[thirdparty/git.git] / t / helper / test-lazy-init-name-hash.c

#define USE_THE_INDEX_VARIABLE
#include "test-tool.h"
#include "cache.h"
#include "environment.h"
#include "parse-options.h"

static int single;
static int multi;
static int count = 1;
static int dump;
static int perf;
static int analyze;
static int analyze_step;

/*
 * Dump the contents of the "dir" and "name" hash tables to stdout.
 * If you sort the result, you can compare it with the other type
 * mode and verify that both single and multi produce the same set.
 */
static void dump_run(void)
{
	struct hashmap_iter iter_dir;
	struct hashmap_iter iter_cache;

	/* Stolen from name-hash.c */
	struct dir_entry {
		struct hashmap_entry ent;
		struct dir_entry *parent;
		int nr;
		unsigned int namelen;
		char name[FLEX_ARRAY];
	};

	struct dir_entry *dir;
	struct cache_entry *ce;

	repo_read_index(the_repository);
	if (single) {
		test_lazy_init_name_hash(&the_index, 0);
	} else {
		int nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
		if (!nr_threads_used)
			die("non-threaded code path used");
	}

	hashmap_for_each_entry(&the_index.dir_hash, &iter_dir, dir,
				ent /* member name */)
		printf("dir %08x %7d %s\n", dir->ent.hash, dir->nr, dir->name);

	hashmap_for_each_entry(&the_index.name_hash, &iter_cache, ce,
				ent /* member name */)
		printf("name %08x %s\n", ce->ent.hash, ce->name);

	discard_index(&the_index);
}

/*
 * Run the single or multi threaded version "count" times and
 * report on the time taken.
 */
static uint64_t time_runs(int try_threaded)
{
	uint64_t t0, t1, t2;
	uint64_t sum = 0;
	uint64_t avg;
	int nr_threads_used;
	int i;

	for (i = 0; i < count; i++) {
		t0 = getnanotime();
		repo_read_index(the_repository);
		t1 = getnanotime();
		nr_threads_used = test_lazy_init_name_hash(&the_index, try_threaded);
		t2 = getnanotime();

		sum += (t2 - t1);

		if (try_threaded && !nr_threads_used)
			die("non-threaded code path used");

		if (nr_threads_used)
			printf("%f %f %d multi %d\n",
				   ((double)(t1 - t0))/1000000000,
				   ((double)(t2 - t1))/1000000000,
				   the_index.cache_nr,
				   nr_threads_used);
		else
			printf("%f %f %d single\n",
				   ((double)(t1 - t0))/1000000000,
				   ((double)(t2 - t1))/1000000000,
				   the_index.cache_nr);
		fflush(stdout);

		discard_index(&the_index);
	}

	avg = sum / count;
	if (count > 1)
		printf("avg %f %s\n",
			   (double)avg/1000000000,
			   (try_threaded) ? "multi" : "single");

	return avg;
}

/*
 * Try a series of runs varying the "istate->cache_nr" and
 * try to find a good value for the multi-threaded criteria.
 */
static void analyze_run(void)
{
	uint64_t t1s, t1m, t2s, t2m;
	int cache_nr_limit;
	int nr_threads_used = 0;
	int i;
	int nr;

	repo_read_index(the_repository);
	cache_nr_limit = the_index.cache_nr;
	discard_index(&the_index);

	nr = analyze;
	while (1) {
		uint64_t sum_single = 0;
		uint64_t sum_multi = 0;
		uint64_t avg_single;
		uint64_t avg_multi;

		if (nr > cache_nr_limit)
			nr = cache_nr_limit;

		for (i = 0; i < count; i++) {
			repo_read_index(the_repository);
			the_index.cache_nr = nr; /* cheap truncate of index */
			t1s = getnanotime();
			test_lazy_init_name_hash(&the_index, 0);
			t2s = getnanotime();
			sum_single += (t2s - t1s);
			the_index.cache_nr = cache_nr_limit;
			discard_index(&the_index);

			repo_read_index(the_repository);
			the_index.cache_nr = nr; /* cheap truncate of index */
			t1m = getnanotime();
			nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
			t2m = getnanotime();
			sum_multi += (t2m - t1m);
			the_index.cache_nr = cache_nr_limit;
			discard_index(&the_index);

			if (!nr_threads_used)
				printf("    [size %8d] [single %f]   non-threaded code path used\n",
					   nr, ((double)(t2s - t1s))/1000000000);
			else
				printf("    [size %8d] [single %f] %c [multi %f %d]\n",
					   nr,
					   ((double)(t2s - t1s))/1000000000,
					   (((t2s - t1s) < (t2m - t1m)) ? '<' : '>'),
					   ((double)(t2m - t1m))/1000000000,
					   nr_threads_used);
			fflush(stdout);
		}
		if (count > 1) {
			avg_single = sum_single / count;
			avg_multi = sum_multi / count;
			if (!nr_threads_used)
				printf("avg [size %8d] [single %f]\n",
					   nr,
					   (double)avg_single/1000000000);
			else
				printf("avg [size %8d] [single %f] %c [multi %f %d]\n",
					   nr,
					   (double)avg_single/1000000000,
					   (avg_single < avg_multi ? '<' : '>'),
					   (double)avg_multi/1000000000,
					   nr_threads_used);
			fflush(stdout);
		}

		if (nr >= cache_nr_limit)
			return;
		nr += analyze_step;
	}
}

int cmd__lazy_init_name_hash(int argc, const char **argv)
{
	const char *usage[] = {
		"test-tool lazy-init-name-hash -d (-s | -m)",
		"test-tool lazy-init-name-hash -p [-c c]",
		"test-tool lazy-init-name-hash -a a [--step s] [-c c]",
		"test-tool lazy-init-name-hash (-s | -m) [-c c]",
		"test-tool lazy-init-name-hash -s -m [-c c]",
		NULL
	};
	struct option options[] = {
		OPT_BOOL('s', "single", &single, "run single-threaded code"),
		OPT_BOOL('m', "multi", &multi, "run multi-threaded code"),
		OPT_INTEGER('c', "count", &count, "number of passes"),
		OPT_BOOL('d', "dump", &dump, "dump hash tables"),
		OPT_BOOL('p', "perf", &perf, "compare single vs multi"),
		OPT_INTEGER('a', "analyze", &analyze, "analyze different multi sizes"),
		OPT_INTEGER(0, "step", &analyze_step, "analyze step factor"),
		OPT_END(),
	};
	const char *prefix;
	uint64_t avg_single, avg_multi;

	prefix = setup_git_directory();

	argc = parse_options(argc, argv, prefix, options, usage, 0);

	/*
	 * istate->dir_hash is only created when ignore_case is set.
	 */
	ignore_case = 1;

	if (dump) {
		if (perf || analyze > 0)
			die("cannot combine dump, perf, or analyze");
		if (count > 1)
			die("count not valid with dump");
		if (single && multi)
			die("cannot use both single and multi with dump");
		if (!single && !multi)
			die("dump requires either single or multi");
		dump_run();
		return 0;
	}

	if (perf) {
		if (analyze > 0)
			die("cannot combine dump, perf, or analyze");
		if (single || multi)
			die("cannot use single or multi with perf");
		avg_single = time_runs(0);
		avg_multi = time_runs(1);
		if (avg_multi > avg_single)
			die("multi is slower");
		return 0;
	}

	if (analyze) {
		if (analyze < 500)
			die("analyze must be at least 500");
		if (!analyze_step)
			analyze_step = analyze;
		if (single || multi)
			die("cannot use single or multi with analyze");
		analyze_run();
		return 0;
	}

	if (!single && !multi)
		die("require either -s or -m or both");

	if (single)
		time_runs(0);
	if (multi)
		time_runs(1);

	return 0;
}
Commit	Line	Data
bdafeae0	1	#define USE_THE_INDEX_VARIABLE
64eb82fe	2	#include "test-tool.h"
ea194895	3	#include "cache.h"
32a8f510	4	#include "environment.h"
ea194895 JH	5	#include "parse-options.h"
	6
	7	static int single;
	8	static int multi;
	9	static int count = 1;
	10	static int dump;
	11	static int perf;
	12	static int analyze;
	13	static int analyze_step;
	14
	15	/*
	16	* Dump the contents of the "dir" and "name" hash tables to stdout.
	17	* If you sort the result, you can compare it with the other type
	18	* mode and verify that both single and multi produce the same set.
	19	*/
	20	static void dump_run(void)
	21	{
	22	struct hashmap_iter iter_dir;
	23	struct hashmap_iter iter_cache;
	24
	25	/* Stolen from name-hash.c */
	26	struct dir_entry {
	27	struct hashmap_entry ent;
	28	struct dir_entry *parent;
	29	int nr;
	30	unsigned int namelen;
	31	char name[FLEX_ARRAY];
	32	};
	33
	34	struct dir_entry *dir;
	35	struct cache_entry *ce;
	36
0ea414a1	37	repo_read_index(the_repository);
ea194895 JH	38	if (single) {
	39	test_lazy_init_name_hash(&the_index, 0);
	40	} else {
	41	int nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
	42	if (!nr_threads_used)
	43	die("non-threaded code path used");
	44	}
	45
87571c3f	46	hashmap_for_each_entry(&the_index.dir_hash, &iter_dir, dir,
23dee69f	47	ent /* member name */)
ea194895	48	printf("dir %08x %7d %s\n", dir->ent.hash, dir->nr, dir->name);
ea194895	49
87571c3f	50	hashmap_for_each_entry(&the_index.name_hash, &iter_cache, ce,
23dee69f	51	ent /* member name */)
ea194895	52	printf("name %08x %s\n", ce->ent.hash, ce->name);
ea194895	53
0ea414a1	54	discard_index(&the_index);
ea194895 JH	55	}
	56
	57	/*
	58	* Run the single or multi threaded version "count" times and
	59	* report on the time taken.
	60	*/
	61	static uint64_t time_runs(int try_threaded)
	62	{
	63	uint64_t t0, t1, t2;
	64	uint64_t sum = 0;
	65	uint64_t avg;
	66	int nr_threads_used;
	67	int i;
	68
	69	for (i = 0; i < count; i++) {
	70	t0 = getnanotime();
0ea414a1	71	repo_read_index(the_repository);
ea194895 JH	72	t1 = getnanotime();
	73	nr_threads_used = test_lazy_init_name_hash(&the_index, try_threaded);
	74	t2 = getnanotime();
	75
	76	sum += (t2 - t1);
	77
	78	if (try_threaded && !nr_threads_used)
	79	die("non-threaded code path used");
	80
	81	if (nr_threads_used)
	82	printf("%f %f %d multi %d\n",
	83	((double)(t1 - t0))/1000000000,
	84	((double)(t2 - t1))/1000000000,
	85	the_index.cache_nr,
	86	nr_threads_used);
	87	else
	88	printf("%f %f %d single\n",
	89	((double)(t1 - t0))/1000000000,
	90	((double)(t2 - t1))/1000000000,
	91	the_index.cache_nr);
	92	fflush(stdout);
	93
0ea414a1	94	discard_index(&the_index);
ea194895 JH	95	}
	96
	97	avg = sum / count;
	98	if (count > 1)
	99	printf("avg %f %s\n",
	100	(double)avg/1000000000,
	101	(try_threaded) ? "multi" : "single");
	102
	103	return avg;
	104	}
	105
	106	/*
	107	* Try a series of runs varying the "istate->cache_nr" and
	108	* try to find a good value for the multi-threaded criteria.
	109	*/
	110	static void analyze_run(void)
	111	{
	112	uint64_t t1s, t1m, t2s, t2m;
	113	int cache_nr_limit;
74dea0e1	114	int nr_threads_used = 0;
ea194895 JH	115	int i;
	116	int nr;
	117
0ea414a1	118	repo_read_index(the_repository);
ea194895	119	cache_nr_limit = the_index.cache_nr;
0ea414a1	120	discard_index(&the_index);
ea194895 JH	121
	122	nr = analyze;
	123	while (1) {
	124	uint64_t sum_single = 0;
	125	uint64_t sum_multi = 0;
	126	uint64_t avg_single;
	127	uint64_t avg_multi;
	128
	129	if (nr > cache_nr_limit)
	130	nr = cache_nr_limit;
	131
	132	for (i = 0; i < count; i++) {
0ea414a1	133	repo_read_index(the_repository);
ea194895 JH	134	the_index.cache_nr = nr; /* cheap truncate of index */
	135	t1s = getnanotime();
	136	test_lazy_init_name_hash(&the_index, 0);
	137	t2s = getnanotime();
	138	sum_single += (t2s - t1s);
	139	the_index.cache_nr = cache_nr_limit;
0ea414a1	140	discard_index(&the_index);
ea194895	141
0ea414a1	142	repo_read_index(the_repository);
ea194895 JH	143	the_index.cache_nr = nr; /* cheap truncate of index */
	144	t1m = getnanotime();
	145	nr_threads_used = test_lazy_init_name_hash(&the_index, 1);
	146	t2m = getnanotime();
	147	sum_multi += (t2m - t1m);
	148	the_index.cache_nr = cache_nr_limit;
0ea414a1	149	discard_index(&the_index);
ea194895 JH	150
	151	if (!nr_threads_used)
	152	printf(" [size %8d] [single %f] non-threaded code path used\n",
	153	nr, ((double)(t2s - t1s))/1000000000);
	154	else
	155	printf(" [size %8d] [single %f] %c [multi %f %d]\n",
	156	nr,
	157	((double)(t2s - t1s))/1000000000,
	158	(((t2s - t1s) < (t2m - t1m)) ? '<' : '>'),
	159	((double)(t2m - t1m))/1000000000,
	160	nr_threads_used);
	161	fflush(stdout);
	162	}
	163	if (count > 1) {
	164	avg_single = sum_single / count;
	165	avg_multi = sum_multi / count;
	166	if (!nr_threads_used)
	167	printf("avg [size %8d] [single %f]\n",
	168	nr,
	169	(double)avg_single/1000000000);
	170	else
	171	printf("avg [size %8d] [single %f] %c [multi %f %d]\n",
	172	nr,
	173	(double)avg_single/1000000000,
	174	(avg_single < avg_multi ? '<' : '>'),
	175	(double)avg_multi/1000000000,
	176	nr_threads_used);
	177	fflush(stdout);
	178	}
	179
	180	if (nr >= cache_nr_limit)
	181	return;
	182	nr += analyze_step;
	183	}
	184	}
	185
64eb82fe	186	int cmd__lazy_init_name_hash(int argc, const char **argv)
ea194895 JH	187	{
ea194895 JH	188	const char *usage[] = {
64eb82fe NTND	189	"test-tool lazy-init-name-hash -d (-s \| -m)",
	190	"test-tool lazy-init-name-hash -p [-c c]",
	191	"test-tool lazy-init-name-hash -a a [--step s] [-c c]",
	192	"test-tool lazy-init-name-hash (-s \| -m) [-c c]",
	193	"test-tool lazy-init-name-hash -s -m [-c c]",
ea194895 JH	194	NULL
	195	};
	196	struct option options[] = {
	197	OPT_BOOL('s', "single", &single, "run single-threaded code"),
	198	OPT_BOOL('m', "multi", &multi, "run multi-threaded code"),
	199	OPT_INTEGER('c', "count", &count, "number of passes"),
	200	OPT_BOOL('d', "dump", &dump, "dump hash tables"),
	201	OPT_BOOL('p', "perf", &perf, "compare single vs multi"),
	202	OPT_INTEGER('a', "analyze", &analyze, "analyze different multi sizes"),
	203	OPT_INTEGER(0, "step", &analyze_step, "analyze step factor"),
	204	OPT_END(),
	205	};
	206	const char *prefix;
	207	uint64_t avg_single, avg_multi;
	208
	209	prefix = setup_git_directory();
	210
	211	argc = parse_options(argc, argv, prefix, options, usage, 0);
	212
	213	/*
	214	* istate->dir_hash is only created when ignore_case is set.
	215	*/
	216	ignore_case = 1;
	217
	218	if (dump) {
	219	if (perf \|\| analyze > 0)
	220	die("cannot combine dump, perf, or analyze");
	221	if (count > 1)
	222	die("count not valid with dump");
	223	if (single && multi)
	224	die("cannot use both single and multi with dump");
	225	if (!single && !multi)
	226	die("dump requires either single or multi");
	227	dump_run();
	228	return 0;
	229	}
	230
	231	if (perf) {
	232	if (analyze > 0)
	233	die("cannot combine dump, perf, or analyze");
	234	if (single \|\| multi)
	235	die("cannot use single or multi with perf");
	236	avg_single = time_runs(0);
	237	avg_multi = time_runs(1);
	238	if (avg_multi > avg_single)
	239	die("multi is slower");
	240	return 0;
	241	}
	242
	243	if (analyze) {
	244	if (analyze < 500)
	245	die("analyze must be at least 500");
	246	if (!analyze_step)
	247	analyze_step = analyze;
	248	if (single \|\| multi)
	249	die("cannot use single or multi with analyze");
	250	analyze_run();
	251	return 0;
	252	}
	253
	254	if (!single && !multi)
	255	die("require either -s or -m or both");
	256
	257	if (single)
258	time_runs(0);
259	if (multi)
260	time_runs(1);
261
262	return 0;
263	}