test-wildmatch: add "perf" command to compare wildmatch and fnmatch

It takes a text file, a pattern, a number <n> and pathname flag. Each line in the text file is matched against the pattern <n> times. If "pathname" is given, FNM_PATHNAME is used. test-wildmatch is built with -O2 and tested against glibc 2.14.1 (also -O2) and compat/fnmatch. The input file is linux-2.6.git file list. <n> is 2000. The complete command list is at the end. wildmatch is beaten in the following cases. Apparently it needs some improvement in FNM_PATHNAME case: glibc, '*/*/*' with FNM_PATHNAME: wildmatch 8s 1559us fnmatch 1s 11877us or 12.65% faster compat, '*/*/*' with FNM_PATHNAME: wildmatch 7s 922458us fnmatch 2s 905111us or 36.67% faster compat, '*/*/*' without FNM_PATHNAME: wildmatch 7s 264201us fnmatch 2s 1897us or 27.56% faster compat, '[a-z]*/[a-z]*/[a-z]*' with FNM_PATHNAME: wildmatch 8s 742827us fnmatch 0s 922943us or 10.56% faster compat, '[a-z]*/[a-z]*/[a-z]*' without FNM_PATHNAME: wildmatch 8s 284520us fnmatch 0s 6936us or 0.08% faster The rest of glibc numbers ------------------------- 'Documentation/*' wildmatch 1s 529479us fnmatch 1s 98263us or 71.81% slower 'drivers/*' wildmatch 1s 988288us fnmatch 1s 192049us or 59.95% slower 'Documentation/*' pathname wildmatch 1s 557507us fnmatch 1s 93696us or 70.22% slower 'drivers/*' pathname wildmatch 2s 161626us fnmatch 1s 230372us or 56.92% slower '[Dd]ocu[Mn]entation/*' wildmatch 1s 776581us fnmatch 1s 471693us or 82.84% slower '[Dd]o?u[Mn]en?ati?n/*' wildmatch 1s 770770us fnmatch 1s 555727us or 87.86% slower '[Dd]o?u[Mn]en?ati?n/*' pathname wildmatch 1s 783507us fnmatch 1s 537029us or 86.18% slower '[A-Za-z][A-Za-z]??*' wildmatch 4s 110386us fnmatch 4s 926306us or 119.85% slower '[A-Za-z][A-Za-z]??' wildmatch 3s 918114us fnmatch 3s 686175us or 94.08% slower '[A-Za-z][A-Za-z]??*' pathname wildmatch 4s 453746us fnmatch 4s 955856us or 111.27% slower '[A-Za-z][A-Za-z]??' pathname wildmatch 3s 896646us fnmatch 3s 733828us or 95.82% slower '*/*/*' wildmatch 7s 287985us fnmatch 1s 74083us or 14.74% slower '[a-z]*/[a-z]*/[a-z]*' pathname wildmatch 8s 796659us fnmatch 1s 568409us or 17.83% slower '[a-z]*/[a-z]*/[a-z]*' wildmatch 8s 316559us fnmatch 3s 430652us or 41.25% slower The rest of compat numbers -------------------------- 'Documentation/*' wildmatch 1s 520389us fnmatch 0s 62579us or 4.12% slower 'drivers/*' wildmatch 1s 955354us fnmatch 0s 190109us or 9.72% slower 'Documentation/*' pathname wildmatch 1s 561675us fnmatch 0s 55336us or 3.54% slower 'drivers/*' pathname wildmatch 2s 106100us fnmatch 0s 219680us or 10.43% slower '[Dd]ocu[Mn]entation/*' wildmatch 1s 750810us fnmatch 0s 542721us or 31.00% slower '[Dd]o?u[Mn]en?ati?n/*' wildmatch 1s 724791us fnmatch 0s 538948us or 31.25% slower '[Dd]o?u[Mn]en?ati?n/*' pathname wildmatch 1s 731403us fnmatch 0s 537474us or 31.04% slower '[A-Za-z][A-Za-z]??*' wildmatch 4s 28555us fnmatch 1s 67297us or 26.49% slower '[A-Za-z][A-Za-z]??' wildmatch 3s 838279us fnmatch 0s 880005us or 22.93% slower '[A-Za-z][A-Za-z]??*' pathname wildmatch 4s 379476us fnmatch 1s 55643us or 24.10% slower '[A-Za-z][A-Za-z]??' pathname wildmatch 3s 830910us fnmatch 0s 849699us or 22.18% slower The following commands are used: LANG=C ./test-wildmatch perf /tmp/filelist.txt 'Documentation/*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt 'drivers/*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt 'Documentation/*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt 'drivers/*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '[Dd]ocu[Mn]entation/*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt '[Dd]o?u[Mn]en?ati?n/*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt '[Dd]o?u[Mn]en?ati?n/*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '[A-Za-z][A-Za-z]??*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt '[A-Za-z][A-Za-z]??' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt '[A-Za-z][A-Za-z]??*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '[A-Za-z][A-Za-z]??' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '*/*/*' 2000 LANG=C ./test-wildmatch perf /tmp/filelist.txt '*/*/*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '[a-z]*/[a-z]*/[a-z]*' 2000 pathname LANG=C ./test-wildmatch perf /tmp/filelist.txt '[a-z]*/[a-z]*/[a-z]*' 2000 Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>

Nguyễn Thái Ngọc Duy committed Jan 1, 2013 at 09:44 UTC 1b25892636d2f250eb2163301440ee8c8c1ac944
1 file changed +73
test-wildmatch.c
+73
@@ -1,9 +1,82 @@
1 #include "cache.h"
2 #include "wildmatch.h"
3
4 +static int perf(int ac, char **av)
5 +{
6 + struct timeval tv1, tv2;
7 + struct stat st;
8 + int fd, i, n, flags1 = 0, flags2 = 0;
9 + char *buffer, *p;
10 + uint32_t usec1, usec2;
11 + const char *lang;
12 + const char *file = av[0];
13 + const char *pattern = av[1];
14 +
15 + lang = getenv("LANG");
16 + if (lang && strcmp(lang, "C"))
17 + die("Please test it on C locale.");
18 +
19 + if ((fd = open(file, O_RDONLY)) == -1 || fstat(fd, &st))
20 + die_errno("file open");
21 +
22 + buffer = xmalloc(st.st_size + 2);
23 + if (read(fd, buffer, st.st_size) != st.st_size)
24 + die_errno("read");
25 +
26 + buffer[st.st_size] = '\0';
27 + buffer[st.st_size + 1] = '\0';
28 + for (i = 0; i < st.st_size; i++)
29 + if (buffer[i] == '\n')
30 + buffer[i] = '\0';
31 +
32 + n = atoi(av[2]);
33 + if (av[3] && !strcmp(av[3], "pathname")) {
34 + flags1 = WM_PATHNAME;
35 + flags2 = FNM_PATHNAME;
36 + }
37 +
38 + gettimeofday(&tv1, NULL);
39 + for (i = 0; i < n; i++) {
40 + for (p = buffer; *p; p += strlen(p) + 1)
41 + wildmatch(pattern, p, flags1, NULL);
42 + }
43 + gettimeofday(&tv2, NULL);
44 +
45 + usec1 = (uint32_t)tv2.tv_sec * 1000000 + tv2.tv_usec;
46 + usec1 -= (uint32_t)tv1.tv_sec * 1000000 + tv1.tv_usec;
47 + printf("wildmatch %ds %dus\n",
48 + (int)(usec1 / 1000000),
49 + (int)(usec1 % 1000000));
50 +
51 + gettimeofday(&tv1, NULL);
52 + for (i = 0; i < n; i++) {
53 + for (p = buffer; *p; p += strlen(p) + 1)
54 + fnmatch(pattern, p, flags2);
55 + }
56 + gettimeofday(&tv2, NULL);
57 +
58 + usec2 = (uint32_t)tv2.tv_sec * 1000000 + tv2.tv_usec;
59 + usec2 -= (uint32_t)tv1.tv_sec * 1000000 + tv1.tv_usec;
60 + if (usec2 > usec1)
61 + printf("fnmatch %ds %dus or %.2f%% slower\n",
62 + (int)((usec2 - usec1) / 1000000),
63 + (int)((usec2 - usec1) % 1000000),
64 + (float)(usec2 - usec1) / usec1 * 100);
65 + else
66 + printf("fnmatch %ds %dus or %.2f%% faster\n",
67 + (int)((usec1 - usec2) / 1000000),
68 + (int)((usec1 - usec2) % 1000000),
69 + (float)(usec1 - usec2) / usec1 * 100);
70 + return 0;
71 +}
72 +
73 int main(int argc, char **argv)
74 {
75 int i;
76 +
77 + if (!strcmp(argv[1], "perf"))
78 + return perf(argc - 2, argv + 2);
79 +
80 for (i = 2; i < argc; i++) {
81 if (argv[i][0] == '/')
82 die("Forward slash is not allowed at the beginning of the\n"