X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=lib%2Fzipf.c;h=681df70069ac27419b4545dcbac0ee2de81fdbee;hp=21aaa36b1910655ccda093053a03470988736a8d;hb=8c5e96a4ee7004b2982b5d5247ecea1cb96bccb7;hpb=1442ba18b74e353ecbdf828b43f965648de2d546

diff --git a/lib/zipf.c b/lib/zipf.c
index 21aaa36b..681df700 100644
--- a/lib/zipf.c
+++ b/lib/zipf.c
@@ -10,101 +10,44 @@
 #include "zipf.h"
 #include "../minmax.h"
 #include "../hash.h"
-#include "../os/os.h"
 
-struct fio_zipf_disk {
-	uint64_t ver_magic;
-	uint64_t nranges;
-	uint64_t zetan;
-};
-
-#define FIO_ZIPF_DISK_MAGIC	0x7a697066
-#define FIO_ZIPF_DISK_VER	1
-#define FIO_ZIPF_MAGIC		((FIO_ZIPF_DISK_MAGIC << 16) | FIO_ZIPF_DISK_VER)
-
-static void write_zipf(struct zipf_state *zs)
-{
-	struct fio_zipf_disk f;
-	char tmp[80];
-	int fd;
-
-	sprintf(tmp, "fio.zipf.%f.%llu", zs->theta, (unsigned long long) zs->nranges);
-	fd = open(tmp, O_CREAT | O_WRONLY, 0644);
-	if (fd == -1)
-		return;
-
-	f.ver_magic = __cpu_to_le64(FIO_ZIPF_MAGIC);
-	f.nranges = __cpu_to_le64(zs->nranges);
-	f.zetan = __cpu_to_le64(fio_double_to_uint64(zs->zetan));
-	if (write(fd, &f, sizeof(f)) != sizeof(f))
-		unlink(tmp);
-
-	close(fd);
-}
+#define ZIPF_MAX_GEN	10000000UL
 
 static void zipf_update(struct zipf_state *zs)
 {
+	unsigned long to_gen;
 	unsigned int i;
 
-	log_info("fio: generating zetan for theta=%f, ranges=%lu\n", zs->theta, zs->nranges);
+	/*
+	 * It can become very costly to generate long sequences. Just cap it at
+	 * 10M max, that should be doable in 1-2s on even slow machines.
+	 * Precision will take a slight hit, but nothing major.
+	 */
+	to_gen = min(zs->nranges, (uint64_t) ZIPF_MAX_GEN);
 
-	for (i = 0; i < zs->nranges; i++)
+	for (i = 0; i < to_gen; i++)
 		zs->zetan += pow(1.0 / (double) (i + 1), zs->theta);
-
-	write_zipf(zs);
-}
-
-static void zipf_load_gen_zeta(struct zipf_state *zs)
-{
-	struct fio_zipf_disk f;
-	char tmp[80];
-	int fd;
-
-	sprintf(tmp, "fio.zipf.%f.%llu", zs->theta, (unsigned long long) zs->nranges);
-	fd = open(tmp, O_RDONLY);
-	if (fd == -1) {
-punt:
-		zipf_update(zs);
-		return;
-	}
-
-	if (read(fd, &f, sizeof(f)) != sizeof(f)) {
-		close(fd);
-		goto punt;
-	}
-
-	close(fd);
-
-	f.ver_magic = le64_to_cpu(f.ver_magic);
-	f.nranges = le64_to_cpu(f.nranges);
-	f.zetan = le64_to_cpu(f.zetan);
-
-	if (f.ver_magic != FIO_ZIPF_MAGIC) {
-		unlink(tmp);
-		goto punt;
-	}
-
-	zs->zetan = fio_uint64_to_double(f.zetan);
-	zs->nranges = f.nranges;
 }
 
-static void shared_rand_init(struct zipf_state *zs, unsigned long nranges)
+static void shared_rand_init(struct zipf_state *zs, unsigned long nranges,
+			     unsigned int seed)
 {
 	memset(zs, 0, sizeof(*zs));
 	zs->nranges = nranges;
 
-	init_rand(&zs->rand);
+	init_rand_seed(&zs->rand, seed, 0);
 	zs->rand_off = __rand(&zs->rand);
 }
 
-void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta)
+void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta,
+	       unsigned int seed)
 {
-	shared_rand_init(zs, nranges);
+	shared_rand_init(zs, nranges, seed);
 
 	zs->theta = theta;
 	zs->zeta2 = pow(1.0, zs->theta) + pow(0.5, zs->theta);
 
-	zipf_load_gen_zeta(zs);
+	zipf_update(zs);
 }
 
 unsigned long long zipf_next(struct zipf_state *zs)
@@ -116,7 +59,7 @@ unsigned long long zipf_next(struct zipf_state *zs)
 	alpha = 1.0 / (1.0 - zs->theta);
 	eta = (1.0 - pow(2.0 / n, 1.0 - zs->theta)) / (1.0 - zs->zeta2 / zs->zetan);
 
-	rand_uni = (double) __rand(&zs->rand) / (double) FRAND_MAX;
+	rand_uni = (double) __rand(&zs->rand) / (double) FRAND32_MAX;
 	rand_z = rand_uni * zs->zetan;
 
 	if (rand_z < 1.0)
@@ -126,19 +69,35 @@ unsigned long long zipf_next(struct zipf_state *zs)
 	else
 		val = 1 + (unsigned long long)(n * pow(eta*rand_uni - eta + 1.0, alpha));
 
-	return (__hash_long(val - 1) + zs->rand_off) % zs->nranges;
+	val--;
+
+	if (!zs->disable_hash)
+		val = __hash_u64(val);
+
+	return (val + zs->rand_off) % zs->nranges;
 }
 
-void pareto_init(struct zipf_state *zs, unsigned long nranges, double h)
+void pareto_init(struct zipf_state *zs, unsigned long nranges, double h,
+		 unsigned int seed)
 {
-	shared_rand_init(zs, nranges);
+	shared_rand_init(zs, nranges, seed);
 	zs->pareto_pow = log(h) / log(1.0 - h);
 }
 
 unsigned long long pareto_next(struct zipf_state *zs)
 {
-	double rand = (double) __rand(&zs->rand) / (double) FRAND_MAX;
-	unsigned long long n = zs->nranges - 1;
+	double rand = (double) __rand(&zs->rand) / (double) FRAND32_MAX;
+	unsigned long long n;
+
+	n = (zs->nranges - 1) * pow(rand, zs->pareto_pow);
+
+	if (!zs->disable_hash)
+		n = __hash_u64(n);
 
-	return (__hash_long(n * pow(rand, zs->pareto_pow)) + zs->rand_off) % zs->nranges;
+	return (n + zs->rand_off)  % zs->nranges;
+}
+
+void zipf_disable_hash(struct zipf_state *zs)
+{
+	zs->disable_hash = true;
 }