Home | History | Annotate | Download | only in dicttrieperf
      1 /*
      2  **********************************************************************
      3  *   Copyright (C) 2010-2011, International Business Machines
      4  *   Corporation and others.  All Rights Reserved.
      5  **********************************************************************
      6  *  file name:  dicttrieperf.cpp
      7  *  encoding:   US-ASCII
      8  *  tab size:   8 (not used)
      9  *  indentation:4
     10  *
     11  *  created on: 2010dec09
     12  *  created by: Markus W. Scherer
     13  *
     14  *  Performance test program for dictionary-type tries.
     15  *
     16  * Usage from within <ICU build tree>/test/perf/dicttrieperf/ :
     17  * (Linux)
     18  *  make
     19  *  export LD_LIBRARY_PATH=../../../lib:../../../stubdata:../../../tools/ctestfw
     20  *  ./dicttrieperf --sourcedir <ICU build tree>/data/out/tmp --passes 3 --iterations 1000
     21  * or
     22  *  ./dicttrieperf -f <ICU source tree>/source/data/brkitr/thaidict.txt --passes 3 --iterations 250
     23  */
     24 
     25 #include <stdio.h>
     26 #include <stdlib.h>
     27 #include "unicode/bytestrie.h"
     28 #include "unicode/bytestriebuilder.h"
     29 #include "unicode/localpointer.h"
     30 #include "unicode/ucharstrie.h"
     31 #include "unicode/ucharstriebuilder.h"
     32 #include "unicode/uperf.h"
     33 #include "unicode/utext.h"
     34 #include "charstr.h"
     35 #include "package.h"
     36 #include "toolutil.h"
     37 #include "triedict.h"
     38 #include "ucbuf.h"  // struct ULine
     39 #include "uoptions.h"
     40 #include "uvectr32.h"
     41 
     42 #define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
     43 
     44 // Test object.
     45 class DictionaryTriePerfTest : public UPerfTest {
     46 public:
     47     DictionaryTriePerfTest(int32_t argc, const char *argv[], UErrorCode &status)
     48             : UPerfTest(argc, argv, NULL, 0, "", status), numTextLines(0) {
     49         if(hasFile()) {
     50             getLines(status);
     51             for(int32_t i=0; i<numLines; ++i) {
     52                 // Skip comment lines (start with a character below 'A').
     53                 if(lines[i].name[0]>=0x41) {
     54                     ++numTextLines;
     55                     // Remove trailing CR LF.
     56                     int32_t len=lines[i].len;
     57                     UChar c;
     58                     while(len>0 && ((c=lines[i].name[len-1])==0xa || c==0xd)) {
     59                         --len;
     60                     }
     61                     lines[i].len=len;
     62                 }
     63             }
     64         }
     65     }
     66 
     67     virtual UPerfFunction *runIndexedTest(int32_t index, UBool exec, const char *&name, char *par=NULL);
     68 
     69     const char *getSourceDir() const { return sourceDir; }
     70 
     71     UBool hasFile() const { return ucharBuf!=NULL; }
     72     const ULine *getCachedLines() const { return lines; }
     73     int32_t getNumLines() const { return numLines; }
     74     int32_t numTextLines;  // excluding comment lines
     75 };
     76 
     77 // Performance test function object.
     78 // Loads icudt46l.dat (or whatever its current versioned filename)
     79 // from the -s or --sourcedir path.
     80 class PackageLookup : public UPerfFunction {
     81 protected:
     82     PackageLookup(const DictionaryTriePerfTest &perf) {
     83         IcuToolErrorCode errorCode("PackageLookup()");
     84         CharString filename(perf.getSourceDir(), errorCode);
     85         int32_t filenameLength=filename.length();
     86         if(filenameLength>0 && filename[filenameLength-1]!=U_FILE_SEP_CHAR &&
     87                                filename[filenameLength-1]!=U_FILE_ALT_SEP_CHAR) {
     88             filename.append(U_FILE_SEP_CHAR, errorCode);
     89         }
     90         filename.append(U_ICUDATA_NAME, errorCode);
     91         filename.append(".dat", errorCode);
     92         pkg.readPackage(filename.data());
     93     }
     94 
     95 public:
     96     virtual ~PackageLookup() {}
     97 
     98     // virtual void call(UErrorCode* pErrorCode) { ... }
     99 
    100     virtual long getOperationsPerIteration() {
    101         return pkg.getItemCount();
    102     }
    103 
    104     // virtual long getEventsPerIteration();
    105 
    106 protected:
    107     Package pkg;
    108 };
    109 
    110 struct TOCEntry {
    111     int32_t nameOffset, dataOffset;
    112 };
    113 
    114 // Similar to ICU 4.6 offsetTOCLookupFn() (in ucmndata.c).
    115 static int32_t simpleBinarySearch(const char *s, const char *names, const TOCEntry *toc, int32_t count) {
    116     int32_t start=0;
    117     int32_t limit=count;
    118     int32_t lastNumber=limit;
    119     for(;;) {
    120         int32_t number=(start+limit)/2;
    121         if(lastNumber==number) {  // have we moved?
    122             return -1;  // not found
    123         }
    124         lastNumber=number;
    125         int32_t cmp=strcmp(s, names+toc[number].nameOffset);
    126         if(cmp<0) {
    127             limit=number;
    128         } else if(cmp>0) {
    129             start=number;
    130         } else {  // found s
    131             return number;
    132         }
    133     }
    134 }
    135 
    136 class BinarySearchPackageLookup : public PackageLookup {
    137 public:
    138     BinarySearchPackageLookup(const DictionaryTriePerfTest &perf)
    139             : PackageLookup(perf) {
    140         IcuToolErrorCode errorCode("BinarySearchPackageLookup()");
    141         int32_t count=pkg.getItemCount();
    142         toc=new TOCEntry[count];
    143         for(int32_t i=0; i<count; ++i) {
    144             toc[i].nameOffset=itemNames.length();
    145             toc[i].dataOffset=i;  // arbitrary value, see toc comment below
    146             // The Package class removes the "icudt46l/" prefix.
    147             // We restore that here for a fair performance test.
    148             const char *name=pkg.getItem(i)->name;
    149             itemNames.append("icudt46l/", errorCode);
    150             itemNames.append(name, strlen(name)+1, errorCode);
    151         }
    152         printf("size of item names: %6ld\n", (long)itemNames.length());
    153         printf("size of TOC:        %6ld\n", (long)(count*8));
    154         printf("total index size:   %6ld\n", (long)(itemNames.length()+count*8));
    155     }
    156     virtual ~BinarySearchPackageLookup() {
    157         delete[] toc;
    158     }
    159 
    160     virtual void call(UErrorCode * /*pErrorCode*/) {
    161         int32_t count=pkg.getItemCount();
    162         const char *itemNameChars=itemNames.data();
    163         const char *name=itemNameChars;
    164         for(int32_t i=0; i<count; ++i) {
    165             if(simpleBinarySearch(name, itemNameChars, toc, count)<0) {
    166                 fprintf(stderr, "item not found: %s\n", name);
    167             }
    168             name=strchr(name, 0)+1;
    169         }
    170     }
    171 
    172 protected:
    173     CharString itemNames;
    174     // toc imitates a .dat file's array of UDataOffsetTOCEntry
    175     // with nameOffset and dataOffset.
    176     // We don't need the dataOffsets, but we want to imitate the real
    177     // memory density, to measure equivalent CPU cache usage.
    178     TOCEntry *toc;
    179 };
    180 
    181 #ifndef MIN
    182 #define MIN(a,b) (((a)<(b)) ? (a) : (b))
    183 #endif
    184 
    185 // Compare strings where we know the shared prefix length,
    186 // and advance the prefix length as we find that the strings share even more characters.
    187 static int32_t strcmpAfterPrefix(const char *s1, const char *s2, int32_t &prefixLength) {
    188     int32_t pl=prefixLength;
    189     s1+=pl;
    190     s2+=pl;
    191     int32_t cmp=0;
    192     for(;;) {
    193         int32_t c1=(uint8_t)*s1++;
    194         int32_t c2=(uint8_t)*s2++;
    195         cmp=c1-c2;
    196         if(cmp!=0 || c1==0) {  // different or done
    197             break;
    198         }
    199         ++pl;  // increment shared same-prefix length
    200     }
    201     prefixLength=pl;
    202     return cmp;
    203 }
    204 
    205 static int32_t prefixBinarySearch(const char *s, const char *names, const TOCEntry *toc, int32_t count) {
    206     if(count==0) {
    207         return -1;
    208     }
    209     int32_t start=0;
    210     int32_t limit=count;
    211     // Remember the shared prefix between s, start and limit,
    212     // and don't compare that shared prefix again.
    213     // The shared prefix should get longer as we narrow the [start, limit[ range.
    214     int32_t startPrefixLength=0;
    215     int32_t limitPrefixLength=0;
    216     // Prime the prefix lengths so that we don't keep prefixLength at 0 until
    217     // both the start and limit indexes have moved.
    218     // At the same time, we find if s is one of the start and (limit-1) names,
    219     // and if not, exclude them from the actual binary search.
    220     if(0==strcmpAfterPrefix(s, names+toc[0].nameOffset, startPrefixLength)) {
    221         return 0;
    222     }
    223     ++start;
    224     --limit;
    225     if(0==strcmpAfterPrefix(s, names+toc[limit].nameOffset, limitPrefixLength)) {
    226         return limit;
    227     }
    228     while(start<limit) {
    229         int32_t i=(start+limit)/2;
    230         int32_t prefixLength=MIN(startPrefixLength, limitPrefixLength);
    231         int32_t cmp=strcmpAfterPrefix(s, names+toc[i].nameOffset, prefixLength);
    232         if(cmp<0) {
    233             limit=i;
    234             limitPrefixLength=prefixLength;
    235         } else if(cmp==0) {
    236             return i;
    237         } else {
    238             start=i+1;
    239             startPrefixLength=prefixLength;
    240         }
    241     }
    242     return -1;
    243 }
    244 
    245 class PrefixBinarySearchPackageLookup : public BinarySearchPackageLookup {
    246 public:
    247     PrefixBinarySearchPackageLookup(const DictionaryTriePerfTest &perf)
    248             : BinarySearchPackageLookup(perf) {}
    249 
    250     virtual void call(UErrorCode * /*pErrorCode*/) {
    251         int32_t count=pkg.getItemCount();
    252         const char *itemNameChars=itemNames.data();
    253         const char *name=itemNameChars;
    254         for(int32_t i=0; i<count; ++i) {
    255             if(prefixBinarySearch(name, itemNameChars, toc, count)<0) {
    256                 fprintf(stderr, "item not found: %s\n", name);
    257             }
    258             name=strchr(name, 0)+1;
    259         }
    260     }
    261 };
    262 
    263 static int32_t bytesTrieLookup(const char *s, const char *nameTrieBytes) {
    264     BytesTrie trie(nameTrieBytes);
    265     if(USTRINGTRIE_HAS_VALUE(trie.next(s, -1))) {
    266         return trie.getValue();
    267     } else {
    268         return -1;
    269     }
    270 }
    271 
    272 class BytesTriePackageLookup : public PackageLookup {
    273 public:
    274     BytesTriePackageLookup(const DictionaryTriePerfTest &perf)
    275             : PackageLookup(perf) {
    276         IcuToolErrorCode errorCode("BinarySearchPackageLookup()");
    277         builder=new BytesTrieBuilder(errorCode);
    278         int32_t count=pkg.getItemCount();
    279         for(int32_t i=0; i<count; ++i) {
    280             // The Package class removes the "icudt46l/" prefix.
    281             // We restore that here for a fair performance test.
    282             // We store all full names so that we do not have to reconstruct them
    283             // in the call() function.
    284             const char *name=pkg.getItem(i)->name;
    285             int32_t offset=itemNames.length();
    286             itemNames.append("icudt46l/", errorCode);
    287             itemNames.append(name, -1, errorCode);
    288             // As value, set the data item index.
    289             // In a real implementation, we would use that to get the
    290             // start and limit offset of the data item.
    291             StringPiece fullName(itemNames.toStringPiece());
    292             fullName.remove_prefix(offset);
    293             builder->add(fullName, i, errorCode);
    294             // NUL-terminate the name for call() to find the next one.
    295             itemNames.append(0, errorCode);
    296         }
    297         int32_t length=builder->buildStringPiece(USTRINGTRIE_BUILD_SMALL, errorCode).length();
    298         printf("size of BytesTrie:   %6ld\n", (long)length);
    299         // count+1: +1 for the last-item limit offset which we should have always had
    300         printf("size of dataOffsets:%6ld\n", (long)((count+1)*4));
    301         printf("total index size:   %6ld\n", (long)(length+(count+1)*4));
    302     }
    303     virtual ~BytesTriePackageLookup() {
    304         delete builder;
    305     }
    306 
    307     virtual void call(UErrorCode *pErrorCode) {
    308         int32_t count=pkg.getItemCount();
    309         const char *nameTrieBytes=builder->buildStringPiece(USTRINGTRIE_BUILD_SMALL, *pErrorCode).data();
    310         const char *name=itemNames.data();
    311         for(int32_t i=0; i<count; ++i) {
    312             if(bytesTrieLookup(name, nameTrieBytes)<0) {
    313                 fprintf(stderr, "item not found: %s\n", name);
    314             }
    315             name=strchr(name, 0)+1;
    316         }
    317     }
    318 
    319 protected:
    320     BytesTrieBuilder *builder;
    321     CharString itemNames;
    322 };
    323 
    324 // Performance test function object.
    325 // Each subclass loads a dictionary text file
    326 // from the -s or --sourcedir path plus -f or --file-name.
    327 // For example, <ICU source dir>/source/data/brkitr/thaidict.txt.
    328 class DictLookup : public UPerfFunction {
    329 public:
    330     DictLookup(const DictionaryTriePerfTest &perfTest) : perf(perfTest) {}
    331 
    332     virtual long getOperationsPerIteration() {
    333         return perf.numTextLines;
    334     }
    335 
    336 protected:
    337     const DictionaryTriePerfTest &perf;
    338 };
    339 
    340 class CompactTrieDictLookup : public DictLookup {
    341 public:
    342     CompactTrieDictLookup(const DictionaryTriePerfTest &perfTest)
    343             : DictLookup(perfTest), ctd(NULL) {
    344         IcuToolErrorCode errorCode("UCharsTrieDictLookup()");
    345         // U+0E1C is the median code unit, from
    346         // the UCharsTrie root node (split-branch node) for thaidict.txt.
    347         MutableTrieDictionary builder(0xe1c, errorCode);
    348         const ULine *lines=perf.getCachedLines();
    349         int32_t numLines=perf.getNumLines();
    350         for(int32_t i=0; i<numLines; ++i) {
    351             // Skip comment lines (start with a character below 'A').
    352             if(lines[i].name[0]<0x41) {
    353                 continue;
    354             }
    355             builder.addWord(lines[i].name, lines[i].len, errorCode);
    356         }
    357         ctd=new CompactTrieDictionary(builder, errorCode);
    358         int32_t length=(int32_t)ctd->dataSize();
    359         printf("size of CompactTrieDict:    %6ld bytes\n", (long)length);
    360     }
    361 
    362     virtual ~CompactTrieDictLookup() {
    363         delete ctd;
    364     }
    365 
    366     virtual void call(UErrorCode *pErrorCode) {
    367         UText text=UTEXT_INITIALIZER;
    368         int32_t lengths[20];
    369         const ULine *lines=perf.getCachedLines();
    370         int32_t numLines=perf.getNumLines();
    371         for(int32_t i=0; i<numLines; ++i) {
    372             // Skip comment lines (start with a character below 'A').
    373             if(lines[i].name[0]<0x41) {
    374                 continue;
    375             }
    376             utext_openUChars(&text, lines[i].name, lines[i].len, pErrorCode);
    377             int32_t count;
    378             ctd->matches(&text, lines[i].len,
    379                          lengths, count, LENGTHOF(lengths));
    380             if(count==0 || lengths[count-1]!=lines[i].len) {
    381                 fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    382             }
    383         }
    384     }
    385 
    386 protected:
    387     CompactTrieDictionary *ctd;
    388 };
    389 
    390 // Closely imitate CompactTrieDictionary::matches().
    391 // Note: CompactTrieDictionary::matches() is part of its trie implementation,
    392 // and while it loops over the text, it knows the current state.
    393 // By contrast, this implementation uses UCharsTrie API functions that have to
    394 // check the trie state each time and load/store state in the object.
    395 // (Whether it hasNext() and whether it is in the middle of a linear-match node.)
    396 static int32_t
    397 ucharsTrieMatches(UCharsTrie &trie,
    398                   UText *text, int32_t textLimit,
    399                   int32_t *lengths, int &count, int limit ) {
    400     UChar32 c=utext_next32(text);
    401     // Notes:
    402     // a) CompactTrieDictionary::matches() does not check for U_SENTINEL.
    403     // b) It also ignores non-BMP code points by casting to UChar!
    404     if(c<0) {
    405         return 0;
    406     }
    407     // Should be firstForCodePoint() but CompactTrieDictionary
    408     // handles only code units.
    409     UStringTrieResult result=trie.first(c);
    410     int32_t numChars=1;
    411     count=0;
    412     for(;;) {
    413         if(USTRINGTRIE_HAS_VALUE(result)) {
    414             if(count<limit) {
    415                 // lengths[count++]=(int32_t)utext_getNativeIndex(text);
    416                 lengths[count++]=numChars;  // CompactTrieDictionary just counts chars too.
    417             }
    418             if(result==USTRINGTRIE_FINAL_VALUE) {
    419                 break;
    420             }
    421         } else if(result==USTRINGTRIE_NO_MATCH) {
    422             break;
    423         }
    424         if(numChars>=textLimit) {
    425             // Note: Why do we have both a text limit and a UText that knows its length?
    426             break;
    427         }
    428         UChar32 c=utext_next32(text);
    429         // Notes:
    430         // a) CompactTrieDictionary::matches() does not check for U_SENTINEL.
    431         // b) It also ignores non-BMP code points by casting to UChar!
    432         if(c<0) {
    433             break;
    434         }
    435         ++numChars;
    436         // Should be nextForCodePoint() but CompactTrieDictionary
    437         // handles only code units.
    438         result=trie.next(c);
    439     }
    440 #if 0
    441     // Note: CompactTrieDictionary::matches() comments say that it leaves the UText
    442     // after the longest prefix match and returns the number of characters
    443     // that were matched.
    444     if(index!=lastMatch) {
    445         utext_setNativeIndex(text, lastMatch);
    446     }
    447     return lastMatch-start;
    448     // However, it does not do either of these, so I am not trying to
    449     // imitate it (or its docs) 100%.
    450 #endif
    451     return numChars;
    452 }
    453 
    454 class UCharsTrieDictLookup : public DictLookup {
    455 public:
    456     UCharsTrieDictLookup(const DictionaryTriePerfTest &perfTest)
    457             : DictLookup(perfTest), trie(NULL) {
    458         IcuToolErrorCode errorCode("UCharsTrieDictLookup()");
    459         builder=new UCharsTrieBuilder(errorCode);
    460         const ULine *lines=perf.getCachedLines();
    461         int32_t numLines=perf.getNumLines();
    462         for(int32_t i=0; i<numLines; ++i) {
    463             // Skip comment lines (start with a character below 'A').
    464             if(lines[i].name[0]<0x41) {
    465                 continue;
    466             }
    467             builder->add(UnicodeString(FALSE, lines[i].name, lines[i].len), 0, errorCode);
    468         }
    469         UnicodeString trieUChars;
    470         int32_t length=builder->buildUnicodeString(USTRINGTRIE_BUILD_SMALL, trieUChars, errorCode).length();
    471         printf("size of UCharsTrie:          %6ld bytes\n", (long)length*2);
    472         trie=builder->build(USTRINGTRIE_BUILD_SMALL, errorCode);
    473     }
    474 
    475     virtual ~UCharsTrieDictLookup() {
    476         delete builder;
    477         delete trie;
    478     }
    479 
    480 protected:
    481     UCharsTrieBuilder *builder;
    482     UCharsTrie *trie;
    483 };
    484 
    485 class UCharsTrieDictMatches : public UCharsTrieDictLookup {
    486 public:
    487     UCharsTrieDictMatches(const DictionaryTriePerfTest &perfTest)
    488             : UCharsTrieDictLookup(perfTest) {}
    489 
    490     virtual void call(UErrorCode *pErrorCode) {
    491         UText text=UTEXT_INITIALIZER;
    492         int32_t lengths[20];
    493         const ULine *lines=perf.getCachedLines();
    494         int32_t numLines=perf.getNumLines();
    495         for(int32_t i=0; i<numLines; ++i) {
    496             // Skip comment lines (start with a character below 'A').
    497             if(lines[i].name[0]<0x41) {
    498                 continue;
    499             }
    500             utext_openUChars(&text, lines[i].name, lines[i].len, pErrorCode);
    501             int32_t count=0;
    502             ucharsTrieMatches(*trie, &text, lines[i].len,
    503                               lengths, count, LENGTHOF(lengths));
    504             if(count==0 || lengths[count-1]!=lines[i].len) {
    505                 fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    506             }
    507         }
    508     }
    509 };
    510 
    511 class UCharsTrieDictContains : public UCharsTrieDictLookup {
    512 public:
    513     UCharsTrieDictContains(const DictionaryTriePerfTest &perfTest)
    514             : UCharsTrieDictLookup(perfTest) {}
    515 
    516     virtual void call(UErrorCode * /*pErrorCode*/) {
    517         const ULine *lines=perf.getCachedLines();
    518         int32_t numLines=perf.getNumLines();
    519         for(int32_t i=0; i<numLines; ++i) {
    520             // Skip comment lines (which start with a character below 'A').
    521             if(lines[i].name[0]<0x41) {
    522                 continue;
    523             }
    524             if(!USTRINGTRIE_HAS_VALUE(trie->reset().next(lines[i].name, lines[i].len))) {
    525                 fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    526             }
    527         }
    528     }
    529 };
    530 
    531 static inline int32_t thaiCharToByte(UChar32 c) {
    532     if(0xe00<=c && c<=0xefe) {
    533         return c&0xff;
    534     } else if(c==0x2e) {
    535         return 0xff;
    536     } else {
    537         return -1;
    538     }
    539 }
    540 
    541 static UBool thaiWordToBytes(const UChar *s, int32_t length,
    542                              CharString &str, UErrorCode &errorCode) {
    543     for(int32_t i=0; i<length; ++i) {
    544         UChar c=s[i];
    545         int32_t b=thaiCharToByte(c);
    546         if(b>=0) {
    547             str.append((char)b, errorCode);
    548         } else {
    549             fprintf(stderr, "thaiWordToBytes(): unable to encode U+%04X as a byte\n", c);
    550             return FALSE;
    551         }
    552     }
    553     return TRUE;
    554 }
    555 
    556 class BytesTrieDictLookup : public DictLookup {
    557 public:
    558     BytesTrieDictLookup(const DictionaryTriePerfTest &perfTest)
    559             : DictLookup(perfTest), trie(NULL), noDict(FALSE) {
    560         IcuToolErrorCode errorCode("BytesTrieDictLookup()");
    561         builder=new BytesTrieBuilder(errorCode);
    562         CharString str;
    563         const ULine *lines=perf.getCachedLines();
    564         int32_t numLines=perf.getNumLines();
    565         for(int32_t i=0; i<numLines; ++i) {
    566             // Skip comment lines (start with a character below 'A').
    567             if(lines[i].name[0]<0x41) {
    568                 continue;
    569             }
    570             if(!thaiWordToBytes(lines[i].name, lines[i].len, str.clear(), errorCode)) {
    571                 fprintf(stderr, "thaiWordToBytes(): failed for word %ld (0-based)\n", (long)i);
    572                 noDict=TRUE;
    573                 break;
    574             }
    575             builder->add(str.toStringPiece(), 0, errorCode);
    576         }
    577         if(!noDict) {
    578             int32_t length=builder->buildStringPiece(USTRINGTRIE_BUILD_SMALL, errorCode).length();
    579             printf("size of BytesTrie:           %6ld bytes\n", (long)length);
    580             trie=builder->build(USTRINGTRIE_BUILD_SMALL, errorCode);
    581         }
    582     }
    583 
    584     virtual ~BytesTrieDictLookup() {
    585         delete builder;
    586         delete trie;
    587     }
    588 
    589 protected:
    590     BytesTrieBuilder *builder;
    591     BytesTrie *trie;
    592     UBool noDict;
    593 };
    594 
    595 static int32_t
    596 bytesTrieMatches(BytesTrie &trie,
    597                  UText *text, int32_t textLimit,
    598                  int32_t *lengths, int &count, int limit ) {
    599     UChar32 c=utext_next32(text);
    600     if(c<0) {
    601         return 0;
    602     }
    603     UStringTrieResult result=trie.first(thaiCharToByte(c));
    604     int32_t numChars=1;
    605     count=0;
    606     for(;;) {
    607         if(USTRINGTRIE_HAS_VALUE(result)) {
    608             if(count<limit) {
    609                 // lengths[count++]=(int32_t)utext_getNativeIndex(text);
    610                 lengths[count++]=numChars;  // CompactTrieDictionary just counts chars too.
    611             }
    612             if(result==USTRINGTRIE_FINAL_VALUE) {
    613                 break;
    614             }
    615         } else if(result==USTRINGTRIE_NO_MATCH) {
    616             break;
    617         }
    618         if(numChars>=textLimit) {
    619             break;
    620         }
    621         UChar32 c=utext_next32(text);
    622         if(c<0) {
    623             break;
    624         }
    625         ++numChars;
    626         result=trie.next(thaiCharToByte(c));
    627     }
    628     return numChars;
    629 }
    630 
    631 class BytesTrieDictMatches : public BytesTrieDictLookup {
    632 public:
    633     BytesTrieDictMatches(const DictionaryTriePerfTest &perfTest)
    634             : BytesTrieDictLookup(perfTest) {}
    635 
    636     virtual void call(UErrorCode *pErrorCode) {
    637         if(noDict) {
    638             return;
    639         }
    640         UText text=UTEXT_INITIALIZER;
    641         int32_t lengths[20];
    642         const ULine *lines=perf.getCachedLines();
    643         int32_t numLines=perf.getNumLines();
    644         for(int32_t i=0; i<numLines; ++i) {
    645             // Skip comment lines (start with a character below 'A').
    646             if(lines[i].name[0]<0x41) {
    647                 continue;
    648             }
    649             utext_openUChars(&text, lines[i].name, lines[i].len, pErrorCode);
    650             int32_t count=0;
    651             bytesTrieMatches(*trie, &text, lines[i].len,
    652                              lengths, count, LENGTHOF(lengths));
    653             if(count==0 || lengths[count-1]!=lines[i].len) {
    654                 fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    655             }
    656         }
    657     }
    658 };
    659 
    660 class BytesTrieDictContains : public BytesTrieDictLookup {
    661 public:
    662     BytesTrieDictContains(const DictionaryTriePerfTest &perfTest)
    663             : BytesTrieDictLookup(perfTest) {}
    664 
    665     virtual void call(UErrorCode * /*pErrorCode*/) {
    666         if(noDict) {
    667             return;
    668         }
    669         const ULine *lines=perf.getCachedLines();
    670         int32_t numLines=perf.getNumLines();
    671         for(int32_t i=0; i<numLines; ++i) {
    672             const UChar *line=lines[i].name;
    673             // Skip comment lines (start with a character below 'A').
    674             if(line[0]<0x41) {
    675                 continue;
    676             }
    677             UStringTrieResult result=trie->first(thaiCharToByte(line[0]));
    678             int32_t lineLength=lines[i].len;
    679             for(int32_t j=1; j<lineLength; ++j) {
    680                 if(!USTRINGTRIE_HAS_NEXT(result)) {
    681                     fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    682                     break;
    683                 }
    684                 result=trie->next(thaiCharToByte(line[j]));
    685             }
    686             if(!USTRINGTRIE_HAS_VALUE(result)) {
    687                 fprintf(stderr, "word %ld (0-based) not found\n", (long)i);
    688             }
    689         }
    690     }
    691 };
    692 
    693 UPerfFunction *DictionaryTriePerfTest::runIndexedTest(int32_t index, UBool exec,
    694                                                       const char *&name, char * /*par*/) {
    695     if(hasFile()) {
    696         switch(index) {
    697         case 0:
    698             name="compacttriematches";
    699             if(exec) {
    700                 return new CompactTrieDictLookup(*this);
    701             }
    702             break;
    703         case 1:
    704             name="ucharstriematches";
    705             if(exec) {
    706                 return new UCharsTrieDictMatches(*this);
    707             }
    708             break;
    709         case 2:
    710             name="ucharstriecontains";
    711             if(exec) {
    712                 return new UCharsTrieDictContains(*this);
    713             }
    714             break;
    715         case 3:
    716             name="bytestriematches";
    717             if(exec) {
    718                 return new BytesTrieDictMatches(*this);
    719             }
    720             break;
    721         case 4:
    722             name="bytestriecontains";
    723             if(exec) {
    724                 return new BytesTrieDictContains(*this);
    725             }
    726             break;
    727         default:
    728             name="";
    729             break;
    730         }
    731     } else {
    732         if(index==0 && exec) {
    733             puts("Running BytesTrie perf tests on the .dat package file from the --sourcedir.\n"
    734                  "For UCharsTrie perf tests on a dictionary text file, specify the -f or --file-name.\n");
    735         }
    736         switch(index) {
    737         case 0:
    738             name="simplebinarysearch";
    739             if(exec) {
    740                 return new BinarySearchPackageLookup(*this);
    741             }
    742             break;
    743         case 1:
    744             name="prefixbinarysearch";
    745             if(exec) {
    746                 return new PrefixBinarySearchPackageLookup(*this);
    747             }
    748             break;
    749         case 2:
    750             name="bytestrie";
    751             if(exec) {
    752                 return new BytesTriePackageLookup(*this);
    753             }
    754             break;
    755         default:
    756             name="";
    757             break;
    758         }
    759     }
    760     return NULL;
    761 }
    762 
    763 int main(int argc, const char *argv[]) {
    764     IcuToolErrorCode errorCode("dicttrieperf main()");
    765     DictionaryTriePerfTest test(argc, argv, errorCode);
    766     if(errorCode.isFailure()) {
    767         fprintf(stderr, "DictionaryTriePerfTest() failed: %s\n", errorCode.errorName());
    768         test.usage();
    769         return errorCode.reset();
    770     }
    771     if(!test.run()) {
    772         fprintf(stderr, "FAILED: Tests could not be run, please check the arguments.\n");
    773         return -1;
    774     }
    775     return 0;
    776 }
    777