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

#include "test-tool.h"
#include "cache.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;

	read_cache();
	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_cache();
}

/*
 * 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();
		read_cache();
		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_cache();
	}

	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;

	read_cache();
	cache_nr_limit = the_index.cache_nr;
	discard_cache();

	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++) {
			read_cache();
			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_cache();

			read_cache();
			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_cache();

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