-/*
+/*
* Written by Josh Dybnis and released to the public domain, as explained at
* http://creativecommons.org/licenses/publicdomain
- *
- * C implementation of Cliff Click's lock-free hash table from
+ *
+ * C implementation of Cliff Click's lock-free hash table from
* http://www.azulsystems.com/events/javaone_2008/2008_CodingNonBlock.pdf
* http://sourceforge.net/projects/high-scale-lib
*
- * Note: This is code uses synchronous atomic operations because that is all that x86 provides.
+ * Note: This is code uses synchronous atomic operations because that is all that x86 provides.
* Every atomic operation is also an implicit full memory barrier. The upshot is that it simplifies
- * the code a bit, but it won't be as fast as it could be on platforms that provide weaker
- * operations like and unfenced CAS which would still do the job.
+ * the code a bit, but it won't be as fast as it could be on platforms that provide weaker
+ * operations like unfenced CAS which would still do the job.
+ *
+ * 11FebO9 - Bug fix in ht_iter_next() from Rui Ueyama
*/
#include <stdio.h>
volatile entry_t *table;
hashtable_t *ht; // parent ht;
struct hti *next;
- unsigned scale;
- int max_probe;
+#ifdef USE_SYSTEM_MALLOC
+ void *unaligned_table_ptr; // system malloc doesn't guarentee cache-line alignment
+#endif
+ size_t count; // TODO: make these counters distributed
+ size_t key_count;
+ size_t copy_scan;
+ size_t num_entries_copied;
+ int probe;
int ref_count;
- int count; // TODO: make these counters distributed
- int num_entries_copied;
- int copy_scan;
+ uint8_t scale;
} hti_t;
struct ht_iter {
struct ht {
hti_t *hti;
const datatype_t *key_type;
+ uint32_t hti_copies;
+ double density;
+ int probe;
};
-static const map_val_t COPIED_VALUE = TAG_VALUE(DOES_NOT_EXIST, TAG1);
-static const map_val_t TOMBSTONE = STRIP_TAG(-1, TAG1);
+static const map_val_t COPIED_VALUE = TAG_VALUE(DOES_NOT_EXIST, TAG1);
+static const map_val_t TOMBSTONE = STRIP_TAG(-1, TAG1);
static const unsigned ENTRIES_PER_BUCKET = CACHE_LINE_SIZE/sizeof(entry_t);
static const unsigned ENTRIES_PER_COPY_CHUNK = CACHE_LINE_SIZE/sizeof(entry_t)*2;
static const unsigned MIN_SCALE = 4; // min 16 entries (4 buckets)
-static const unsigned MAX_BUCKETS_TO_PROBE = 250;
static int hti_copy_entry (hti_t *ht1, volatile entry_t *ent, uint32_t ent_key_hash, hti_t *ht2);
// Choose the next bucket to probe using the high-order bits of <key_hash>.
static inline int get_next_ndx(int old_ndx, uint32_t key_hash, int ht_scale) {
+#if 1
int incr = (key_hash >> (32 - ht_scale));
- incr += !incr; // If the increment is 0, make it 1.
+ if (incr < ENTRIES_PER_BUCKET) { incr += ENTRIES_PER_BUCKET; }
return (old_ndx + incr) & MASK(ht_scale);
+#else
+ return (old_ndx + ENTRIES_PER_BUCKET) & MASK(ht_scale);
+#endif
}
-// Lookup <key> in <hti>.
+// Lookup <key> in <hti>.
//
-// Return the entry that <key> is in, or if <key> isn't in <hti> return the entry that it would be
-// in if it were inserted into <hti>. If there is no room for <key> in <hti> then return NULL, to
+// Return the entry that <key> is in, or if <key> isn't in <hti> return the entry that it would be
+// in if it were inserted into <hti>. If there is no room for <key> in <hti> then return NULL, to
// indicate that the caller should look in <hti->next>.
//
-// Record if the entry being returned is empty. Otherwise the caller will have to waste time
+// Record if the entry being returned is empty. Otherwise the caller will have to waste time
// re-comparing the keys to confirm that it did not lose a race to fill an empty entry.
static volatile entry_t *hti_lookup (hti_t *hti, map_key_t key, uint32_t key_hash, int *is_empty) {
TRACE("h2", "hti_lookup(key %p in hti %p)", key, hti);
// Probe one cache line at a time
int ndx = key_hash & MASK(hti->scale); // the first entry to search
- for (int i = 0; i < hti->max_probe; ++i) {
+ for (int i = 0; i < hti->probe; ++i) {
// The start of the bucket is the first entry in the cache line.
- volatile entry_t *bucket = hti->table + (ndx & ~(ENTRIES_PER_BUCKET-1));
+ volatile entry_t *bucket = hti->table + (ndx & ~(ENTRIES_PER_BUCKET-1));
// Start searching at the indexed entry. Then loop around to the begining of the cache line.
for (int j = 0; j < ENTRIES_PER_BUCKET; ++j) {
map_key_t ent_key = ent->key;
if (ent_key == DOES_NOT_EXIST) {
- TRACE("h1", "hti_lookup: entry %p for key %p is empty", ent,
+ TRACE("h1", "hti_lookup: entry %p for key %p is empty", ent,
(hti->ht->key_type == NULL) ? (void *)key : GET_PTR(key));
*is_empty = 1; // indicate an empty so the caller avoids an expensive key compare
return ent;
}
- // Compare <key> with the key in the entry.
+ // Compare <key> with the key in the entry.
if (EXPECT_TRUE(hti->ht->key_type == NULL)) {
// fast path for integer keys
if (ent_key == key) {
static hti_t *hti_alloc (hashtable_t *parent, int scale) {
hti_t *hti = (hti_t *)nbd_malloc(sizeof(hti_t));
memset(hti, 0, sizeof(hti_t));
-
- size_t sz = sizeof(entry_t) * (1 << scale);
- entry_t *table = nbd_malloc(sz);
- memset(table, 0, sz);
- hti->table = table;
-
hti->scale = scale;
- // When searching for a key probe a maximum of 1/4 of the buckets up to 1000 buckets.
- hti->max_probe = ((1 << (hti->scale - 2)) / ENTRIES_PER_BUCKET) + 4;
- if (hti->max_probe > MAX_BUCKETS_TO_PROBE) {
- hti->max_probe = MAX_BUCKETS_TO_PROBE;
- }
+ size_t sz = sizeof(entry_t) * (1ULL << scale);
+#ifdef USE_SYSTEM_MALLOC
+ hti->unaligned_table_ptr = nbd_malloc(sz + CACHE_LINE_SIZE - 1);
+ hti->table = (void *)(((size_t)hti->unaligned_table_ptr + CACHE_LINE_SIZE - 1) & ~(CACHE_LINE_SIZE - 1));
+#else
+ hti->table = nbd_malloc(sz);
+#endif
+ memset((void *)hti->table, 0, sz);
+ hti->probe = (int)(hti->scale * 1.5) + 2;
+ int quarter = (1ULL << (hti->scale - 2)) / ENTRIES_PER_BUCKET;
+ if (hti->probe > quarter && quarter > 4) {
+ // When searching for a key probe a maximum of 1/4
+ hti->probe = quarter;
+ }
+ ASSERT(hti->probe);
hti->ht = parent;
hti->ref_count = 1; // one for the parent
// heuristics to determine the size of the new table
size_t count = ht_count(hti->ht);
unsigned int new_scale = hti->scale;
- new_scale += (count > (1 << (new_scale - 2))); // double size if more than 1/4 full
- new_scale += (count > (1 << (new_scale - 2))); // double size again if more than 1/2 full
+ new_scale += (count > (1ULL << (hti->scale - 1))) || (hti->key_count > (1ULL << (hti->scale - 2)) + (1ULL << (hti->scale - 3))); // double size if more than 1/2 full
// Allocate the new table and attempt to install it.
hti_t *next = hti_alloc(hti->ht, new_scale);
if (old_next != NULL) {
// Another thread beat us to it.
TRACE("h0", "hti_start_copy: lost race to install new hti; found %p", old_next, 0);
- nbd_free(next);
+#ifdef USE_SYSTEM_MALLOC
+ nbd_free(next->unaligned_table_ptr);
+#else
+ nbd_free((void *)next->table);
+#endif
return;
}
TRACE("h0", "hti_start_copy: new hti %p scale %llu", next, next->scale);
+ SYNC_ADD(&hti->ht->hti_copies, 1);
+ hti->ht->density = (double)hti->key_count / (1ULL << hti->scale) * 100;
+ hti->ht->probe = hti->probe;
}
-// Copy the key and value stored in <ht1_ent> (which must be an entry in <ht1>) to <ht2>.
+// Copy the key and value stored in <ht1_ent> (which must be an entry in <ht1>) to <ht2>.
//
// Return 1 unless <ht1_ent> is already copied (then return 0), so the caller can account for the total
// number of entries left to copy.
assert(ht1);
assert(ht1->next);
assert(ht2);
- assert(ht1_ent >= ht1->table && ht1_ent < ht1->table + (1 << ht1->scale));
+ assert(ht1_ent >= ht1->table && ht1_ent < ht1->table + (1ULL << ht1->scale));
#ifndef NBD32
assert(key_hash == 0 || ht1->ht->key_type == NULL || (key_hash >> 16) == (ht1_ent->key >> 48));
#endif
// The old table's dead entries don't need to be copied to the new table
if (ht1_ent_val == TOMBSTONE)
- return TRUE;
+ return TRUE;
// Install the key in the new table.
map_key_t ht1_ent_key = ht1_ent->key;
map_key_t key = (ht1->ht->key_type == NULL) ? (map_key_t)ht1_ent_key : (map_key_t)GET_PTR(ht1_ent_key);
// We use 0 to indicate that <key_hash> is uninitiallized. Occasionally the key's hash will really be 0 and we
- // waste time recomputing it every time. It is rare enough that it won't hurt performance.
- if (key_hash == 0) {
+ // waste time recomputing it every time. It is rare enough that it won't hurt performance.
+ if (key_hash == 0) {
#ifdef NBD32
key_hash = (ht1->ht->key_type == NULL) ? murmur32_4b(ht1_ent_key) : ht1->ht->key_type->hash((void *)key);
#else
ht1_ent_key, old_ht2_ent_key);
return hti_copy_entry(ht1, ht1_ent, key_hash, ht2); // recursive tail-call
}
+ SYNC_ADD(&ht2->key_count, 1);
}
// Copy the value to the entry in the new table.
// Update the count if we were the one that completed the copy.
if (old_ht2_ent_val == DOES_NOT_EXIST) {
TRACE("h0", "hti_copy_entry: key %p value %p copied to new entry", key, ht1_ent_val);
- SYNC_ADD(&ht1->count, -1);
- SYNC_ADD(&ht2->count, 1);
+ (void)SYNC_ADD(&ht1->count, -1);
+ (void)SYNC_ADD(&ht2->count, 1);
return TRUE;
}
- TRACE("h0", "hti_copy_entry: lost race to install value %p in new entry; found value %p",
+ TRACE("h0", "hti_copy_entry: lost race to install value %p in new entry; found value %p",
ht1_ent_val, old_ht2_ent_val);
return FALSE; // another thread completed the copy
}
-// Compare <expected> with the existing value associated with <key>. If the values match then
-// replace the existing value with <new>. If <new> is DOES_NOT_EXIST, delete the value associated with
+// Compare <expected> with the existing value associated with <key>. If the values match then
+// replace the existing value with <new>. If <new> is DOES_NOT_EXIST, delete the value associated with
// the key by replacing it with a TOMBSTONE.
//
// Return the previous value associated with <key>, or DOES_NOT_EXIST if <key> is not in the table
-// or associated with a TOMBSTONE. If a copy is in progress and <key> has been copied to the next
-// table then return COPIED_VALUE.
+// or associated with a TOMBSTONE. If a copy is in progress and <key> has been copied to the next
+// table then return COPIED_VALUE.
//
// NOTE: the returned value matches <expected> iff the set succeeds
//
-// Certain values of <expected> have special meaning. If <expected> is CAS_EXPECT_EXISTS then any
+// Certain values of <expected> have special meaning. If <expected> is CAS_EXPECT_EXISTS then any
// real value matches (i.ent. not a TOMBSTONE or DOES_NOT_EXIST) as long as <key> is in the table. If
// <expected> is CAS_EXPECT_WHATEVER then skip the test entirely.
//
return DOES_NOT_EXIST;
// Allocate <new_key>.
- map_key_t new_key = (hti->ht->key_type == NULL)
- ? (map_key_t)key
+ map_key_t new_key = (hti->ht->key_type == NULL)
+ ? (map_key_t)key
: (map_key_t)hti->ht->key_type->clone((void *)key);
#ifndef NBD32
if (EXPECT_FALSE(hti->ht->key_type != NULL)) {
- // Combine <new_key> pointer with bits from its hash
- new_key = ((uint64_t)(key_hash >> 16) << 48) | new_key;
+ // Combine <new_key> pointer with bits from its hash
+ new_key = ((uint64_t)(key_hash >> 16) << 48) | new_key;
}
#endif
// Retry if another thread stole the entry out from under us.
if (old_ent_key != DOES_NOT_EXIST) {
TRACE("h0", "hti_cas: lost race to install key %p in entry %p", new_key, ent);
- TRACE("h0", "hti_cas: found %p instead of NULL",
+ TRACE("h0", "hti_cas: found %p instead of NULL",
(hti->ht->key_type == NULL) ? (void *)old_ent_key : GET_PTR(old_ent_key), 0);
if (hti->ht->key_type != NULL) {
nbd_free(GET_PTR(new_key));
return hti_cas(hti, key, key_hash, expected, new); // tail-call
}
TRACE("h2", "hti_cas: installed key %p in entry %p", new_key, ent);
+ SYNC_ADD(&hti->key_count, 1);
}
- TRACE("h0", "hti_cas: entry for key %p is %p",
+ TRACE("h0", "hti_cas: entry for key %p is %p",
(hti->ht->key_type == NULL) ? (void *)ent->key : GET_PTR(ent->key), ent);
// If the entry is in the middle of a copy, the copy must be completed first.
map_val_t ent_val = ent->val;
if (EXPECT_FALSE(IS_TAGGED(ent_val, TAG1))) {
if (ent_val != COPIED_VALUE && ent_val != TAG_VALUE(TOMBSTONE, TAG1)) {
- int did_copy = hti_copy_entry(hti, ent, key_hash, ((volatile hti_t *)hti)->next);
+ int did_copy = hti_copy_entry(hti, ent, key_hash, VOLATILE_DEREF(hti).next);
if (did_copy) {
- SYNC_ADD(&hti->num_entries_copied, 1);
+ (void)SYNC_ADD(&hti->num_entries_copied, 1);
}
- TRACE("h0", "hti_cas: value in the middle of a copy, copy completed by %s",
+ TRACE("h0", "hti_cas: value in the middle of a copy, copy completed by %s",
(did_copy ? "self" : "other"), 0);
}
TRACE("h0", "hti_cas: value copied to next table, retry on next table", 0, 0);
// The set succeeded. Adjust the value count.
if (old_existed && new == DOES_NOT_EXIST) {
- SYNC_ADD(&hti->count, -1);
+ (void)SYNC_ADD(&hti->count, -1);
} else if (!old_existed && new != DOES_NOT_EXIST) {
- SYNC_ADD(&hti->count, 1);
+ (void)SYNC_ADD(&hti->count, 1);
}
// Return the previous value.
volatile entry_t *ent = hti_lookup(hti, key, key_hash, &is_empty);
// When hti_lookup() returns NULL it means we hit the reprobe limit while
- // searching the table. In that case, if a copy is in progress the key
+ // searching the table. In that case, if a copy is in progress the key
// might exist in the copy.
if (EXPECT_FALSE(ent == NULL)) {
- if (((volatile hti_t *)hti)->next != NULL)
+ if (VOLATILE_DEREF(hti).next != NULL)
return hti_get(hti->next, key, key_hash); // recursive tail-call
return DOES_NOT_EXIST;
}
map_val_t ent_val = ent->val;
if (EXPECT_FALSE(IS_TAGGED(ent_val, TAG1))) {
if (EXPECT_FALSE(ent_val != COPIED_VALUE && ent_val != TAG_VALUE(TOMBSTONE, TAG1))) {
- int did_copy = hti_copy_entry(hti, ent, key_hash, ((volatile hti_t *)hti)->next);
+ int did_copy = hti_copy_entry(hti, ent, key_hash, VOLATILE_DEREF(hti).next);
if (did_copy) {
- SYNC_ADD(&hti->num_entries_copied, 1);
+ (void)SYNC_ADD(&hti->num_entries_copied, 1);
}
}
- return hti_get(((volatile hti_t *)hti)->next, key, key_hash); // tail-call
+ return hti_get(VOLATILE_DEREF(hti).next, key, key_hash); // tail-call
}
return (ent_val == TOMBSTONE) ? DOES_NOT_EXIST : ent_val;
// returns TRUE if copy is done
static int hti_help_copy (hti_t *hti) {
volatile entry_t *ent;
- size_t limit;
+ size_t limit;
size_t total_copied = hti->num_entries_copied;
size_t num_copied = 0;
- size_t x = hti->copy_scan;
+ size_t x = hti->copy_scan;
TRACE("h1", "ht_cas: help copy. scan is %llu, size is %llu", x, 1<<hti->scale);
- if (total_copied != (1 << hti->scale)) {
+ if (total_copied != (1ULL << hti->scale)) {
// Panic if we've been around the array twice and still haven't finished the copy.
- int panic = (x >= (1 << (hti->scale + 1)));
+ int panic = (x >= (1ULL << (hti->scale + 1)));
if (!panic) {
limit = ENTRIES_PER_COPY_CHUNK;
// Reserve some entries for this thread to copy. There is a race condition here because the
// fetch and add isn't atomic, but that is ok.
- hti->copy_scan = x + ENTRIES_PER_COPY_CHUNK;
+ hti->copy_scan = x + ENTRIES_PER_COPY_CHUNK;
- // <copy_scan> might be larger than the size of the table, if some thread stalls while
+ // <copy_scan> might be larger than the size of the table, if some thread stalls while
// copying. In that case we just wrap around to the begining and make another pass through
// the table.
ent = hti->table + (x & MASK(hti->scale));
TRACE("h1", "ht_cas: help copy panic", 0, 0);
// scan the whole table
ent = hti->table;
- limit = (1 << hti->scale);
+ limit = (1ULL << hti->scale);
}
// Copy the entries
for (int i = 0; i < limit; ++i) {
num_copied += hti_copy_entry(hti, ent++, 0, hti->next);
- assert(ent <= hti->table + (1 << hti->scale));
+ assert(ent <= hti->table + (1ULL << hti->scale));
}
if (num_copied != 0) {
total_copied = SYNC_ADD(&hti->num_entries_copied, num_copied);
}
}
- return (total_copied == (1 << hti->scale));
+ return (total_copied == (1ULL << hti->scale));
}
static void hti_defer_free (hti_t *hti) {
assert(hti->ref_count == 0);
- for (uint32_t i = 0; i < (1 << hti->scale); ++i) {
+ for (uint32_t i = 0; i < (1ULL << hti->scale); ++i) {
map_key_t key = hti->table[i].key;
map_val_t val = hti->table[i].val;
if (val == COPIED_VALUE)
rcu_defer_free(GET_PTR(key));
}
}
+#ifdef USE_SYSTEM_MALLOC
+ rcu_defer_free(hti->unaligned_table_ptr);
+#else
rcu_defer_free((void *)hti->table);
+#endif
rcu_defer_free(hti);
}
size_t count = 0;
while (hti) {
count += hti->count;
- hti = hti->next;
+ hti = hti->next;
}
return count;
}
hashtable_t *ht = nbd_malloc(sizeof(hashtable_t));
ht->key_type = key_type;
ht->hti = (hti_t *)hti_alloc(ht, MIN_SCALE);
+ ht->hti_copies = 0;
+ ht->density = 0.0;
return ht;
}
nbd_free(ht);
}
-void ht_print (hashtable_t *ht) {
+void ht_print (hashtable_t *ht, int verbose) {
+ printf("probe:%-2d density:%.1f%% count:%-8lld ", ht->probe, ht->density, (uint64_t)ht_count(ht));
hti_t *hti = ht->hti;
while (hti) {
- printf("hti:%p scale:%u count:%d copied:%d\n", hti, hti->scale, hti->count, hti->num_entries_copied);
- for (int i = 0; i < (1 << hti->scale); ++i) {
- volatile entry_t *ent = hti->table + i;
- printf("[0x%x] 0x%llx:0x%llx\n", i, (uint64_t)ent->key, (uint64_t)ent->val);
- if (i > 30) {
- printf("...\n");
- break;
+ if (verbose) {
+ for (int i = 0; i < (1ULL << hti->scale); ++i) {
+ volatile entry_t *ent = hti->table + i;
+ printf("[0x%x] 0x%llx:0x%llx\n", i, (uint64_t)ent->key, (uint64_t)ent->val);
+ if (i > 30) {
+ printf("...\n");
+ break;
+ }
}
}
+ int scale = hti->scale;
+ printf("hti count:%lld scale:%d key density:%.1f%% value density:%.1f%% probe:%d\n",
+ (uint64_t)hti->count, scale, (double)hti->key_count / (1ULL << scale) * 100,
+ (double)hti->count / (1ULL << scale) * 100, hti->probe);
hti = hti->next;
}
}
volatile entry_t *ent;
map_key_t key;
map_val_t val;
- size_t table_size = (1 << iter->hti->scale);
+ size_t table_size = (1ULL << iter->hti->scale);
do {
iter->idx++;
if (iter->idx == table_size) {
} while (key == DOES_NOT_EXIST || val == DOES_NOT_EXIST || val == TOMBSTONE);
- if (key_ptr) {
- *key_ptr = key;
- }
if (val == COPIED_VALUE) {
const datatype_t *key_type = iter->hti->ht->key_type;
#ifdef NBD32
uint32_t hash = (key_type == NULL) ? murmur32_8b((uint64_t)key) : key_type->hash((void *)key);
#endif
val = hti_get(iter->hti->next, (map_key_t)ent->key, hash);
- }
+ // Go to the next entry if key is already deleted.
+ if (val == DOES_NOT_EXIST)
+ return ht_iter_next(iter, key_ptr); // recursive tail-call
+ }
+
+ if (key_ptr) {
+ *key_ptr = key;
+ }
return val;
}
projects / nbds / blobdiff