1 /***********************************************************************
2 * Copyright (C) 2016 and later: Unicode, Inc. and others.
3 * License & terms of use: http://www.unicode.org/copyright.html#License
4 ***********************************************************************
5 ***********************************************************************
7 * Copyright (C) 2001-2012 IBM, Inc. All Rights Reserved.
9 ***********************************************************************/
10 /********************************************************************************
14 * Modification History:
16 * Andy Heninger First Version
18 *********************************************************************************
22 // This program tests string collation and sort key generation performance.
23 // Three APIs can be teste: ICU C , Unix strcoll, strxfrm and Windows LCMapString
24 // A file of names is required as input, one per line. It must be in utf-8 or utf-16 format,
25 // and include a byte order mark. Either LE or BE format is OK.
28 const char gUsageString[] =
29 "usage: collperf options...\n"
30 "-help Display this message.\n"
31 "-file file_name utf-16 format file of names.\n"
32 "-locale name ICU locale to use. Default is en_US\n"
33 "-rules file_name Collation rules file (overrides locale)\n"
34 "-langid 0x1234 Windows Language ID number. Default to value for -locale option\n"
35 " see http://msdn.microsoft.com/library/psdk/winbase/nls_8xo3.htm\n"
36 "-win Run test using Windows native services. (ICU is default)\n"
37 "-unix Run test using Unix strxfrm, strcoll services.\n"
38 "-uselen Use API with string lengths. Default is null-terminated strings\n"
39 "-usekeys Run tests using sortkeys rather than strcoll\n"
40 "-strcmp Run tests using u_strcmp rather than strcoll\n"
41 "-strcmpCPO Run tests using u_strcmpCodePointOrder rather than strcoll\n"
42 "-loop nnnn Loopcount for test. Adjust for reasonable total running time.\n"
43 "-iloop n Inner Loop Count. Default = 1. Number of calls to function\n"
44 " under test at each call point. For measuring test overhead.\n"
45 "-terse Terse numbers-only output. Intended for use by scripts.\n"
46 "-french French accent ordering\n"
47 "-frenchoff No French accent ordering (for use with French locales.)\n"
48 "-norm Normalizing mode on\n"
49 "-shifted Shifted mode\n"
50 "-lower Lower case first\n"
51 "-upper Upper case first\n"
52 "-case Enable separate case level\n"
53 "-level n Sort level, 1 to 5, for Primary, Secndary, Tertiary, Quaternary, Identical\n"
54 "-keyhist Produce a table sort key size vs. string length\n"
55 "-binsearch Binary Search timing test\n"
56 "-keygen Sort Key Generation timing test\n"
57 "-qsort Quicksort timing test\n"
58 "-iter Iteration Performance Test\n"
59 "-dump Display strings, sort keys and CEs.\n"
71 #include <unicode/utypes.h>
72 #include <unicode/ucol.h>
73 #include <unicode/ucoleitr.h>
74 #include <unicode/uloc.h>
75 #include <unicode/ustring.h>
76 #include <unicode/ures.h>
77 #include <unicode/uchar.h>
78 #include <unicode/ucnv.h>
79 #include <unicode/utf8.h>
85 // Stubs for Windows API functions when building on UNIXes.
88 inline int CompareStringW(DWORD, DWORD, UChar *, int, UChar *, int) {return 0;}
90 unsigned long timeGetTime() {
93 unsigned long val = t.tv_sec * 1000; // Let it overflow. Who cares.
94 val += t.tv_usec / 1000;
97 inline int LCMapStringW(DWORD, DWORD, UChar *, int, UChar *, int) {return 0;}
98 const int LCMAP_SORTKEY = 0;
99 #define MAKELCID(a,b) 0
100 const int SORT_DEFAULT = 0;
106 // Command line option variables
107 // These global variables are set according to the options specified
108 // on the command line by the user.
109 char * opt_fName = 0;
110 const char * opt_locale = "en_US";
111 int opt_langid = 0; // Defaults to value corresponding to opt_locale.
112 char * opt_rules = 0;
113 UBool opt_help = FALSE;
114 int opt_loopCount = 1;
115 int opt_iLoopCount = 1;
116 UBool opt_terse = FALSE;
117 UBool opt_qsort = FALSE;
118 UBool opt_binsearch = FALSE;
119 UBool opt_icu = TRUE;
120 UBool opt_win = FALSE; // Run with Windows native functions.
121 UBool opt_unix = FALSE; // Run with UNIX strcoll, strxfrm functions.
122 UBool opt_uselen = FALSE;
123 UBool opt_usekeys = FALSE;
124 UBool opt_strcmp = FALSE;
125 UBool opt_strcmpCPO = FALSE;
126 UBool opt_norm = FALSE;
127 UBool opt_keygen = FALSE;
128 UBool opt_french = FALSE;
129 UBool opt_frenchoff = FALSE;
130 UBool opt_shifted = FALSE;
131 UBool opt_lower = FALSE;
132 UBool opt_upper = FALSE;
133 UBool opt_case = FALSE;
135 UBool opt_keyhist = FALSE;
136 UBool opt_itertest = FALSE;
137 UBool opt_dump = FALSE;
142 // Definitions for the command line options
146 enum {FLAG, NUM, STRING} type;
151 {"-file", OptSpec::STRING, &opt_fName},
152 {"-locale", OptSpec::STRING, &opt_locale},
153 {"-langid", OptSpec::NUM, &opt_langid},
154 {"-rules", OptSpec::STRING, &opt_rules},
155 {"-qsort", OptSpec::FLAG, &opt_qsort},
156 {"-binsearch", OptSpec::FLAG, &opt_binsearch},
157 {"-iter", OptSpec::FLAG, &opt_itertest},
158 {"-win", OptSpec::FLAG, &opt_win},
159 {"-unix", OptSpec::FLAG, &opt_unix},
160 {"-uselen", OptSpec::FLAG, &opt_uselen},
161 {"-usekeys", OptSpec::FLAG, &opt_usekeys},
162 {"-strcmp", OptSpec::FLAG, &opt_strcmp},
163 {"-strcmpCPO", OptSpec::FLAG, &opt_strcmpCPO},
164 {"-norm", OptSpec::FLAG, &opt_norm},
165 {"-french", OptSpec::FLAG, &opt_french},
166 {"-frenchoff", OptSpec::FLAG, &opt_frenchoff},
167 {"-shifted", OptSpec::FLAG, &opt_shifted},
168 {"-lower", OptSpec::FLAG, &opt_lower},
169 {"-upper", OptSpec::FLAG, &opt_upper},
170 {"-case", OptSpec::FLAG, &opt_case},
171 {"-level", OptSpec::NUM, &opt_level},
172 {"-keyhist", OptSpec::FLAG, &opt_keyhist},
173 {"-keygen", OptSpec::FLAG, &opt_keygen},
174 {"-loop", OptSpec::NUM, &opt_loopCount},
175 {"-iloop", OptSpec::NUM, &opt_iLoopCount},
176 {"-terse", OptSpec::FLAG, &opt_terse},
177 {"-dump", OptSpec::FLAG, &opt_dump},
178 {"-help", OptSpec::FLAG, &opt_help},
179 {"-?", OptSpec::FLAG, &opt_help},
180 {0, OptSpec::FLAG, 0}
184 //---------------------------------------------------------------------------
186 // Global variables pointing to and describing the test file
188 //---------------------------------------------------------------------------
193 // Each line from the source file (containing a name, presumably) gets
194 // one of these structs.
207 Line *gFileLines; // Ptr to array of Line structs, one per line in the file.
218 //---------------------------------------------------------------------------
220 // ProcessOptions() Function to read the command line options.
222 //---------------------------------------------------------------------------
223 UBool ProcessOptions(int argc, const char **argv, OptSpec opts[])
227 const char *pArgName;
230 for (argNum=1; argNum<argc; argNum++) {
231 pArgName = argv[argNum];
232 for (pOpt = opts; pOpt->name != 0; pOpt++) {
233 if (strcmp(pOpt->name, pArgName) == 0) {
234 switch (pOpt->type) {
236 *(UBool *)(pOpt->pVar) = TRUE;
238 case OptSpec::STRING:
240 if (argNum >= argc) {
241 fprintf(stderr, "value expected for \"%s\" option.\n", pOpt->name);
244 *(const char **)(pOpt->pVar) = argv[argNum];
248 if (argNum >= argc) {
249 fprintf(stderr, "value expected for \"%s\" option.\n", pOpt->name);
253 i = strtol(argv[argNum], &endp, 0);
254 if (endp == argv[argNum]) {
255 fprintf(stderr, "integer value expected for \"%s\" option.\n", pOpt->name);
258 *(int *)(pOpt->pVar) = i;
265 fprintf(stderr, "Unrecognized option \"%s\"\n", pArgName);
272 //---------------------------------------------------------------------------------------
274 // Comparison functions for use by qsort.
276 // Six flavors, ICU or Windows, SortKey or String Compare, Strings with length
277 // or null terminated.
279 //---------------------------------------------------------------------------------------
280 int ICUstrcmpK(const void *a, const void *b) {
282 int t = strcmp((*(Line **)a)->icuSortKey, (*(Line **)b)->icuSortKey);
287 int ICUstrcmpL(const void *a, const void *b) {
290 t = ucol_strcoll(gCol, (*(Line **)a)->name, (*(Line **)a)->len, (*(Line **)b)->name, (*(Line **)b)->len);
291 if (t == UCOL_LESS) return -1;
292 if (t == UCOL_GREATER) return +1;
297 int ICUstrcmp(const void *a, const void *b) {
300 t = ucol_strcoll(gCol, (*(Line **)a)->name, -1, (*(Line **)b)->name, -1);
301 if (t == UCOL_LESS) return -1;
302 if (t == UCOL_GREATER) return +1;
307 int Winstrcmp(const void *a, const void *b) {
310 t = CompareStringW(gWinLCID, 0, (*(Line **)a)->name, -1, (*(Line **)b)->name, -1);
315 int UNIXstrcmp(const void *a, const void *b) {
318 t = strcoll((*(Line **)a)->unixName, (*(Line **)b)->unixName);
323 int WinstrcmpL(const void *a, const void *b) {
326 t = CompareStringW(gWinLCID, 0, (*(Line **)a)->name, (*(Line **)a)->len, (*(Line **)b)->name, (*(Line **)b)->len);
331 int WinstrcmpK(const void *a, const void *b) {
333 int t = strcmp((*(Line **)a)->winSortKey, (*(Line **)b)->winSortKey);
338 //---------------------------------------------------------------------------------------
340 // Function for sorting the names (lines) into a random order.
341 // Order is based on a hash of the ICU Sort key for the lines
342 // The randomized order is used as input for the sorting timing tests.
344 //---------------------------------------------------------------------------------------
345 int ICURandomCmp(const void *a, const void *b) {
346 char *ask = (*(Line **)a)->icuSortKey;
347 char *bsk = (*(Line **)b)->icuSortKey;
352 aVal += aVal*37 + *ask++;
355 bVal += bVal*37 + *bsk++;
361 else if (aVal > bVal) {
367 //---------------------------------------------------------------------------------------
369 // doKeyGen() Key Generation Timing Test
371 //---------------------------------------------------------------------------------------
380 // Adjust loop count to compensate for file size. Should be order n
381 double dLoopCount = double(opt_loopCount) * (1000. / double(gNumFileLines));
382 int adj_loopCount = int(dLoopCount);
383 if (adj_loopCount < 1) adj_loopCount = 1;
386 unsigned long startTime = timeGetTime();
389 for (loops=0; loops<adj_loopCount; loops++) {
390 for (line=0; line < gNumFileLines; line++) {
392 len = gFileLines[line].len;
394 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
395 t=LCMapStringW(gWinLCID, LCMAP_SORTKEY,
396 gFileLines[line].name, len,
397 (unsigned short *)gFileLines[line].winSortKey, 5000); // TODO something with length.
404 for (loops=0; loops<adj_loopCount; loops++) {
405 for (line=0; line < gNumFileLines; line++) {
407 len = gFileLines[line].len;
409 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
410 t = ucol_getSortKey(gCol, gFileLines[line].name, len, (unsigned char *)gFileLines[line].icuSortKey, 5000);
417 for (loops=0; loops<adj_loopCount; loops++) {
418 for (line=0; line < gNumFileLines; line++) {
419 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
420 t = strxfrm(gFileLines[line].unixSortKey, gFileLines[line].unixName, 5000);
426 unsigned long elapsedTime = timeGetTime() - startTime;
427 int ns = (int)(float(1000000) * (float)elapsedTime / (float)(adj_loopCount*gNumFileLines));
429 if (opt_terse == FALSE) {
430 printf("Sort Key Generation: total # of keys = %d\n", loops*gNumFileLines);
431 printf("Sort Key Generation: time per key = %d ns\n", ns);
439 for (line=0; line<gNumFileLines; line++) {
440 totalChars += u_strlen(gFileLines[line].name);
442 totalKeyLen += strlen(gFileLines[line].winSortKey);
445 totalKeyLen += strlen(gFileLines[line].icuSortKey);
448 totalKeyLen += strlen(gFileLines[line].unixSortKey);
452 if (opt_terse == FALSE) {
453 printf("Key Length / character = %f\n", (float)totalKeyLen / (float)totalChars);
455 printf("%f, ", (float)totalKeyLen / (float)totalChars);
461 //---------------------------------------------------------------------------------------
463 // doBinarySearch() Binary Search timing test. Each name from the list
464 // is looked up in the full sorted list of names.
466 //---------------------------------------------------------------------------------------
467 void doBinarySearch()
474 unsigned long elapsedTime = 0;
476 // Adjust loop count to compensate for file size. Should be order n (lookups) * log n (compares/lookup)
477 // Accurate timings do not depend on this being perfect. The correction is just to try to
478 // get total running times of about the right order, so the that user doesn't need to
479 // manually adjust the loop count for every different file size.
480 double dLoopCount = double(opt_loopCount) * 3000. / (log10((double)gNumFileLines) * double(gNumFileLines));
481 if (opt_usekeys) dLoopCount *= 5;
482 int adj_loopCount = int(dLoopCount);
483 if (adj_loopCount < 1) adj_loopCount = 1;
486 for (;;) { // not really a loop, just allows "break" to work, to simplify
487 // inadvertantly running more than one test through here.
488 if (opt_strcmp || opt_strcmpCPO)
490 unsigned long startTime = timeGetTime();
491 typedef int32_t (U_EXPORT2 *PF)(const UChar *, const UChar *);
493 if (opt_strcmpCPO) {pf = u_strcmpCodePointOrder;}
494 //if (opt_strcmp && opt_win) {pf = (PF)wcscmp;} // Damn the difference between int32_t and int
495 // which forces the use of a cast here.
498 for (loops=0; loops<adj_loopCount; loops++) {
500 for (line=0; line < gNumFileLines; line++) {
501 int hi = gNumFileLines-1;
505 int newGuess = (hi + lo) / 2;
506 if (newGuess == guess)
509 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
510 r = (*pf)((gSortedLines[line])->name, (gSortedLines[guess])->name);
522 elapsedTime = timeGetTime() - startTime;
529 unsigned long startTime = timeGetTime();
530 UCollationResult r = UCOL_EQUAL;
531 for (loops=0; loops<adj_loopCount; loops++) {
533 for (line=0; line < gNumFileLines; line++) {
537 lineLen = (gSortedLines[line])->len;
539 int hi = gNumFileLines-1;
543 int newGuess = (hi + lo) / 2;
544 if (newGuess == guess)
549 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
550 ri = strcmp((gSortedLines[line])->icuSortKey, (gSortedLines[guess])->icuSortKey);
553 r=UCOL_GREATER; if(ri<0) {r=UCOL_LESS;} else if (ri==0) {r=UCOL_EQUAL;}
558 guessLen = (gSortedLines[guess])->len;
560 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
561 r = ucol_strcoll(gCol, (gSortedLines[line])->name, lineLen, (gSortedLines[guess])->name, guessLen);
574 elapsedTime = timeGetTime() - startTime;
580 unsigned long startTime = timeGetTime();
582 for (loops=0; loops<adj_loopCount; loops++) {
584 for (line=0; line < gNumFileLines; line++) {
588 lineLen = (gSortedLines[line])->len;
590 int hi = gNumFileLines-1;
594 int newGuess = (hi + lo) / 2;
595 if (newGuess == guess)
599 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
600 r = strcmp((gSortedLines[line])->winSortKey, (gSortedLines[guess])->winSortKey);
608 guessLen = (gSortedLines[guess])->len;
610 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
611 r = CompareStringW(gWinLCID, 0, (gSortedLines[line])->name, lineLen, (gSortedLines[guess])->name, guessLen);
614 if (opt_terse == FALSE) {
615 fprintf(stderr, "Error returned from Windows CompareStringW.\n");
621 if (r== 2) // strings ==
623 if (r == 1) // line < guess
630 elapsedTime = timeGetTime() - startTime;
636 unsigned long startTime = timeGetTime();
638 for (loops=0; loops<adj_loopCount; loops++) {
640 for (line=0; line < gNumFileLines; line++) {
641 int hi = gNumFileLines-1;
645 int newGuess = (hi + lo) / 2;
646 if (newGuess == guess)
650 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
651 r = strcmp((gSortedLines[line])->unixSortKey, (gSortedLines[guess])->unixSortKey);
657 for (iLoop=0; iLoop < opt_iLoopCount; iLoop++) {
658 r = strcoll((gSortedLines[line])->unixName, (gSortedLines[guess])->unixName);
662 fprintf(stderr, "Error %d returned from strcoll.\n", errno);
667 if (r == 0) // strings ==
669 if (r < 0) // line < guess
676 elapsedTime = timeGetTime() - startTime;
682 int ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
683 if (opt_terse == FALSE) {
684 printf("binary search: total # of string compares = %d\n", gCount);
685 printf("binary search: compares per loop = %d\n", gCount / loops);
686 printf("binary search: time per compare = %d ns\n", ns);
696 //---------------------------------------------------------------------------------------
698 // doQSort() The quick sort timing test. Uses the C library qsort function.
700 //---------------------------------------------------------------------------------------
703 Line **sortBuf = new Line *[gNumFileLines];
705 // Adjust loop count to compensate for file size. QSort should be n log(n)
706 double dLoopCount = double(opt_loopCount) * 3000. / (log10((double)gNumFileLines) * double(gNumFileLines));
707 if (opt_usekeys) dLoopCount *= 5;
708 int adj_loopCount = int(dLoopCount);
709 if (adj_loopCount < 1) adj_loopCount = 1;
713 unsigned long startTime = timeGetTime();
714 if (opt_win && opt_usekeys) {
715 for (i=0; i<opt_loopCount; i++) {
716 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
717 qsort(sortBuf, gNumFileLines, sizeof(Line *), WinstrcmpK);
721 else if (opt_win && opt_uselen) {
722 for (i=0; i<adj_loopCount; i++) {
723 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
724 qsort(sortBuf, gNumFileLines, sizeof(Line *), WinstrcmpL);
729 else if (opt_win && !opt_uselen) {
730 for (i=0; i<adj_loopCount; i++) {
731 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
732 qsort(sortBuf, gNumFileLines, sizeof(Line *), Winstrcmp);
736 else if (opt_icu && opt_usekeys) {
737 for (i=0; i<adj_loopCount; i++) {
738 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
739 qsort(sortBuf, gNumFileLines, sizeof(Line *), ICUstrcmpK);
743 else if (opt_icu && opt_uselen) {
744 for (i=0; i<adj_loopCount; i++) {
745 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
746 qsort(sortBuf, gNumFileLines, sizeof(Line *), ICUstrcmpL);
751 else if (opt_icu && !opt_uselen) {
752 for (i=0; i<adj_loopCount; i++) {
753 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
754 qsort(sortBuf, gNumFileLines, sizeof(Line *), ICUstrcmp);
758 else if (opt_unix && !opt_usekeys) {
759 for (i=0; i<adj_loopCount; i++) {
760 memcpy(sortBuf, gRandomLines, gNumFileLines * sizeof(Line *));
761 qsort(sortBuf, gNumFileLines, sizeof(Line *), UNIXstrcmp);
765 unsigned long elapsedTime = timeGetTime() - startTime;
766 int ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
767 if (opt_terse == FALSE) {
768 printf("qsort: total # of string compares = %d\n", gCount);
769 printf("qsort: time per compare = %d ns\n", ns);
777 //---------------------------------------------------------------------------------------
779 // doKeyHist() Output a table of data for
780 // average sort key size vs. string length.
782 //---------------------------------------------------------------------------------------
787 // Find the maximum string length
788 for (i=0; i<gNumFileLines; i++) {
789 if (gFileLines[i].len > maxLen) maxLen = gFileLines[i].len;
792 // Allocate arrays to hold the histogram data
793 int *accumulatedLen = new int[maxLen+1];
794 int *numKeysOfSize = new int[maxLen+1];
795 for (i=0; i<=maxLen; i++) {
796 accumulatedLen[i] = 0;
797 numKeysOfSize[i] = 0;
800 // Fill the arrays...
801 for (i=0; i<gNumFileLines; i++) {
802 int len = gFileLines[i].len;
803 accumulatedLen[len] += strlen(gFileLines[i].icuSortKey);
804 numKeysOfSize[len] += 1;
807 // And write out averages
808 printf("String Length, Avg Key Length, Avg Key Len per char\n");
809 for (i=1; i<=maxLen; i++) {
810 if (numKeysOfSize[i] > 0) {
811 printf("%d, %f, %f\n", i, (float)accumulatedLen[i] / (float)numKeysOfSize[i],
812 (float)accumulatedLen[i] / (float)(numKeysOfSize[i] * i));
815 delete []accumulatedLen;
816 delete []numKeysOfSize ;
819 //---------------------------------------------------------------------------------------
821 // doForwardIterTest(UBool) Forward iteration test
822 // argument null-terminated string used
824 //---------------------------------------------------------------------------------------
825 void doForwardIterTest(UBool haslen) {
828 UErrorCode error = U_ZERO_ERROR;
829 printf("\n\nPerforming forward iteration performance test with ");
832 printf("non-null terminated data -----------\n");
835 printf("null terminated data -----------\n");
837 printf("performance test on strings from file -----------\n");
839 UChar dummytext[] = {0, 0};
840 UCollationElements *iter = ucol_openElements(gCol, NULL, 0, &error);
841 ucol_setText(iter, dummytext, 1, &error);
844 unsigned long startTime = timeGetTime();
845 while (count < opt_loopCount) {
847 while (linecount < gNumFileLines) {
848 UChar *str = gFileLines[linecount].name;
849 int strlen = haslen?gFileLines[linecount].len:-1;
850 ucol_setText(iter, str, strlen, &error);
851 while (ucol_next(iter, &error) != UCOL_NULLORDER) {
859 unsigned long elapsedTime = timeGetTime() - startTime;
860 printf("elapsedTime %ld\n", elapsedTime);
862 // empty loop recalculation
864 startTime = timeGetTime();
865 while (count < opt_loopCount) {
867 while (linecount < gNumFileLines) {
868 UChar *str = gFileLines[linecount].name;
869 int strlen = haslen?gFileLines[linecount].len:-1;
870 ucol_setText(iter, str, strlen, &error);
875 elapsedTime -= (timeGetTime() - startTime);
876 printf("elapsedTime %ld\n", elapsedTime);
878 ucol_closeElements(iter);
880 int ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
881 printf("Total number of strings compared %d in %d loops\n", gNumFileLines,
883 printf("Average time per ucol_next() nano seconds %d\n", ns);
885 printf("performance test on skipped-5 concatenated strings from file -----------\n");
889 // appending all the strings
891 while (linecount < gNumFileLines) {
892 strlen += haslen?gFileLines[linecount].len:
893 u_strlen(gFileLines[linecount].name);
896 str = (UChar *)malloc(sizeof(UChar) * strlen);
899 while (strindex < strlen) {
901 len += haslen?gFileLines[linecount].len:
902 u_strlen(gFileLines[linecount].name);
903 memcpy(str + strindex, gFileLines[linecount].name,
904 sizeof(UChar) * len);
909 printf("Total size of strings %d\n", strlen);
917 iter = ucol_openElements(gCol, str, strlen, &error);
919 strlen = u_strlen(str);
921 strlen -= 5; // any left over characters are not iterated,
922 // this is to ensure the backwards and forwards iterators
923 // gets the same position
924 startTime = timeGetTime();
925 while (count < opt_loopCount) {
928 ucol_setOffset(iter, strindex, &error);
930 if (ucol_next(iter, &error) == UCOL_NULLORDER) {
937 if (strindex > strlen) {
940 ucol_setOffset(iter, strindex, &error);
947 elapsedTime = timeGetTime() - startTime;
948 printf("elapsedTime %ld\n", elapsedTime);
950 // empty loop recalculation
953 startTime = timeGetTime();
954 while (count < opt_loopCount) {
957 ucol_setOffset(iter, strindex, &error);
963 if (strindex > strlen) {
966 ucol_setOffset(iter, strindex, &error);
972 elapsedTime -= (timeGetTime() - startTime);
973 printf("elapsedTime %ld\n", elapsedTime);
975 ucol_closeElements(iter);
977 printf("gCount %d\n", gCount);
978 ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
979 printf("Average time per ucol_next() nano seconds %d\n", ns);
982 //---------------------------------------------------------------------------------------
984 // doBackwardIterTest(UBool) Backwards iteration test
985 // argument null-terminated string used
987 //---------------------------------------------------------------------------------------
988 void doBackwardIterTest(UBool haslen) {
990 UErrorCode error = U_ZERO_ERROR;
991 printf("\n\nPerforming backward iteration performance test with ");
994 printf("non-null terminated data -----------\n");
997 printf("null terminated data -----------\n");
1000 printf("performance test on strings from file -----------\n");
1002 UCollationElements *iter = ucol_openElements(gCol, NULL, 0, &error);
1003 UChar dummytext[] = {0, 0};
1004 ucol_setText(iter, dummytext, 1, &error);
1007 unsigned long startTime = timeGetTime();
1008 while (count < opt_loopCount) {
1010 while (linecount < gNumFileLines) {
1011 UChar *str = gFileLines[linecount].name;
1012 int strlen = haslen?gFileLines[linecount].len:-1;
1013 ucol_setText(iter, str, strlen, &error);
1014 while (ucol_previous(iter, &error) != UCOL_NULLORDER) {
1022 unsigned long elapsedTime = timeGetTime() - startTime;
1024 printf("elapsedTime %ld\n", elapsedTime);
1026 // empty loop recalculation
1028 startTime = timeGetTime();
1029 while (count < opt_loopCount) {
1031 while (linecount < gNumFileLines) {
1032 UChar *str = gFileLines[linecount].name;
1033 int strlen = haslen?gFileLines[linecount].len:-1;
1034 ucol_setText(iter, str, strlen, &error);
1039 elapsedTime -= (timeGetTime() - startTime);
1041 printf("elapsedTime %ld\n", elapsedTime);
1042 ucol_closeElements(iter);
1044 int ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
1045 printf("Total number of strings compared %d in %d loops\n", gNumFileLines,
1047 printf("Average time per ucol_previous() nano seconds %d\n", ns);
1049 printf("performance test on skipped-5 concatenated strings from file -----------\n");
1053 // appending all the strings
1055 while (linecount < gNumFileLines) {
1056 strlen += haslen?gFileLines[linecount].len:
1057 u_strlen(gFileLines[linecount].name);
1060 str = (UChar *)malloc(sizeof(UChar) * strlen);
1063 while (strindex < strlen) {
1065 len += haslen?gFileLines[linecount].len:
1066 u_strlen(gFileLines[linecount].name);
1067 memcpy(str + strindex, gFileLines[linecount].name,
1068 sizeof(UChar) * len);
1073 printf("Total size of strings %d\n", strlen);
1082 iter = ucol_openElements(gCol, str, strlen, &error);
1084 strlen = u_strlen(str);
1087 startTime = timeGetTime();
1088 while (count < opt_loopCount) {
1091 ucol_setOffset(iter, strindex, &error);
1093 if (ucol_previous(iter, &error) == UCOL_NULLORDER) {
1100 if (strindex > strlen) {
1103 ucol_setOffset(iter, strindex, &error);
1110 elapsedTime = timeGetTime() - startTime;
1111 printf("elapsedTime %ld\n", elapsedTime);
1113 // empty loop recalculation
1116 startTime = timeGetTime();
1117 while (count < opt_loopCount) {
1120 ucol_setOffset(iter, strindex, &error);
1126 if (strindex > strlen) {
1129 ucol_setOffset(iter, strindex, &error);
1135 elapsedTime -= (timeGetTime() - startTime);
1136 printf("elapsedTime %ld\n", elapsedTime);
1137 ucol_closeElements(iter);
1139 printf("gCount %d\n", gCount);
1140 ns = (int)(float(1000000) * (float)elapsedTime / (float)gCount);
1141 printf("Average time per ucol_previous() nano seconds %d\n", ns);
1144 //---------------------------------------------------------------------------------------
1146 // doIterTest() Iteration test
1148 //---------------------------------------------------------------------------------------
1150 doForwardIterTest(opt_uselen);
1151 doBackwardIterTest(opt_uselen);
1155 //----------------------------------------------------------------------------------------
1157 // UnixConvert -- Convert the lines of the file to the encoding for UNIX
1158 // Since it appears that Unicode support is going in the general
1159 // direction of the use of UTF-8 locales, that is the approach
1160 // that is used here.
1162 //----------------------------------------------------------------------------------------
1163 void UnixConvert() {
1166 UConverter *cvrtr; // An ICU code page converter.
1167 UErrorCode status = U_ZERO_ERROR;
1170 cvrtr = ucnv_open("utf-8", &status); // we are just doing UTF-8 locales for now.
1171 if (U_FAILURE(status)) {
1172 fprintf(stderr, "ICU Converter open failed.: %s\n", u_errorName(status));
1176 for (line=0; line < gNumFileLines; line++) {
1177 int sizeNeeded = ucnv_fromUChars(cvrtr,
1178 0, // ptr to target buffer.
1179 0, // length of target buffer.
1180 gFileLines[line].name,
1181 -1, // source is null terminated
1183 if (status != U_BUFFER_OVERFLOW_ERROR && status != U_ZERO_ERROR) {
1184 //fprintf(stderr, "Conversion from Unicode, something is wrong.\n");
1187 status = U_ZERO_ERROR;
1188 gFileLines[line].unixName = new char[sizeNeeded+1];
1189 sizeNeeded = ucnv_fromUChars(cvrtr,
1190 gFileLines[line].unixName, // ptr to target buffer.
1191 sizeNeeded+1, // length of target buffer.
1192 gFileLines[line].name,
1193 -1, // source is null terminated
1195 if (U_FAILURE(status)) {
1196 fprintf(stderr, "ICU Conversion Failed.: %d\n", status);
1199 gFileLines[line].unixName[sizeNeeded] = 0;
1205 //----------------------------------------------------------------------------------------
1207 // class UCharFile Class to hide all the gorp to read a file in
1208 // and produce a stream of UChars.
1210 //----------------------------------------------------------------------------------------
1213 UCharFile(const char *fileName);
1216 UBool eof() {return fEof;};
1217 UBool error() {return fError;};
1220 UCharFile (const UCharFile & /*other*/) {}; // No copy constructor.
1221 UCharFile & operator = (const UCharFile &/*other*/) {return *this;}; // No assignment op
1227 UChar fPending2ndSurrogate;
1229 enum {UTF16LE, UTF16BE, UTF8} fEncoding;
1232 UCharFile::UCharFile(const char * fileName) {
1236 fFile = fopen(fName, "rb");
1237 fPending2ndSurrogate = 0;
1238 if (fFile == NULL) {
1239 fprintf(stderr, "Can not open file \"%s\"\n", opt_fName);
1244 // Look for the byte order mark at the start of the file.
1246 int BOMC1, BOMC2, BOMC3;
1247 BOMC1 = fgetc(fFile);
1248 BOMC2 = fgetc(fFile);
1250 if (BOMC1 == 0xff && BOMC2 == 0xfe) {
1251 fEncoding = UTF16LE; }
1252 else if (BOMC1 == 0xfe && BOMC2 == 0xff) {
1253 fEncoding = UTF16BE; }
1254 else if (BOMC1 == 0xEF && BOMC2 == 0xBB && (BOMC3 = fgetc(fFile)) == 0xBF ) {
1258 fprintf(stderr, "collperf: file \"%s\" encoding must be UTF-8 or UTF-16, and "
1259 "must include a BOM.\n", fileName);
1266 UCharFile::~UCharFile() {
1272 UChar UCharFile::get() {
1274 switch (fEncoding) {
1301 if (fPending2ndSurrogate != 0) {
1302 c = fPending2ndSurrogate;
1303 fPending2ndSurrogate = 0;
1307 int ch = fgetc(fFile); // Note: c and ch are separate cause eof test doesn't work on UChar type.
1315 // It's ascii. No further utf-8 conversion.
1320 // Figure out the lenght of the char and read the rest of the bytes
1321 // into a temp array.
1323 if (ch >= 0xF0) {nBytes=4;}
1324 else if (ch >= 0xE0) {nBytes=3;}
1325 else if (ch >= 0xC0) {nBytes=2;}
1327 fprintf(stderr, "utf-8 encoded file contains corrupt data.\n");
1332 unsigned char bytes[10];
1333 bytes[0] = (unsigned char)ch;
1335 for (i=1; i<nBytes; i++) {
1336 bytes[i] = fgetc(fFile);
1337 if (bytes[i] < 0x80 || bytes[i] >= 0xc0) {
1338 fprintf(stderr, "utf-8 encoded file contains corrupt data.\n");
1344 // Convert the bytes from the temp array to a Unicode char.
1347 U8_NEXT_UNSAFE(bytes, i, cp);
1350 if (cp >= 0x10000) {
1351 // The code point needs to be broken up into a utf-16 surrogate pair.
1352 // Process first half this time through the main loop, and
1353 // remember the other half for the next time through.
1356 UTF16_APPEND_CHAR_UNSAFE(utf16Buf, i, cp);
1357 fPending2ndSurrogate = utf16Buf[1];
1363 c = 0xFFFD; /* Error, unspecified codepage*/
1364 fprintf(stderr, "UCharFile: Error: unknown fEncoding\n");
1370 //----------------------------------------------------------------------------------------
1372 // openRulesCollator - Command line specified a rules file. Read it in
1373 // and open a collator with it.
1375 //----------------------------------------------------------------------------------------
1376 UCollator *openRulesCollator() {
1377 UCharFile f(opt_rules);
1383 UChar *buf = (UChar *)malloc(bufLen * sizeof(UChar));
1399 buf = (UChar *)realloc(buf, bufLen);
1408 UErrorCode status = U_ZERO_ERROR;
1409 UCollator *coll = ucol_openRules(buf, u_strlen(buf), UCOL_OFF,
1410 UCOL_DEFAULT_STRENGTH, NULL, &status);
1411 if (U_FAILURE(status)) {
1412 fprintf(stderr, "ICU ucol_openRules() open failed.: %d\n", status);
1423 //----------------------------------------------------------------------------------------
1425 // Main -- process command line, read in and pre-process the test file,
1426 // call other functions to do the actual tests.
1428 //----------------------------------------------------------------------------------------
1429 int main(int argc, const char** argv) {
1430 if (ProcessOptions(argc, argv, opts) != TRUE || opt_help || opt_fName == 0) {
1431 printf(gUsageString);
1435 // Make sure that we've only got one API selected.
1436 if (opt_unix || opt_win) opt_icu = FALSE;
1437 if (opt_unix) opt_win = FALSE;
1440 // Set up an ICU collator
1442 UErrorCode status = U_ZERO_ERROR;
1444 if (opt_rules != 0) {
1445 gCol = openRulesCollator();
1446 if (gCol == 0) {return -1;}
1449 gCol = ucol_open(opt_locale, &status);
1450 if (U_FAILURE(status)) {
1451 fprintf(stderr, "Collator creation failed.: %d\n", status);
1455 if (status==U_USING_DEFAULT_WARNING && opt_terse==FALSE) {
1456 fprintf(stderr, "Warning, U_USING_DEFAULT_WARNING for %s\n", opt_locale);
1458 if (status==U_USING_FALLBACK_WARNING && opt_terse==FALSE) {
1459 fprintf(stderr, "Warning, U_USING_FALLBACK_ERROR for %s\n", opt_locale);
1463 ucol_setAttribute(gCol, UCOL_NORMALIZATION_MODE, UCOL_ON, &status);
1465 if (opt_french && opt_frenchoff) {
1466 fprintf(stderr, "collperf: Error, specified both -french and -frenchoff options.");
1470 ucol_setAttribute(gCol, UCOL_FRENCH_COLLATION, UCOL_ON, &status);
1472 if (opt_frenchoff) {
1473 ucol_setAttribute(gCol, UCOL_FRENCH_COLLATION, UCOL_OFF, &status);
1476 ucol_setAttribute(gCol, UCOL_CASE_FIRST, UCOL_LOWER_FIRST, &status);
1479 ucol_setAttribute(gCol, UCOL_CASE_FIRST, UCOL_UPPER_FIRST, &status);
1482 ucol_setAttribute(gCol, UCOL_CASE_LEVEL, UCOL_ON, &status);
1485 ucol_setAttribute(gCol, UCOL_ALTERNATE_HANDLING, UCOL_SHIFTED, &status);
1487 if (opt_level != 0) {
1488 switch (opt_level) {
1490 ucol_setAttribute(gCol, UCOL_STRENGTH, UCOL_PRIMARY, &status);
1493 ucol_setAttribute(gCol, UCOL_STRENGTH, UCOL_SECONDARY, &status);
1496 ucol_setAttribute(gCol, UCOL_STRENGTH, UCOL_TERTIARY, &status);
1499 ucol_setAttribute(gCol, UCOL_STRENGTH, UCOL_QUATERNARY, &status);
1502 ucol_setAttribute(gCol, UCOL_STRENGTH, UCOL_IDENTICAL, &status);
1505 fprintf(stderr, "-level param must be between 1 and 5\n");
1510 if (U_FAILURE(status)) {
1511 fprintf(stderr, "Collator attribute setting failed.: %d\n", status);
1517 // Set up a Windows LCID
1519 if (opt_langid != 0) {
1520 gWinLCID = MAKELCID(opt_langid, SORT_DEFAULT);
1523 gWinLCID = uloc_getLCID(opt_locale);
1528 // Set the UNIX locale
1531 if (setlocale(LC_ALL, opt_locale) == 0) {
1532 fprintf(stderr, "setlocale(LC_ALL, %s) failed.\n", opt_locale);
1537 // Read in the input file.
1538 // File assumed to be utf-16.
1539 // Lines go onto heap buffers. Global index array to line starts is created.
1540 // Lines themselves are null terminated.
1543 UCharFile f(opt_fName);
1548 const int MAXLINES = 100000;
1549 gFileLines = new Line[MAXLINES];
1553 // Read the file, split into lines, and save in memory.
1554 // Loop runs once per utf-16 value from the input file,
1555 // (The number of bytes read from file per loop iteration depends on external encoding.)
1564 // We now have a good UTF-16 value in c.
1566 // Watch for CR, LF, EOF; these finish off a line.
1571 if (f.eof() || c == 0x0a || c==0x2028) { // Unipad inserts 2028 line separators!
1574 gFileLines[gNumFileLines].name = new UChar[column];
1575 gFileLines[gNumFileLines].len = column-1;
1576 memcpy(gFileLines[gNumFileLines].name, buf, column * sizeof(UChar));
1579 if (gNumFileLines >= MAXLINES) {
1580 fprintf(stderr, "File too big. Max number of lines is %d\n", MAXLINES);
1585 if (c == 0xa || c == 0x2028)
1593 static UBool warnFlag = TRUE;
1595 fprintf(stderr, "Warning - file line longer than 1023 chars truncated.\n");
1602 if (opt_terse == FALSE) {
1603 printf("file \"%s\", %d lines.\n", opt_fName, gNumFileLines);
1607 // Convert the lines to the UNIX encoding.
1613 // Pre-compute ICU sort keys for the lines of the file.
1618 for (line=0; line<gNumFileLines; line++) {
1619 t = ucol_getSortKey(gCol, gFileLines[line].name, -1, (unsigned char *)buf, sizeof(buf));
1620 gFileLines[line].icuSortKey = new char[t];
1622 if (t > (int32_t)sizeof(buf)) {
1623 t = ucol_getSortKey(gCol, gFileLines[line].name, -1, (unsigned char *)gFileLines[line].icuSortKey , t);
1627 memcpy(gFileLines[line].icuSortKey, buf, t);
1634 // Pre-compute Windows sort keys for the lines of the file.
1636 for (line=0; line<gNumFileLines; line++) {
1637 t=LCMapStringW(gWinLCID, LCMAP_SORTKEY, gFileLines[line].name, -1, buf, sizeof(buf));
1638 gFileLines[line].winSortKey = new char[t];
1639 if (t > (int32_t)sizeof(buf)) {
1640 t = LCMapStringW(gWinLCID, LCMAP_SORTKEY, gFileLines[line].name, -1, (unsigned short *)(gFileLines[line].winSortKey), t);
1644 memcpy(gFileLines[line].winSortKey, buf, t);
1649 // Pre-compute UNIX sort keys for the lines of the file.
1652 for (line=0; line<gNumFileLines; line++) {
1653 t=strxfrm((char *)buf, gFileLines[line].unixName, sizeof(buf));
1654 gFileLines[line].unixSortKey = new char[t];
1655 if (t > (int32_t)sizeof(buf)) {
1656 t = strxfrm(gFileLines[line].unixSortKey, gFileLines[line].unixName, sizeof(buf));
1660 memcpy(gFileLines[line].unixSortKey, buf, t);
1667 // Dump file lines, CEs, Sort Keys if requested.
1671 for (line=0; line<gNumFileLines; line++) {
1673 UChar c = gFileLines[line].name[i];
1676 if (c < 0x20 || c > 0x7e) {
1677 printf("\\u%.4x", c);
1686 UCollationElements *CEiter = ucol_openElements(gCol, gFileLines[line].name, -1, &status);
1690 ce = ucol_next(CEiter, &status);
1691 if (ce == UCOL_NULLORDER) {
1694 printf(" %.8x", ce);
1701 ucol_closeElements(CEiter);
1704 printf(" ICU Sort Key: ");
1706 unsigned char c = gFileLines[line].icuSortKey[i];
1711 if (i > 0 && i % 20 == 0) {
1721 // Pre-sort the lines.
1724 gSortedLines = new Line *[gNumFileLines];
1725 for (i=0; i<gNumFileLines; i++) {
1726 gSortedLines[i] = &gFileLines[i];
1730 qsort(gSortedLines, gNumFileLines, sizeof(Line *), Winstrcmp);
1732 else if (opt_unix) {
1733 qsort(gSortedLines, gNumFileLines, sizeof(Line *), UNIXstrcmp);
1737 qsort(gSortedLines, gNumFileLines, sizeof(Line *), ICUstrcmp);
1742 // Make up a randomized order, will be used for sorting tests.
1744 gRandomLines = new Line *[gNumFileLines];
1745 for (i=0; i<gNumFileLines; i++) {
1746 gRandomLines[i] = &gFileLines[i];
1748 qsort(gRandomLines, gNumFileLines, sizeof(Line *), ICURandomCmp);
1754 // We've got the file read into memory. Go do something with it.
1757 if (opt_qsort) doQSort();
1758 if (opt_binsearch) doBinarySearch();
1759 if (opt_keygen) doKeyGen();
1760 if (opt_keyhist) doKeyHist();
1761 if (opt_itertest) doIterTest();