Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

refactor: use std::unordered_map instead of bespoke utpHashTable #13

Merged
merged 9 commits into from
Nov 10, 2023
Merged

refactor: use std::unordered_map instead of bespoke utpHashTable #13

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
d339167
feat: make UTPSocketKey std::hash-friendly
ckerr Nov 3, 2023
a271636
chore: savepoint
ckerr Nov 3, 2023
9bbf6b3
perf: do not rebuild UTPSocketKey in LookupAdjacent
ckerr Nov 3, 2023
165a316
refactor: replace UTPSocketHT method with one more narrowly-scoped
ckerr Nov 3, 2023
958758b
copyediting: tab damage, constify
ckerr Nov 3, 2023
13b5cb4
chore: remove unnecessary #include utp_hash.h
ckerr Nov 3, 2023
b109597
chore: remove extra include string.h
ckerr Nov 3, 2023
d0f70e6
chore: add code comments for new code that needs it
ckerr Nov 3, 2023
13545a9
fixup! chore: savepoint
ckerr Nov 9, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 1 addition & 5 deletions utp_api.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
* THE SOFTWARE.
*/

#include <assert.h>
#include <stdio.h>
#include "utp_internal.h"
#include "utp_utils.h"
Expand Down Expand Up @@ -72,7 +73,6 @@ struct_utp_context::struct_utp_context()
memset(&context_stats, 0, sizeof(context_stats));
memset(callbacks, 0, sizeof(callbacks));
target_delay = CCONTROL_TARGET;
utp_sockets = new UTPSocketHT;

callbacks[UTP_GET_UDP_MTU] = &utp_default_get_udp_mtu;
callbacks[UTP_GET_UDP_OVERHEAD] = &utp_default_get_udp_overhead;
Expand All @@ -92,10 +92,6 @@ struct_utp_context::struct_utp_context()
last_check = 0;
}

struct_utp_context::~struct_utp_context() {
delete this->utp_sockets;
}

utp_context* utp_init (int version)
{
assert(version == 2);
Expand Down
194 changes: 0 additions & 194 deletions utp_hash.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,73 +20,20 @@
* THE SOFTWARE.
*/

#include <assert.h>

#include "utp_hash.h"
#include "utp_types.h"

#define LIBUTP_HASH_UNUSED ((utp_link_t)-1)

#ifdef STRICT_ALIGN
inline uint32 Read32(const void *p)
{
uint32 tmp;
memcpy(&tmp, p, sizeof tmp);
return tmp;
}

#else
inline uint32 Read32(const void *p) { return *(uint32*)p; }
#endif


// Get the amount of memory required for the hash parameters and the bucket set
// Waste a space for an unused bucket in order to ensure the following managed memory have 32-bit aligned addresses
// TODO: make this 64-bit clean
#define BASE_SIZE(bc) (sizeof(utp_hash_t) + sizeof(utp_link_t) * ((bc) + 1 - 1))

// Get a pointer to the base of the structure array managed by the hash table
#define get_bep(h) ((byte*)(h)) + BASE_SIZE((h)->N)

// Get the address of the information associated with a specific structure in the array,
// given the address of the base of the structure.
// This assumes a utp_link_t link member is at the end of the structure.
// Given compilers filling out the memory to a 32-bit clean value, this may mean that
// the location named in the structure may not be the location actually used by the hash table,
// since the compiler may have padded the end of the structure with 2 bytes after the utp_link_t member.
// TODO: this macro should not require that the variable pointing at the hash table be named 'hash'
#define ptr_to_link(p) (utp_link_t *) (((byte *) (p)) + hash->E - sizeof(utp_link_t))

// Calculate how much to allocate for a hash table with bucket count, total size, and structure count
// TODO: make this 64-bit clean
#define ALLOCATION_SIZE(bc, ts, sc) (BASE_SIZE((bc)) + (ts) * (sc))

utp_hash_t *utp_hash_create(int N, int key_size, int total_size, int initial, utp_hash_compute_t hashfun, utp_hash_equal_t compfun)
{
// Must have odd number of hash buckets (prime number is best)
assert(N % 2);
// Ensure structures will be at aligned memory addresses
// TODO: make this 64-bit clean
assert(0 == (total_size % 4));

int size = ALLOCATION_SIZE(N, total_size, initial);
utp_hash_t *hash = (utp_hash_t *) malloc( size );
memset( hash, 0, size );

for (int i = 0; i < N + 1; ++i)
hash->inits[i] = LIBUTP_HASH_UNUSED;
hash->N = N;
hash->K = key_size;
hash->E = total_size;
hash->hash_compute = hashfun;
hash->hash_equal = compfun;
hash->allocated = initial;
hash->count = 0;
hash->used = 0;
hash->free = LIBUTP_HASH_UNUSED;
return hash;
}

uint utp_hash_mem(const void *keyp, size_t keysize)
{
uint hash = 0;
Expand All @@ -105,144 +52,3 @@ uint utp_hash_mem(const void *keyp, size_t keysize)
}
return hash;
}

uint utp_hash_mkidx(utp_hash_t *hash, const void *keyp)
{
// Generate a key from the hash
return hash->hash_compute(keyp, hash->K) % hash->N;
}

static inline bool compare(byte *a, byte *b,int n)
{
assert(n >= 4);
if (Read32(a) != Read32(b)) return false;
return memcmp(a+4, b+4, n-4) == 0;
}

#define COMPARE(h,k1,k2,ks) (((h)->hash_equal) ? (h)->hash_equal((void*)k1,(void*)k2,ks) : compare(k1,k2,ks))

// Look-up a key in the hash table.
// Returns NULL if not found
void *utp_hash_lookup(utp_hash_t *hash, const void *key)
{
utp_link_t idx = utp_hash_mkidx(hash, key);

// base pointer
byte *bep = get_bep(hash);

utp_link_t cur = hash->inits[idx];
while (cur != LIBUTP_HASH_UNUSED) {
byte *key2 = bep + (cur * hash->E);
if (COMPARE(hash, (byte*)key, key2, hash->K))
return key2;
cur = *ptr_to_link(key2);
}

return NULL;
}

// Add a new element to the hash table.
// Returns a pointer to the new element.
// This assumes the element is not already present!
void *utp_hash_add(utp_hash_t **hashp, const void *key)
{
//Allocate a new entry
byte *elemp;
utp_link_t elem;
utp_hash_t *hash = *hashp;
utp_link_t idx = utp_hash_mkidx(hash, key);

if ((elem=hash->free) == LIBUTP_HASH_UNUSED) {
utp_link_t all = hash->allocated;
if (hash->used == all) {
utp_hash_t *nhash;
if (all <= (LIBUTP_HASH_UNUSED/2)) {
all *= 2;
} else if (all != LIBUTP_HASH_UNUSED) {
all = LIBUTP_HASH_UNUSED;
} else {
// too many items! can't grow!
assert(0);
return NULL;
}
// otherwise need to allocate.
nhash = (utp_hash_t*)realloc(hash, ALLOCATION_SIZE(hash->N, hash->E, all));
if (!nhash) {
// out of memory (or too big to allocate)
assert(nhash);
return NULL;
}
hash = *hashp = nhash;
hash->allocated = all;
}

elem = hash->used++;
elemp = get_bep(hash) + elem * hash->E;
} else {
elemp = get_bep(hash) + elem * hash->E;
hash->free = *ptr_to_link(elemp);
}

*ptr_to_link(elemp) = hash->inits[idx];
hash->inits[idx] = elem;
hash->count++;

// copy key into it
memcpy(elemp, key, hash->K);
return elemp;
}

// Delete an element from the utp_hash_t
// Returns a pointer to the already deleted element.
void *utp_hash_del(utp_hash_t *hash, const void *key)
{
utp_link_t idx = utp_hash_mkidx(hash, key);

// base pointer
byte *bep = get_bep(hash);

utp_link_t *curp = &hash->inits[idx];
utp_link_t cur;
while ((cur=*curp) != LIBUTP_HASH_UNUSED) {
byte *key2 = bep + (cur * hash->E);
if (COMPARE(hash,(byte*)key,(byte*)key2, hash->K )) {
// found an item that matched. unlink it
*curp = *ptr_to_link(key2);
// Insert into freelist
*ptr_to_link(key2) = hash->free;
hash->free = cur;
hash->count--;
return key2;
}
curp = ptr_to_link(key2);
}

return NULL;
}

void *utp_hash_iterate(utp_hash_t *hash, utp_hash_iterator_t *iter)
{
utp_link_t elem;

if ((elem=iter->elem) == LIBUTP_HASH_UNUSED) {
// Find a bucket with an element
utp_link_t buck = iter->bucket + 1;
for(;;) {
if (buck >= hash->N)
return NULL;
if ((elem = hash->inits[buck]) != LIBUTP_HASH_UNUSED)
break;
buck++;
}
iter->bucket = buck;
}

byte *elemp = get_bep(hash) + (elem * hash->E);
iter->elem = *ptr_to_link(elemp);
return elemp;
}

void utp_hash_free_mem(utp_hash_t* hash)
{
free(hash);
}
116 changes: 0 additions & 116 deletions utp_hash.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,124 +23,8 @@
#ifndef __UTP_HASH_H__
#define __UTP_HASH_H__

#include <string.h> // memset
#include <stdlib.h> // malloc

#include "utp_types.h"
#include "utp_templates.h"

// TODO: make utp_link_t a template parameter to HashTable
typedef uint32 utp_link_t;

#ifdef _MSC_VER
// Silence the warning about the C99-compliant zero-length array at the end of the structure
#pragma warning (disable: 4200)
#endif

typedef uint32 (*utp_hash_compute_t)(const void *keyp, size_t keysize);
typedef uint (*utp_hash_equal_t)(const void *key_a, const void *key_b, size_t keysize);

// In memory the HashTable is laid out as follows:
// ---------------------------- low
// | hash table data members |
// ---------------------------- _
// | indices | ^
// | . | | utp_link_t indices into the key-values.
// | . | .
// ---------------------------- - <----- bep
// | keys and values | each key-value pair has size total_size
// | . |
// | . |
// ---------------------------- high
//
// The code depends on the ability of the compiler to pad the length
// of the hash table data members structure to
// a length divisible by 32-bits with no remainder.
//
// Since the number of hash buckets (indices) should be odd, the code
// asserts this and adds one to the hash bucket count to ensure that the
// following key-value pairs array starts on a 32-bit boundary.
//
// The key-value pairs array should start on a 32-bit boundary, otherwise
// processors like the ARM will silently mangle 32-bit data in these structures
// (e.g., turning 0xABCD into 0XCDAB when moving a value from memory to register
// when the memory address is 16 bits offset from a 32-bit boundary),
// also, the value will be stored at an address two bytes lower than the address
// value would ordinarily indicate.
//
// The key-value pair is of type T. The first field in T must
// be the key, i.e., the first K bytes of T contains the key.
// total_size = sizeof(T) and thus sizeof(T) >= sizeof(K)
//
// N is the number of buckets.
//
struct utp_hash_t {
utp_link_t N;
byte K;
byte E;
size_t count;
utp_hash_compute_t hash_compute;
utp_hash_equal_t hash_equal;
utp_link_t allocated;
utp_link_t used;
utp_link_t free;
utp_link_t inits[1]; // dynamic
};

#ifdef _MSC_VER
#pragma warning (default: 4200)
#endif

struct utp_hash_iterator_t {
utp_link_t bucket;
utp_link_t elem;

utp_hash_iterator_t() : bucket(0xffffffff), elem(0xffffffff) {}
};

uint utp_hash_mem(const void *keyp, size_t keysize);
uint utp_hash_comp(const void *key_a, const void *key_b, size_t keysize);

utp_hash_t *utp_hash_create(int N, int key_size, int total_size, int initial, utp_hash_compute_t hashfun = utp_hash_mem, utp_hash_equal_t eqfun = NULL);
void *utp_hash_lookup(utp_hash_t *hash, const void *key);
void *utp_hash_add(utp_hash_t **hashp, const void *key);
void *utp_hash_del(utp_hash_t *hash, const void *key);

void *utp_hash_iterate(utp_hash_t *hash, utp_hash_iterator_t *iter);
void utp_hash_free_mem(utp_hash_t *hash);

/*
This HashTable requires that T have at least sizeof(K)+sizeof(utp_link_t) bytes.
Usually done like this:

struct K {
int whatever;
};

struct T {
K wtf;
utp_link_t link; // also wtf
};
*/

template<typename K, typename T> class utpHashTable {
utp_hash_t *hash;
public:
static uint compare(const void *k1, const void *k2, size_t /*ks*/) {
return *((K*)k1) == *((K*)k2);
}
static uint32 compute_hash(const void *k, size_t /*ks*/) {
return ((K*)k)->compute_hash();
}
void Init() { hash = NULL; }
bool Allocated() { return (hash != NULL); }
void Free() { utp_hash_free_mem(hash); hash = NULL; }
void Create(int N, int initial) { hash = utp_hash_create(N, sizeof(K), sizeof(T), initial, &compute_hash, &compare); }
T *Lookup(const K &key) { return (T*)utp_hash_lookup(hash, &key); }
T *Add(const K &key) { return (T*)utp_hash_add(&hash, &key); }
T *Delete(const K &key) { return (T*)utp_hash_del(hash, &key); }
T *Iterate(utp_hash_iterator_t &iterator) { return (T*)utp_hash_iterate(hash, &iterator); }
size_t GetCount() { return hash->count; }
};

#endif //__UTP_HASH_H__
Loading