shaba/lvm2
1
0
Fork 0
mirror of git://sourceware.org/git/lvm2.git synced 2024-12-22 17:35:59 +03:00
Code Releases Activity
0181c77e3f
lvm2/base/data-struct/radix-tree.c

750 lines
16 KiB
C
Raw Normal View History

radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
// Copyright (C) 2018 Red Hat, Inc. All rights reserved.
//
// This file is part of LVM2.
//
// This copyrighted material is made available to anyone wishing to use,
// modify, copy, or redistribute it subject to the terms and conditions
// of the GNU Lesser General Public License v.2.1.
//
// You should have received a copy of the GNU Lesser General Public License
// along with this program; if not, write to the Free Software Foundation,
// Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
#include "radix-tree.h"
#include "base/memory/container_of.h"
#include "base/memory/zalloc.h"
#include <assert.h>
#include <stdlib.h>
#include <stdio.h>
//----------------------------------------------------------------
enum node_type {
UNSET = 0,
VALUE,
VALUE_CHAIN,
PREFIX_CHAIN,
NODE4,
NODE16,
NODE48,
NODE256
};
struct value {
enum node_type type;
union radix_value value;
};
// This is used for entries that have a key which is a prefix of another key.
struct value_chain {
union radix_value value;
struct value child;
};
struct prefix_chain {
struct value child;
unsigned len;
uint8_t prefix[0];
};
struct node4 {
uint32_t nr_entries;
uint8_t keys[4];
struct value values[4];
};
struct node16 {
uint32_t nr_entries;
uint8_t keys[16];
struct value values[16];
};
struct node48 {
uint32_t nr_entries;
uint8_t keys[256];
struct value values[48];
};
struct node256 {
radix_tree: add remove method
2018-05-23 14:48:06 +03:00
uint32_t nr_entries;
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
struct value values[256];
};
struct radix_tree {
unsigned nr_entries;
struct value root;
data-struct/radix-tree: pass the value dtr into create. Rather than having to pass it into every method that removes items.
2018-05-29 13:03:10 +03:00
radix_value_dtr dtr;
void *dtr_context;
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
};
//----------------------------------------------------------------
data-struct/radix-tree: pass the value dtr into create. Rather than having to pass it into every method that removes items.
2018-05-29 13:03:10 +03:00
struct radix_tree *radix_tree_create(radix_value_dtr dtr, void *dtr_context)
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
{
struct radix_tree *rt = malloc(sizeof(*rt));
if (rt) {
rt->nr_entries = 0;
rt->root.type = UNSET;
data-struct/radix-tree: pass the value dtr into create. Rather than having to pass it into every method that removes items.
2018-05-29 13:03:10 +03:00
rt->dtr = dtr;
rt->dtr_context = dtr_context;
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
}
return rt;
}
static void _free_node(struct value v, radix_value_dtr dtr, void *context)
{
unsigned i;
struct value_chain *vc;
struct prefix_chain *pc;
struct node4 *n4;
struct node16 *n16;
struct node48 *n48;
struct node256 *n256;
switch (v.type) {
case UNSET:
break;
case VALUE:
if (dtr)
dtr(context, v.value);
break;
case VALUE_CHAIN:
vc = v.value.ptr;
if (dtr)
dtr(context, vc->value);
_free_node(vc->child, dtr, context);
free(vc);
break;
case PREFIX_CHAIN:
pc = v.value.ptr;
_free_node(pc->child, dtr, context);
free(pc);
break;
case NODE4:
n4 = (struct node4 *) v.value.ptr;
for (i = 0; i < n4->nr_entries; i++)
_free_node(n4->values[i], dtr, context);
free(n4);
break;
case NODE16:
n16 = (struct node16 *) v.value.ptr;
for (i = 0; i < n16->nr_entries; i++)
_free_node(n16->values[i], dtr, context);
free(n16);
break;
case NODE48:
n48 = (struct node48 *) v.value.ptr;
for (i = 0; i < n48->nr_entries; i++)
_free_node(n48->values[i], dtr, context);
free(n48);
break;
case NODE256:
n256 = (struct node256 *) v.value.ptr;
for (i = 0; i < 256; i++)
_free_node(n256->values[i], dtr, context);
free(n256);
break;
}
}
data-struct/radix-tree: pass the value dtr into create. Rather than having to pass it into every method that removes items.
2018-05-29 13:03:10 +03:00
void radix_tree_destroy(struct radix_tree *rt)
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
{
data-struct/radix-tree: pass the value dtr into create. Rather than having to pass it into every method that removes items.
2018-05-29 13:03:10 +03:00
_free_node(rt->root, rt->dtr, rt->dtr_context);
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
free(rt);
}
static bool _insert(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv);
static bool _insert_unset(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
unsigned len = ke - kb;
if (!len) {
// value
v->type = VALUE;
v->value = rv;
} else {
// prefix -> value
struct prefix_chain *pc = zalloc(sizeof(*pc) + len);
if (!pc)
return false;
pc->child.type = VALUE;
pc->child.value = rv;
pc->len = len;
memcpy(pc->prefix, kb, len);
v->type = PREFIX_CHAIN;
v->value.ptr = pc;
}
return true;
}
static bool _insert_value(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
unsigned len = ke - kb;
if (!len)
// overwrite
v->value = rv;
else {
// value_chain -> value
struct value_chain *vc = zalloc(sizeof(*vc));
if (!vc)
return false;
vc->value = v->value;
if (!_insert(&vc->child, kb, ke, rv)) {
free(vc);
return false;
}
v->type = VALUE_CHAIN;
v->value.ptr = vc;
}
return true;
}
static bool _insert_value_chain(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct value_chain *vc = v->value.ptr;
return _insert(&vc->child, kb, ke, rv);
}
static unsigned min(unsigned lhs, unsigned rhs)
{
if (lhs <= rhs)
return lhs;
else
return rhs;
}
static bool _insert_prefix_chain(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct prefix_chain *pc = v->value.ptr;
if (*kb == pc->prefix[0]) {
// There's a common prefix let's split the chain into two and
// recurse.
struct prefix_chain *pc2;
unsigned i, len = min(pc->len, ke - kb);
for (i = 0; i < len; i++)
if (kb[i] != pc->prefix[i])
break;
pc2 = zalloc(sizeof(*pc2) + pc->len - i);
pc2->len = pc->len - i;
memmove(pc2->prefix, pc->prefix + i, pc2->len);
pc2->child = pc->child;
// FIXME: this trashes pc so we can't back out
pc->child.type = PREFIX_CHAIN;
pc->child.value.ptr = pc2;
pc->len = i;
if (!_insert(&pc->child, kb + i, ke, rv)) {
free(pc2);
return false;
}
} else {
// Stick an n4 in front.
struct node4 *n4 = zalloc(sizeof(*n4));
if (!n4)
return false;
n4->keys[0] = *kb;
if (!_insert(n4->values, kb + 1, ke, rv)) {
free(n4);
return false;
}
if (pc->len) {
n4->keys[1] = pc->prefix[0];
if (pc->len == 1) {
n4->values[1] = pc->child;
free(pc);
} else {
memmove(pc->prefix, pc->prefix + 1, pc->len - 1);
pc->len--;
n4->values[1] = *v;
}
n4->nr_entries = 2;
} else
n4->nr_entries = 1;
v->type = NODE4;
v->value.ptr = n4;
}
return true;
}
static bool _insert_node4(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct node4 *n4 = v->value.ptr;
if (n4->nr_entries == 4) {
struct node16 *n16 = zalloc(sizeof(*n16));
if (!n16)
return false;
n16->nr_entries = 5;
memcpy(n16->keys, n4->keys, sizeof(n4->keys));
memcpy(n16->values, n4->values, sizeof(n4->values));
n16->keys[4] = *kb;
if (!_insert(n16->values + 4, kb + 1, ke, rv)) {
free(n16);
return false;
}
free(n4);
v->type = NODE16;
v->value.ptr = n16;
} else {
n4 = v->value.ptr;
if (!_insert(n4->values + n4->nr_entries, kb + 1, ke, rv))
return false;
n4->keys[n4->nr_entries] = *kb;
n4->nr_entries++;
}
return true;
}
static bool _insert_node16(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct node16 *n16 = v->value.ptr;
if (n16->nr_entries == 16) {
unsigned i;
struct node48 *n48 = zalloc(sizeof(*n48));
if (!n48)
return false;
n48->nr_entries = 17;
memset(n48->keys, 48, sizeof(n48->keys));
for (i = 0; i < 16; i++) {
n48->keys[n16->keys[i]] = i;
n48->values[i] = n16->values[i];
}
n48->keys[*kb] = 16;
if (!_insert(n48->values + 16, kb + 1, ke, rv)) {
free(n48);
return false;
}
free(n16);
v->type = NODE48;
v->value.ptr = n48;
} else {
if (!_insert(n16->values + n16->nr_entries, kb + 1, ke, rv))
return false;
n16->keys[n16->nr_entries] = *kb;
n16->nr_entries++;
}
return true;
}
static bool _insert_node48(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct node48 *n48 = v->value.ptr;
if (n48->nr_entries == 48) {
unsigned i;
struct node256 *n256 = zalloc(sizeof(*n256));
if (!n256)
return false;
for (i = 0; i < 256; i++) {
if (n48->keys[i] >= 48)
continue;
n256->values[i] = n48->values[n48->keys[i]];
}
if (!_insert(n256->values + *kb, kb + 1, ke, rv)) {
free(n256);
return false;
}
free(n48);
v->type = NODE256;
v->value.ptr = n256;
} else {
if (!_insert(n48->values + n48->nr_entries, kb + 1, ke, rv))
return false;
n48->keys[*kb] = n48->nr_entries;
n48->nr_entries++;
}
return true;
}
static bool _insert_node256(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct node256 *n256 = v->value.ptr;
radix_tree: add remove method
2018-05-23 14:48:06 +03:00
bool was_unset = n256->values[*kb].type == UNSET;
if (!_insert(n256->values + *kb, kb + 1, ke, rv))
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
return false;
radix_tree: add remove method
2018-05-23 14:48:06 +03:00
if (was_unset)
n256->nr_entries++;
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
return true;
}
// FIXME: the tree should not be touched if insert fails (eg, OOM)
static bool _insert(struct value *v, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
if (kb == ke) {
if (v->type == UNSET) {
v->type = VALUE;
v->value = rv;
} else if (v->type == VALUE) {
v->value = rv;
} else {
struct value_chain *vc = zalloc(sizeof(*vc));
if (!vc)
return false;
vc->value = rv;
vc->child = *v;
v->type = VALUE_CHAIN;
v->value.ptr = vc;
}
return true;
}
switch (v->type) {
case UNSET:
return _insert_unset(v, kb, ke, rv);
case VALUE:
return _insert_value(v, kb, ke, rv);
case VALUE_CHAIN:
return _insert_value_chain(v, kb, ke, rv);
case PREFIX_CHAIN:
return _insert_prefix_chain(v, kb, ke, rv);
case NODE4:
return _insert_node4(v, kb, ke, rv);
case NODE16:
return _insert_node16(v, kb, ke, rv);
case NODE48:
return _insert_node48(v, kb, ke, rv);
case NODE256:
return _insert_node256(v, kb, ke, rv);
}
// can't get here
return false;
}
struct lookup_result {
struct value *v;
uint8_t *kb;
};
static struct lookup_result _lookup_prefix(struct value *v, uint8_t *kb, uint8_t *ke)
{
unsigned i;
struct value_chain *vc;
struct prefix_chain *pc;
struct node4 *n4;
struct node16 *n16;
struct node48 *n48;
struct node256 *n256;
if (kb == ke)
return (struct lookup_result) {.v = v, .kb = kb};
switch (v->type) {
case UNSET:
case VALUE:
break;
case VALUE_CHAIN:
vc = v->value.ptr;
return _lookup_prefix(&vc->child, kb, ke);
case PREFIX_CHAIN:
pc = v->value.ptr;
if (ke - kb < pc->len)
return (struct lookup_result) {.v = v, .kb = kb};
for (i = 0; i < pc->len; i++)
if (kb[i] != pc->prefix[i])
return (struct lookup_result) {.v = v, .kb = kb};
return _lookup_prefix(&pc->child, kb + pc->len, ke);
case NODE4:
n4 = v->value.ptr;
for (i = 0; i < n4->nr_entries; i++)
if (n4->keys[i] == *kb)
return _lookup_prefix(n4->values + i, kb + 1, ke);
break;
case NODE16:
// FIXME: use binary search or simd?
n16 = v->value.ptr;
for (i = 0; i < n16->nr_entries; i++)
if (n16->keys[i] == *kb)
return _lookup_prefix(n16->values + i, kb + 1, ke);
break;
case NODE48:
n48 = v->value.ptr;
i = n48->keys[*kb];
if (i < 48)
return _lookup_prefix(n48->values + i, kb + 1, ke);
break;
case NODE256:
n256 = v->value.ptr;
return _lookup_prefix(n256->values + *kb, kb + 1, ke);
}
return (struct lookup_result) {.v = v, .kb = kb};
}
bool radix_tree_insert(struct radix_tree *rt, uint8_t *kb, uint8_t *ke, union radix_value rv)
{
struct lookup_result lr = _lookup_prefix(&rt->root, kb, ke);
if (_insert(lr.v, lr.kb, ke, rv)) {
rt->nr_entries++;
return true;
}
return false;
}
radix_tree: add remove method
2018-05-23 14:48:06 +03:00
// Note the degrade functions also free the original node.
static void _degrade_to_n4(struct node16 *n16, struct value *result)
{
struct node4 *n4 = zalloc(sizeof(*n4));
n4->nr_entries = n16->nr_entries;
memcpy(n4->keys, n16->keys, n16->nr_entries * sizeof(*n4->keys));
memcpy(n4->values, n16->values, n16->nr_entries * sizeof(*n4->values));
free(n16);
result->type = NODE4;
result->value.ptr = n4;
}
static void _degrade_to_n16(struct node48 *n48, struct value *result)
{
struct node4 *n16 = zalloc(sizeof(*n16));
n16->nr_entries = n48->nr_entries;
memcpy(n16->keys, n48->keys, n48->nr_entries * sizeof(*n16->keys));
memcpy(n16->values, n48->values, n48->nr_entries * sizeof(*n16->values));
free(n48);
result->type = NODE16;
result->value.ptr = n16;
}
static void _degrade_to_n48(struct node256 *n256, struct value *result)
{
unsigned i, count = 0;
struct node4 *n48 = zalloc(sizeof(*n48));
n48->nr_entries = n256->nr_entries;
for (i = 0; i < 256; i++) {
if (n256->values[i].type == UNSET)
continue;
n48->keys[count] = i;
n48->values[count] = n256->values[i];
count++;
}
free(n256);
result->type = NODE48;
result->value.ptr = n48;
}
static bool _remove(struct value *root, uint8_t *kb, uint8_t *ke)
{
bool r;
unsigned i;
struct value_chain *vc;
struct prefix_chain *pc;
struct node4 *n4;
struct node16 *n16;
struct node48 *n48;
struct node256 *n256;
if (kb == ke) {
if (root->type == VALUE) {
root->type = UNSET;
return true;
} else if (root->type == VALUE_CHAIN) {
vc = root->value.ptr;
memcpy(root, &vc->child, sizeof(*root));
free(vc);
return true;
} else
return false;
}
switch (root->type) {
case UNSET:
case VALUE:
// this is a value for a prefix of the key
return false;
case VALUE_CHAIN:
vc = root->value.ptr;
r = _remove(&vc->child, kb, ke);
if (r && (vc->child.type == UNSET)) {
memcpy(root, &vc->child, sizeof(*root));
free(vc);
}
return r;
case PREFIX_CHAIN:
pc = root->value.ptr;
if (ke - kb < pc->len)
return false;
for (i = 0; i < pc->len; i++)
if (kb[i] != pc->prefix[i])
return false;
return _remove(&pc->child, kb + pc->len, ke);
case NODE4:
n4 = root->value.ptr;
for (i = 0; i < n4->nr_entries; i++) {
if (n4->keys[i] == *kb) {
r = _remove(n4->values + i, kb + 1, ke);
if (r && n4->values[i].type == UNSET) {
n4->nr_entries--;
if (i < n4->nr_entries)
// slide the entries down
memmove(n4->keys + i, n4->keys + i + 1,
sizeof(*n4->keys) * (n4->nr_entries - i));
if (!n4->nr_entries)
root->type = UNSET;
}
return r;
}
}
return false;
case NODE16:
n16 = root->value.ptr;
for (i = 0; i < n16->nr_entries; i++) {
if (n16->keys[i] == *kb) {
r = _remove(n16->values + i, kb + 1, ke);
if (r && n16->values[i].type == UNSET) {
n16->nr_entries--;
if (i < n16->nr_entries)
// slide the entries down
memmove(n16->keys + i, n16->keys + i + 1,
sizeof(*n16->keys) * (n16->nr_entries - i));
if (n16->nr_entries <= 4)
_degrade_to_n4(n16, root);
}
return r;
}
}
return false;
case NODE48:
n48 = root->value.ptr;
i = n48->keys[*kb];
if (i < 48) {
r = _remove(n48->values + i, kb + 1, ke);
if (r && n48->values[i].type == UNSET) {
n48->keys[*kb] = 48;
n48->nr_entries--;
if (n48->nr_entries <= 16)
_degrade_to_n16(n48, root);
}
return r;
}
return false;
case NODE256:
n256 = root->value.ptr;
r = _remove(n256->values + (*kb), kb + 1, ke);
if (r && n256->values[*kb].type == UNSET) {
n256->nr_entries--;
if (n256->nr_entries <= 48)
_degrade_to_n48(n256, root);
}
return r;
}
return false;
}
bool radix_tree_remove(struct radix_tree *rt, uint8_t *key_begin, uint8_t *key_end)
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
{
radix_tree: add remove method
2018-05-23 14:48:06 +03:00
if (_remove(&rt->root, key_begin, key_end)) {
rt->nr_entries--;
return true;
}
return false;
radix-tree: First drop of radix tree. An implementation of an adaptive radix tree. Has the following nice properties: - At least as fast as the hash table - Uses less memory - You don't need to give an expected size when you create - It scales nicely (ie. no large reallocations like the hash table). - You can iterate the keys in lexicographical order. Only insert and lookup are implemented so far. Plus there's a lot more performance to come.
2018-05-11 08:10:01 +03:00
}
bool radix_tree_lookup(struct radix_tree *rt,
uint8_t *kb, uint8_t *ke, union radix_value *result)
{
struct value_chain *vc;
struct lookup_result lr = _lookup_prefix(&rt->root, kb, ke);
if (lr.kb == ke) {
switch (lr.v->type) {
case VALUE:
*result = lr.v->value;
return true;
case VALUE_CHAIN:
vc = lr.v->value.ptr;
*result = vc->value;
return true;
default:
return false;
}
}
return false;
}
//----------------------------------------------------------------
Copy Permalink