hashmap: new and fast hashmap with dynamic size
ka-weihe
Alexander Medvednikov
parent
219239eadc
commit
6fd175d9be
|
|
@ -2,120 +2,242 @@
|
|||
// Use of this source code is governed by an MIT license
|
||||
// that can be found in the LICENSE file.
|
||||
module hashmap
|
||||
/*
|
||||
This is work in progress.
|
||||
A very early test version of the Hashmap with a fixed size.
|
||||
Only works with string keys and int values for now.
|
||||
|
||||
I added this to improve performance of the V compiler,
|
||||
which uses lots of O(log n) map get's. Turned out with N < 10 000
|
||||
the performance gains are basically non-existent.
|
||||
*/
|
||||
|
||||
|
||||
struct Hashmap {
|
||||
cap int
|
||||
keys []string
|
||||
table []Hashmapentry
|
||||
elm_size int
|
||||
pub mut:
|
||||
nr_collisions int
|
||||
}
|
||||
|
||||
struct Hashmapentry {
|
||||
mut:
|
||||
key string
|
||||
val int
|
||||
next &Hashmapentry // linked list for collisions
|
||||
}
|
||||
|
||||
const (
|
||||
min_cap = 2<<10
|
||||
max_cap = 2<<20
|
||||
)
|
||||
|
||||
const(
|
||||
fnv64_prime = 1099511628211
|
||||
initial_size = 2<<4
|
||||
initial_cap = initial_size - 1
|
||||
load_factor = 0.5
|
||||
probe_offset = u16(256)
|
||||
fnv64_prime = 1099511628211
|
||||
fnv64_offset_basis = 14695981039346656037
|
||||
fnv32_offset_basis = u32(2166136261)
|
||||
fnv32_prime = u32(16777619)
|
||||
)
|
||||
|
||||
const(
|
||||
fnv32_offset_basis = u32(2166136261)
|
||||
fnv32_prime = u32(16777619)
|
||||
)
|
||||
|
||||
pub fn new_hashmap(planned_nr_items int) Hashmap {
|
||||
mut cap := planned_nr_items * 5
|
||||
if cap < min_cap {
|
||||
cap = min_cap
|
||||
}
|
||||
if cap > max_cap {
|
||||
cap = max_cap
|
||||
}
|
||||
return Hashmap{
|
||||
cap: cap
|
||||
elm_size: 4
|
||||
table: make(cap, cap, sizeof(Hashmapentry))
|
||||
}
|
||||
pub struct Hashmap {
|
||||
mut:
|
||||
info &u16
|
||||
key_values &KeyValue
|
||||
cap int
|
||||
pub mut:
|
||||
size int
|
||||
}
|
||||
|
||||
pub fn (m mut Hashmap) set(key string, val int) {
|
||||
// mut hash := int(b_fabs(key.hash()))
|
||||
// idx := hash % m.cap
|
||||
idx := int(fnv1a32(key) % m.cap)
|
||||
if m.table[idx].key.len != 0 {
|
||||
// println('\nset() idx=$idx key="$key" hash="$hash" val=$val')
|
||||
m.nr_collisions++
|
||||
// println('collision:' + m.table[idx].key)
|
||||
mut e := &m.table[idx]
|
||||
for e.next != 0 {
|
||||
e = e.next
|
||||
}
|
||||
e.next = &Hashmapentry{
|
||||
key,val,0}
|
||||
}
|
||||
else {
|
||||
m.table[idx] = Hashmapentry{
|
||||
key,val,0}
|
||||
}
|
||||
}
|
||||
|
||||
pub fn (m &Hashmap) get(key string) int {
|
||||
// mut hash := int(b_fabs(key.hash()))
|
||||
// idx := hash % m.cap
|
||||
idx := int(fnv1a32(key) % m.cap)
|
||||
mut e := &m.table[idx]
|
||||
for e.next != 0 {
|
||||
// todo unsafe {
|
||||
if e.key == key {
|
||||
return e.val
|
||||
}
|
||||
e = e.next
|
||||
}
|
||||
return e.val
|
||||
}
|
||||
|
||||
[inline]
|
||||
fn b_fabs(v int) f64 {
|
||||
return if v < 0 { -v } else { v }
|
||||
}
|
||||
|
||||
// inline functions here for speed
|
||||
// rather than full impl in vlib
|
||||
[inline]
|
||||
fn fnv1a32(data string) u32 {
|
||||
mut hash := fnv32_offset_basis
|
||||
for i := 0; i < data.len; i++ {
|
||||
hash = (hash ^ u32(data[i])) * fnv32_prime
|
||||
}
|
||||
return hash
|
||||
struct KeyValue {
|
||||
key string
|
||||
mut:
|
||||
value int
|
||||
}
|
||||
|
||||
[inline]
|
||||
fn fnv1a64(data string) u64 {
|
||||
mut hash := fnv64_offset_basis
|
||||
for i := 0; i < data.len; i++ {
|
||||
hash = (hash ^ u64(data[i])) * fnv64_prime
|
||||
}
|
||||
return hash
|
||||
mut hash := fnv64_offset_basis
|
||||
for i := 0; i < data.len; i++ {
|
||||
hash = (hash ^ u64(data[i])) * fnv64_prime
|
||||
}
|
||||
return hash
|
||||
}
|
||||
|
||||
pub fn new_hashmap() Hashmap {
|
||||
return Hashmap{
|
||||
info: &u16(calloc(sizeof(u16) * initial_size))
|
||||
key_values: &KeyValue(calloc(sizeof(KeyValue) * initial_size))
|
||||
cap: initial_cap
|
||||
size: 0
|
||||
}
|
||||
}
|
||||
|
||||
pub fn (h mut Hashmap) set(key string, value int) {
|
||||
// The load factor is 0.5.
|
||||
// It will be adjustable in the future and with
|
||||
// a higher default settings to lower memory usage.
|
||||
if (h.size<<1) == (h.cap - 1) {
|
||||
h.rehash()
|
||||
}
|
||||
// Hash-function will be swapped for wyhash
|
||||
hash := fnv1a64(key)
|
||||
mut info := u16((hash>>56) | probe_offset)
|
||||
mut index := hash & h.cap
|
||||
// While probe count is less
|
||||
for info < h.info[index] {
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// While we might have a match
|
||||
for info == h.info[index] {
|
||||
if key == h.key_values[index].key {
|
||||
h.key_values[index].value = value
|
||||
return
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// Match is not possible anymore.
|
||||
// Probe until an empty index is found.
|
||||
// Swap when probe count is higher/richer (Robin Hood).
|
||||
mut current_key := key
|
||||
mut current_value := value
|
||||
for h.info[index] != 0 {
|
||||
if info > h.info[index] {
|
||||
tmp_kv := h.key_values[index]
|
||||
tmp_info := h.info[index]
|
||||
h.key_values[index] = KeyValue{
|
||||
current_key,current_value}
|
||||
h.info[index] = info
|
||||
current_key = tmp_kv.key
|
||||
current_value = tmp_kv.value
|
||||
info = tmp_info
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// Should almost never happen
|
||||
if (info & 0xFF00) == 0xFF00 {
|
||||
h.rehash()
|
||||
h.set(current_key, current_value)
|
||||
return
|
||||
}
|
||||
h.info[index] = info
|
||||
h.key_values[index] = KeyValue{
|
||||
current_key,current_value}
|
||||
h.size++
|
||||
}
|
||||
|
||||
fn (h mut Hashmap) rehash() {
|
||||
old_cap := h.cap
|
||||
h.cap = ((h.cap + 1)<<1) - 1
|
||||
mut new_key_values := &KeyValue(calloc(sizeof(KeyValue) * (h.cap + 1)))
|
||||
mut new_info := &u16(calloc(sizeof(u16) * (h.cap + 1)))
|
||||
for i in 0 .. (old_cap + 1) {
|
||||
if h.info[i] != 0 {
|
||||
key := h.key_values[i].key
|
||||
value := h.key_values[i].value
|
||||
hash := fnv1a64(key)
|
||||
mut info := u16((hash>>56) | probe_offset)
|
||||
mut index := hash & h.cap
|
||||
// While probe count is less
|
||||
for info < new_info[index] {
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// While we might have a match
|
||||
for info == new_info[index] {
|
||||
if key == new_key_values[index].key {
|
||||
new_key_values[index].value = value
|
||||
return
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// Match is not possible anymore.
|
||||
// Probe until an empty index is found.
|
||||
// Swap when probe count is higher/richer (Robin Hood).
|
||||
mut current_key := key
|
||||
mut current_value := value
|
||||
for new_info[index] != 0 {
|
||||
if info > new_info[index] {
|
||||
tmp_kv := new_key_values[index]
|
||||
tmp_info := new_info[index]
|
||||
new_key_values[index] = KeyValue{
|
||||
current_key,current_value}
|
||||
new_info[index] = info
|
||||
current_key = tmp_kv.key
|
||||
current_value = tmp_kv.value
|
||||
info = tmp_info
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// Should almost never happen
|
||||
if (info & 0xFF00) == 0xFF00 {
|
||||
h.rehash()
|
||||
h.set(current_key, current_value)
|
||||
return
|
||||
}
|
||||
new_info[index] = info
|
||||
new_key_values[index] = KeyValue{
|
||||
current_key,current_value}
|
||||
}
|
||||
}
|
||||
h.key_values = new_key_values
|
||||
h.info = new_info
|
||||
}
|
||||
|
||||
pub fn (h mut Hashmap) delete(key string) {
|
||||
hash := fnv1a64(key)
|
||||
mut index := hash & h.cap
|
||||
mut info := u16((hash>>56) | probe_offset)
|
||||
for info < h.info[index] {
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
// Perform backwards shifting
|
||||
for info == h.info[index] {
|
||||
if key == h.key_values[index].key {
|
||||
mut old_index := index
|
||||
index = (index + 1) & h.cap
|
||||
mut current_info := h.info[index]
|
||||
for (current_info>>8) > 1 {
|
||||
h.info[old_index] = current_info - probe_offset
|
||||
h.key_values[old_index] = h.key_values[index]
|
||||
old_index = index
|
||||
index = (index + 1) & h.cap
|
||||
current_info = h.info[index]
|
||||
}
|
||||
h.info[old_index] = 0
|
||||
h.size--
|
||||
return
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
}
|
||||
|
||||
pub fn (h Hashmap) get(key string) int {
|
||||
hash := fnv1a64(key)
|
||||
mut index := hash & h.cap
|
||||
mut info := u16((hash>>56) | probe_offset)
|
||||
for info < h.info[index] {
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
for info == h.info[index] {
|
||||
if key == h.key_values[index].key {
|
||||
return h.key_values[index].value
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
pub fn (h Hashmap) exists(key string) bool {
|
||||
hash := fnv1a64(key)
|
||||
mut index := hash & h.cap
|
||||
mut info := u16((hash>>56) | probe_offset)
|
||||
for info < h.info[index] {
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
for info == h.info[index] {
|
||||
if key == h.key_values[index].key {
|
||||
return true
|
||||
}
|
||||
index = (index + 1) & h.cap
|
||||
info += probe_offset
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
pub fn (h Hashmap) keys() []string {
|
||||
size := h.size
|
||||
mut keys := [''].repeat(size)
|
||||
mut j := 0
|
||||
for i in 0 .. (h.cap + 1) {
|
||||
if h.info[i] != 0 {
|
||||
keys[j] = h.key_values[i].key
|
||||
j++
|
||||
}
|
||||
}
|
||||
return keys
|
||||
}
|
||||
|
|
|
|||
|
|
@ -3,7 +3,7 @@ module hashmap
|
|||
import rand
|
||||
|
||||
fn test_random_strings() {
|
||||
mut m := new_hashmap(1000)
|
||||
mut m := new_hashmap()
|
||||
for i in 0..1000 {
|
||||
mut buf := []byte
|
||||
for j in 0..10 {
|
||||
|
|
@ -21,12 +21,11 @@ fn test_random_strings() {
|
|||
|
||||
fn test_large_hashmap() {
|
||||
N := 300 * 1000
|
||||
mut nums := new_hashmap(N)
|
||||
mut nums := new_hashmap()
|
||||
for i := 0; i < N; i++ {
|
||||
key := i.str()
|
||||
nums.set(key, i)
|
||||
}
|
||||
println('nr collisions: $nums.nr_collisions')
|
||||
for i := 0; i < N; i++ {
|
||||
key := i.str()
|
||||
assert nums.get(key) == i
|
||||
|
|
|
|||
Loading…
Reference in New Issue