t/helper/test-lazy-init-name-hash.c

   1 #define USE_THE_INDEX_VARIABLE
   2 #include "test-tool.h"
   3 #include "cache.h"
   4 #include "environment.h"
   5 #include "parse-options.h"
   6 #include "repository.h"
   7 #include "setup.h"
   8 #include "trace.h"
   9
  10 static int single;
  11 static int multi;
  12 static int count = 1;
  13 static int dump;
  14 static int perf;
  15 static int analyze;
  16 static int analyze_step;
  17
  18 /*
  19  * Dump the contents of the "dir" and "name" hash tables to stdout.
  20  * If you sort the result, you can compare it with the other type
  21  * mode and verify that both single and multi produce the same set.
  22  */
  23 static void dump_run(void)
  24 {
  25         struct hashmap_iter iter_dir;
  26         struct hashmap_iter iter_cache;
  27
  28         /* Stolen from name-hash.c */
  29         struct dir_entry {
  30                 struct hashmap_entry ent;
  31                 struct dir_entry *parent;
  32                 int nr;
  33                 unsigned int namelen;
  34                 char name[FLEX_ARRAY];
  35         };
  36
  37         struct dir_entry *dir;
  38         struct cache_entry *ce;
  39
  40         repo_read_index(the_repository);
  41         if (single) {
  42                 test_lazy_init_name_hash(&the_index, 0);
  43         } else {
  44                 int nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
  45                 if (!nr_threads_used)
  46                         die("non-threaded code path used");
  47         }
  48
  49         hashmap_for_each_entry(&the_index.dir_hash, &iter_dir, dir,
  50                                 ent /* member name */)
  51                 printf("dir %08x %7d %s\n", dir->ent.hash, dir->nr, dir->name);
  52
  53         hashmap_for_each_entry(&the_index.name_hash, &iter_cache, ce,
  54                                 ent /* member name */)
  55                 printf("name %08x %s\n", ce->ent.hash, ce->name);
  56
  57         discard_index(&the_index);
  58 }
  59
  60 /*
  61  * Run the single or multi threaded version "count" times and
  62  * report on the time taken.
  63  */
  64 static uint64_t time_runs(int try_threaded)
  65 {
  66         uint64_t t0, t1, t2;
  67         uint64_t sum = 0;
  68         uint64_t avg;
  69         int nr_threads_used;
  70         int i;
  71
  72         for (i = 0; i < count; i++) {
  73                 t0 = getnanotime();
  74                 repo_read_index(the_repository);
  75                 t1 = getnanotime();
  76                 nr_threads_used = test_lazy_init_name_hash(&the_index, try_threaded);
  77                 t2 = getnanotime();
  78
  79                 sum += (t2 - t1);
  80
  81                 if (try_threaded && !nr_threads_used)
  82                         die("non-threaded code path used");
  83
  84                 if (nr_threads_used)
  85                         printf("%f %f %d multi %d\n",
  86                                    ((double)(t1 - t0))/1000000000,
  87                                    ((double)(t2 - t1))/1000000000,
  88                                    the_index.cache_nr,
  89                                    nr_threads_used);
  90                 else
  91                         printf("%f %f %d single\n",
  92                                    ((double)(t1 - t0))/1000000000,
  93                                    ((double)(t2 - t1))/1000000000,
  94                                    the_index.cache_nr);
  95                 fflush(stdout);
  96
  97                 discard_index(&the_index);
  98         }
  99
 100         avg = sum / count;
 101         if (count > 1)
 102                 printf("avg %f %s\n",
 103                            (double)avg/1000000000,
 104                            (try_threaded) ? "multi" : "single");
 105
 106         return avg;
 107 }
 108
 109 /*
 110  * Try a series of runs varying the "istate->cache_nr" and
 111  * try to find a good value for the multi-threaded criteria.
 112  */
 113 static void analyze_run(void)
 114 {
 115         uint64_t t1s, t1m, t2s, t2m;
 116         int cache_nr_limit;
 117         int nr_threads_used = 0;
 118         int i;
 119         int nr;
 120
 121         repo_read_index(the_repository);
 122         cache_nr_limit = the_index.cache_nr;
 123         discard_index(&the_index);
 124
 125         nr = analyze;
 126         while (1) {
 127                 uint64_t sum_single = 0;
 128                 uint64_t sum_multi = 0;
 129                 uint64_t avg_single;
 130                 uint64_t avg_multi;
 131
 132                 if (nr > cache_nr_limit)
 133                         nr = cache_nr_limit;
 134
 135                 for (i = 0; i < count; i++) {
 136                         repo_read_index(the_repository);
 137                         the_index.cache_nr = nr; /* cheap truncate of index */
 138                         t1s = getnanotime();
 139                         test_lazy_init_name_hash(&the_index, 0);
 140                         t2s = getnanotime();
 141                         sum_single += (t2s - t1s);
 142                         the_index.cache_nr = cache_nr_limit;
 143                         discard_index(&the_index);
 144
 145                         repo_read_index(the_repository);
 146                         the_index.cache_nr = nr; /* cheap truncate of index */
 147                         t1m = getnanotime();
 148                         nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
 149                         t2m = getnanotime();
 150                         sum_multi += (t2m - t1m);
 151                         the_index.cache_nr = cache_nr_limit;
 152                         discard_index(&the_index);
 153
 154                         if (!nr_threads_used)
 155                                 printf("    [size %8d] [single %f]   non-threaded code path used\n",
 156                                            nr, ((double)(t2s - t1s))/1000000000);
 157                         else
 158                                 printf("    [size %8d] [single %f] %c [multi %f %d]\n",
 159                                            nr,
 160                                            ((double)(t2s - t1s))/1000000000,
 161                                            (((t2s - t1s) < (t2m - t1m)) ? '<' : '>'),
 162                                            ((double)(t2m - t1m))/1000000000,
 163                                            nr_threads_used);
 164                         fflush(stdout);
 165                 }
 166                 if (count > 1) {
 167                         avg_single = sum_single / count;
 168                         avg_multi = sum_multi / count;
 169                         if (!nr_threads_used)
 170                                 printf("avg [size %8d] [single %f]\n",
 171                                            nr,
 172                                            (double)avg_single/1000000000);
 173                         else
 174                                 printf("avg [size %8d] [single %f] %c [multi %f %d]\n",
 175                                            nr,
 176                                            (double)avg_single/1000000000,
 177                                            (avg_single < avg_multi ? '<' : '>'),
 178                                            (double)avg_multi/1000000000,
 179                                            nr_threads_used);
 180                         fflush(stdout);
 181                 }
 182
 183                 if (nr >= cache_nr_limit)
 184                         return;
 185                 nr += analyze_step;
 186         }
 187 }
 188
 189 int cmd__lazy_init_name_hash(int argc, const char **argv)
 190 {
 191         const char *usage[] = {
 192                 "test-tool lazy-init-name-hash -d (-s | -m)",
 193                 "test-tool lazy-init-name-hash -p [-c c]",
 194                 "test-tool lazy-init-name-hash -a a [--step s] [-c c]",
 195                 "test-tool lazy-init-name-hash (-s | -m) [-c c]",
 196                 "test-tool lazy-init-name-hash -s -m [-c c]",
 197                 NULL
 198         };
 199         struct option options[] = {
 200                 OPT_BOOL('s', "single", &single, "run single-threaded code"),
 201                 OPT_BOOL('m', "multi", &multi, "run multi-threaded code"),
 202                 OPT_INTEGER('c', "count", &count, "number of passes"),
 203                 OPT_BOOL('d', "dump", &dump, "dump hash tables"),
 204                 OPT_BOOL('p', "perf", &perf, "compare single vs multi"),
 205                 OPT_INTEGER('a', "analyze", &analyze, "analyze different multi sizes"),
 206                 OPT_INTEGER(0, "step", &analyze_step, "analyze step factor"),
 207                 OPT_END(),
 208         };
 209         const char *prefix;
 210         uint64_t avg_single, avg_multi;
 211
 212         prefix = setup_git_directory();
 213
 214         argc = parse_options(argc, argv, prefix, options, usage, 0);
 215
 216         /*
 217          * istate->dir_hash is only created when ignore_case is set.
 218          */
 219         ignore_case = 1;
 220
 221         if (dump) {
 222                 if (perf || analyze > 0)
 223                         die("cannot combine dump, perf, or analyze");
 224                 if (count > 1)
 225                         die("count not valid with dump");
 226                 if (single && multi)
 227                         die("cannot use both single and multi with dump");
 228                 if (!single && !multi)
 229                         die("dump requires either single or multi");
 230                 dump_run();
 231                 return 0;
 232         }
 233
 234         if (perf) {
 235                 if (analyze > 0)
 236                         die("cannot combine dump, perf, or analyze");
 237                 if (single || multi)
 238                         die("cannot use single or multi with perf");
 239                 avg_single = time_runs(0);
 240                 avg_multi = time_runs(1);
 241                 if (avg_multi > avg_single)
 242                         die("multi is slower");
 243                 return 0;
 244         }
 245
 246         if (analyze) {
 247                 if (analyze < 500)
 248                         die("analyze must be at least 500");
 249                 if (!analyze_step)
 250                         analyze_step = analyze;
 251                 if (single || multi)
 252                         die("cannot use single or multi with analyze");
 253                 analyze_run();
 254                 return 0;
 255         }
 256
 257         if (!single && !multi)
 258                 die("require either -s or -m or both");
 259
 260         if (single)
 261                 time_runs(0);
 262         if (multi)
 263                 time_runs(1);
 264
 265         return 0;
 266 }