while( sp<ep )
  784. if( *sp++ == key )
  785. break;
  786. if( sp==ep )
  787. return cmInvalidIdx;
  788. return (sp-1) - sbp;
  789. }
  790. unsigned VECT_OP_FUNC(Count)( const VECT_OP_TYPE* sp, unsigned sn, VECT_OP_TYPE key )
  791. {
  792. unsigned cnt = 0;
  793. const VECT_OP_TYPE* ep = sp + sn;
  794. while( sp<ep )
  795. if( *sp++ == key )
  796. ++cnt;
  797. return cnt;
  798. }
  799. VECT_OP_TYPE* VECT_OP_FUNC(ReplaceLte)( VECT_OP_TYPE* dp, unsigned dn, const VECT_OP_TYPE* sp, VECT_OP_TYPE lteKeyVal, VECT_OP_TYPE replaceVal )
  800. {
  801. VECT_OP_TYPE* rp = dp;
  802. const VECT_OP_TYPE* ep = dp + dn;
  803. for(; dp < ep; ++sp )
  804. *dp++ = *sp <= lteKeyVal ? replaceVal : *sp;
  805. return rp;
  806. }
  807. VECT_OP_TYPE* VECT_OP_FUNC(Diag)( VECT_OP_TYPE* dbp, unsigned n, const VECT_OP_TYPE* sbp )
  808. {
  809. unsigned i,j;
  810. for(i=0,j=0; i<n && j<n; ++i,++j)
  811. dbp[ (i*n) + j ] = sbp[i];
  812. return dbp;
  813. }
  814. VECT_OP_TYPE* VECT_OP_FUNC(DiagZ)(VECT_OP_TYPE* dbp, unsigned n, const VECT_OP_TYPE* sbp )
  815. {
  816. VECT_OP_FUNC(Fill)(dbp,n*n,0);
  817. return VECT_OP_FUNC(Diag)(dbp,n,sbp);
  818. }
  819. VECT_OP_TYPE* VECT_OP_FUNC(Identity)( VECT_OP_TYPE* dbp, unsigned rn, unsigned cn )
  820. {
  821. unsigned i,j;
  822. for(i=0,j=0; i<cn && j<rn; ++i,++j)
  823. dbp[ (i*rn) + j ] = 1;
  824. return dbp;
  825. }
  826. VECT_OP_TYPE* VECT_OP_FUNC(IdentityZ)( VECT_OP_TYPE* dbp, unsigned rn, unsigned cn )
  827. {
  828. VECT_OP_FUNC(Fill)(dbp,rn*cn,0);
  829. return VECT_OP_FUNC(Identity)(dbp,rn,cn);
  830. }
  831. VECT_OP_TYPE* VECT_OP_FUNC(Transpose)( VECT_OP_TYPE* dbp, const VECT_OP_TYPE* sp, unsigned srn, unsigned scn )
  832. {
  833. VECT_OP_TYPE* dp = dbp;
  834. const VECT_OP_TYPE* dep = dbp + (srn*scn);
  835. while( dbp < dep )
  836. {
  837. const VECT_OP_TYPE* sbp = sp++;
  838. const VECT_OP_TYPE* sep = sbp + (srn*scn);
  839. for(; sbp < sep; sbp+=srn )
  840. *dbp++ = *sbp;
  841. }
  842. return dp;
  843. }
  844. VECT_OP_TYPE VECT_OP_FUNC(Seq)( VECT_OP_TYPE* dbp, unsigned dn, VECT_OP_TYPE beg, VECT_OP_TYPE incr )
  845. {
  846. const VECT_OP_TYPE* dep = dbp + dn;
  847. unsigned i = 0;
  848. for(; dbp<dep; ++i)
  849. *dbp++ = beg + (incr*i);
  850. return beg + (incr*i);
  851. }
  852. void VECT_OP_FUNC(FnThresh)( const VECT_OP_TYPE* xV, unsigned xN, unsigned wndN, VECT_OP_TYPE* yV, unsigned yStride, VECT_OP_TYPE (*fnPtr)(const VECT_OP_TYPE*, unsigned) )
  853. {
  854. int i0 = cmIsOddU(wndN) ? (wndN-1)/2 : wndN/2;
  855. int i1 = cmIsOddU(wndN) ? (wndN-1)/2 : wndN/2 - 1;
  856. int i,j;
  857. i0 = -i0;
  858. if( fnPtr == NULL )
  859. fnPtr = &(VECT_OP_FUNC(Median));
  860. for(i=0; i<xN; ++i,++i0,++i1)
  861. {
  862. j = (i*yStride);
  863. if( i0 < 0 )
  864. if( i1 >= xN )
  865. yV[j] = (*fnPtr)(xV,xN);
  866. else
  867. yV[j] = (*fnPtr)(xV,i1+1);
  868. else if( i1 >= xN )
  869. yV[j] = (*fnPtr)(xV+i0,xN-i0);
  870. else
  871. yV[j] = (*fnPtr)(xV+i0,wndN);
  872. }
  873. }
  874. void VECT_OP_FUNC(MedianFilt)( const VECT_OP_TYPE* xV, unsigned xN, unsigned wndN, VECT_OP_TYPE* yV, unsigned yStride )
  875. {
  876. int i0 = cmIsOddU(wndN) ? (wndN-1)/2 : wndN/2;
  877. int i1 = cmIsOddU(wndN) ? (wndN-1)/2 : wndN/2 - 1;
  878. int i,j;
  879. VECT_OP_TYPE tV[ wndN ];
  880. i0 = -i0;
  881. VECT_OP_FUNC(Fill)(tV,wndN,0);
  882. for(i=0; i<xN; ++i,++i0,++i1)
  883. {
  884. j = (i*yStride);
  885. // note that the position of the zero padding in tV[]
  886. // does not matter because the median calcluation does
  887. // not make any assumptions about the order of the argument
  888. // vector.
  889. if( i0 < 0 )
  890. {
  891. VECT_OP_FUNC(Copy)(tV,wndN+i0,xV);
  892. VECT_OP_FUNC(Fill)(tV+wndN+i0,labs(i0),0);
  893. //VECT_OP_FUNC(Print)(NULL,1,wndN,tV,-1,-1);
  894. yV[j] = VECT_OP_FUNC(Median)(tV,wndN);
  895. continue;
  896. }
  897. if( i1 >= xN )
  898. {
  899. VECT_OP_FUNC(Copy)(tV,wndN-(i1-xN+1),xV+i0);
  900. VECT_OP_FUNC(Fill)(tV+wndN-(i1-xN+1),i1-xN+1,0);
  901. //VECT_OP_FUNC(Print)(NULL,1,wndN,tV,-1,-1);
  902. yV[j] = VECT_OP_FUNC(Median)(tV,wndN);
  903. continue;
  904. }
  905. //VECT_OP_FUNC(Print)(NULL,1,wndN,xV+i0,-1,-1);
  906. yV[j] = VECT_OP_FUNC(Median)(xV+i0,wndN);
  907. }
  908. }
  909. unsigned* VECT_OP_FUNC(LevEditDistAllocMtx)(unsigned maxN)
  910. {
  911. maxN += 1;
  912. unsigned* m = cmMemAllocZ(unsigned,maxN*maxN);
  913. unsigned* p = m;
  914. unsigned i;
  915. // initialize the comparison matrix with the default costs in the
  916. // first row and column
  917. // (Note that this matrix is not oriented in column major order like most 'cm' matrices.)
  918. for(i=0; i<maxN; ++i)
  919. {
  920. p[i] = i; // 0th row
  921. p[ i * maxN ] = i; // 0th col
  922. }
  923. return m;
  924. }
  925. double VECT_OP_FUNC(LevEditDist)(unsigned mtxMaxN, unsigned* m, const VECT_OP_TYPE* s0, int n0, const VECT_OP_TYPE* s1, int n1, unsigned maxN )
  926. {
  927. mtxMaxN += 1;
  928. assert( n0 < mtxMaxN && n1 < mtxMaxN );
  929. int v = 0;
  930. unsigned i;
  931. // Note that m[maxN,maxN] is not oriented in column major order like most 'cm' matrices.
  932. for(i=1; i<n0+1; ++i)
  933. {
  934. unsigned ii = i * mtxMaxN; // current row
  935. unsigned i_1 = ii - mtxMaxN; // previous row
  936. unsigned j;
  937. for( j=1; j<n1+1; ++j)
  938. {
  939. int cost = s0[i-1] == s1[j-1] ? 0 : 1;
  940. //m[i][j] = min( m[i-1][j] + 1, min( m[i][j-1] + 1, m[i-1][j-1] + cost ) );
  941. m[ ii + j ] = v = cmMin( m[ i_1 + j] + 1, cmMin( m[ ii + j - 1] + 1, m[ i_1 + j - 1 ] + cost ) );
  942. }
  943. }
  944. return (double) v / maxN;
  945. }
  946. double VECT_OP_FUNC(LevEditDistWithCostThresh)( int mtxMaxN, unsigned* m, const VECT_OP_TYPE* s0, int n0, const VECT_OP_TYPE* s1, int n1, double maxCost, unsigned maxN )
  947. {
  948. mtxMaxN += 1;
  949. int v = 0;
  950. maxCost = cmMin(1.0,cmMax(0.0,maxCost));
  951. int iMaxCost = ceil( maxCost * maxN );
  952. assert( iMaxCost > 0 && maxCost > 0 );
  953. // If the two strings are different lengths and the min possible distance is
  954. // greater than the threshold then return the threshold as the cost.
  955. // (Note: For strings of different length the min possible distance is the
  956. // difference in length between the two strings).
  957. if( abs(n0-n1) > iMaxCost )
  958. return maxCost;
  959. int i;
  960. // for each row in the matrix ...
  961. for(i=1; i<n0+1; ++i)
  962. {
  963. int ii = i * mtxMaxN; // current row
  964. int i_1 = ii - mtxMaxN; // previous row
  965. // Limit the row to (2*iMaxCost)+1 diagnal strip.
  966. // This strip is based on the idea that the best case can be precomputed for
  967. // all matrix elements in advance - where the best case for position i,j is:
  968. // abs(i-j). This can be justified based on the idea that the least possible
  969. // distance between two strings of length i and j is abs(i-1). The minimum least
  970. // possible distance is therefore found on the matrix diagnal and grows as the
  971. // distance from the diagnal increases.
  972. int ji = cmMax( 1, i - iMaxCost );
  973. int jn = cmMin(iMaxCost + i, n1) + 1;
  974. int j;
  975. // fill in (max cost + 1) as the value in the column before the starting column
  976. // (it will be referred to during the first computation in this row)
  977. if( ji >= 2 )
  978. m[ ii + (ji-1) ] = iMaxCost + 1;
  979. // for each column in the diagnal stripe - beginning with the leftmost column.
  980. for( j=ji; j<jn; ++j)
  981. {
  982. int cost = s0[i-1] == s1[j-1] ? 0 : 1;
  983. m[ ii + j ] = v = cmMin( m[ i_1 + j] + 1, cmMin( m[ ii + j - 1] + 1, m[ i_1 + j - 1 ] + cost ) );
  984. }
  985. // fill in (max cost + 1) in the column following the last column
  986. // (it will be referred to during computation of the following row)
  987. if( j < n1+1 )
  988. m[ii + j] = iMaxCost + 1;
  989. }
  990. assert( v >= 0 );
  991. return cmMin( maxCost , (double) v / maxN);
  992. }
  993. #endif