@ -4,10 +4,12 @@
* * Implementation is heavily inspired by Lua ' s ltable . c . * * Implementation is heavily inspired by Lua ' s ltable . c .
*/ */
# include "upb/table.int.h"
# include <string.h> # include <string.h>
# include "third_party/wyhash/wyhash.h"
# include "upb/table.int.h"
/* Must be last. */
# include "upb/port_def.inc" # include "upb/port_def.inc"
# define UPB_MAXARRSIZE 16 /* 64k. */ # define UPB_MAXARRSIZE 16 /* 64k. */
@ -87,11 +89,7 @@ static upb_tabent *mutable_entries(upb_table *t) {
} }
static bool isfull ( upb_table * t ) { static bool isfull ( upb_table * t ) {
if ( upb_table_size ( t ) = = 0 ) { return t - > count = = t - > max_count ;
return true ;
} else {
return ( ( double ) ( t - > count + 1 ) / upb_table_size ( t ) ) > MAX_LOAD ;
}
} }
static bool init ( upb_table * t , uint8_t size_lg2 , upb_alloc * a ) { static bool init ( upb_table * t , uint8_t size_lg2 , upb_alloc * a ) {
@ -100,6 +98,7 @@ static bool init(upb_table *t, uint8_t size_lg2, upb_alloc *a) {
t - > count = 0 ; t - > count = 0 ;
t - > size_lg2 = size_lg2 ; t - > size_lg2 = size_lg2 ;
t - > mask = upb_table_size ( t ) ? upb_table_size ( t ) - 1 : 0 ; t - > mask = upb_table_size ( t ) ? upb_table_size ( t ) - 1 : 0 ;
t - > max_count = upb_table_size ( t ) * MAX_LOAD ;
bytes = upb_table_size ( t ) * sizeof ( upb_tabent ) ; bytes = upb_table_size ( t ) * sizeof ( upb_tabent ) ;
if ( bytes > 0 ) { if ( bytes > 0 ) {
t - > entries = upb_malloc ( a , bytes ) ; t - > entries = upb_malloc ( a , bytes ) ;
@ -115,9 +114,17 @@ static void uninit(upb_table *t, upb_alloc *a) {
upb_free ( a , mutable_entries ( t ) ) ; upb_free ( a , mutable_entries ( t ) ) ;
} }
static upb_tabent * emptyent ( upb_table * t ) { static upb_tabent * emptyent ( upb_table * t , upb_tabent * e ) {
upb_tabent * e = mutable_entries ( t ) + upb_table_size ( t ) ; upb_tabent * begin = mutable_entries ( t ) ;
while ( 1 ) { if ( upb_tabent_isempty ( - - e ) ) return e ; UPB_ASSERT ( e > t - > entries ) ; } upb_tabent * end = begin + upb_table_size ( t ) ;
for ( e = e + 1 ; e < end ; e + + ) {
if ( upb_tabent_isempty ( e ) ) return e ;
}
for ( e = begin ; e < end ; e + + ) {
if ( upb_tabent_isempty ( e ) ) return e ;
}
UPB_ASSERT ( false ) ;
return NULL ;
} }
static upb_tabent * getentry_mutable ( upb_table * t , uint32_t hash ) { static upb_tabent * getentry_mutable ( upb_table * t , uint32_t hash ) {
@ -173,7 +180,7 @@ static void insert(upb_table *t, lookupkey_t key, upb_tabkey tabkey,
our_e - > next = NULL ; our_e - > next = NULL ;
} else { } else {
/* Collision. */ /* Collision. */
upb_tabent * new_e = emptyent ( t ) ; upb_tabent * new_e = emptyent ( t , mainpos_e ) ;
/* Head of collider's chain. */ /* Head of collider's chain. */
upb_tabent * chain = getentry_mutable ( t , hashfunc ( mainpos_e - > key ) ) ; upb_tabent * chain = getentry_mutable ( t , hashfunc ( mainpos_e - > key ) ) ;
if ( chain = = mainpos_e ) { if ( chain = = mainpos_e ) {
@ -268,10 +275,14 @@ static upb_tabkey strcopy(lookupkey_t k2, upb_alloc *a) {
return ( uintptr_t ) str ; return ( uintptr_t ) str ;
} }
static uint32_t table_hash ( const char * p , size_t n ) {
return wyhash ( p , n , 0 , _wyp ) ;
}
static uint32_t strhash ( upb_tabkey key ) { static uint32_t strhash ( upb_tabkey key ) {
uint32_t len ; uint32_t len ;
char * str = upb_tabstr ( key , & len ) ; char * str = upb_tabstr ( key , & len ) ;
return upb_murmur_hash2 ( str , len , 0 ) ; return table_hash ( str , len ) ;
} }
static bool streql ( upb_tabkey k1 , lookupkey_t k2 ) { static bool streql ( upb_tabkey k1 , lookupkey_t k2 ) {
@ -280,9 +291,14 @@ static bool streql(upb_tabkey k1, lookupkey_t k2) {
return len = = k2 . str . len & & ( len = = 0 | | memcmp ( str , k2 . str . str , len ) = = 0 ) ; return len = = k2 . str . len & & ( len = = 0 | | memcmp ( str , k2 . str . str , len ) = = 0 ) ;
} }
bool upb_strtable_init2 ( upb_strtable * t , upb_ctype_t ctype , upb_alloc * a ) { bool upb_strtable_init2 ( upb_strtable * t , upb_ctype_t ctype ,
size_t expected_size , upb_alloc * a ) {
UPB_UNUSED ( ctype ) ; /* TODO(haberman): rm */ UPB_UNUSED ( ctype ) ; /* TODO(haberman): rm */
return init ( & t - > t , 2 , a ) ; // Multiply by approximate reciprocal of MAX_LOAD (0.85), with pow2 denominator.
size_t need_entries = ( expected_size + 1 ) * 1204 / 1024 ;
UPB_ASSERT ( need_entries > = expected_size * 0.85 ) ;
int size_lg2 = _upb_lg2ceil ( need_entries ) ;
return init ( & t - > t , size_lg2 , a ) ;
} }
void upb_strtable_clear ( upb_strtable * t ) { void upb_strtable_clear ( upb_strtable * t ) {
@ -333,20 +349,20 @@ bool upb_strtable_insert3(upb_strtable *t, const char *k, size_t len,
tabkey = strcopy ( key , a ) ; tabkey = strcopy ( key , a ) ;
if ( tabkey = = 0 ) return false ; if ( tabkey = = 0 ) return false ;
hash = upb_murmur_hash2 ( key . str . str , key . str . len , 0 ) ; hash = table_hash ( key . str . str , key . str . len ) ;
insert ( & t - > t , key , tabkey , v , hash , & strhash , & streql ) ; insert ( & t - > t , key , tabkey , v , hash , & strhash , & streql ) ;
return true ; return true ;
} }
bool upb_strtable_lookup2 ( const upb_strtable * t , const char * key , size_t len , bool upb_strtable_lookup2 ( const upb_strtable * t , const char * key , size_t len ,
upb_value * v ) { upb_value * v ) {
uint32_t hash = upb_murmur_hash2 ( key , len , 0 ) ; uint32_t hash = table_hash ( key , len ) ;
return lookup ( & t - > t , strkey2 ( key , len ) , v , hash , & streql ) ; return lookup ( & t - > t , strkey2 ( key , len ) , v , hash , & streql ) ;
} }
bool upb_strtable_remove3 ( upb_strtable * t , const char * key , size_t len , bool upb_strtable_remove3 ( upb_strtable * t , const char * key , size_t len ,
upb_value * val , upb_alloc * alloc ) { upb_value * val , upb_alloc * alloc ) {
uint32_t hash = upb_murmur_hash2 ( key , len , 0 ) ; uint32_t hash = table_hash ( key , len ) ;
upb_tabkey tabkey ; upb_tabkey tabkey ;
if ( rm ( & t - > t , strkey2 ( key , len ) , val , & tabkey , hash , & streql ) ) { if ( rm ( & t - > t , strkey2 ( key , len ) , val , & tabkey , hash , & streql ) ) {
if ( alloc ) { if ( alloc ) {
@ -699,182 +715,3 @@ bool upb_inttable_iter_isequal(const upb_inttable_iter *i1,
return i1 - > t = = i2 - > t & & i1 - > index = = i2 - > index & & return i1 - > t = = i2 - > t & & i1 - > index = = i2 - > index & &
i1 - > array_part = = i2 - > array_part ; i1 - > array_part = = i2 - > array_part ;
} }
# if defined(UPB_UNALIGNED_READS_OK) || defined(__s390x__)
/* -----------------------------------------------------------------------------
* MurmurHash2 , by Austin Appleby ( released as public domain ) .
* Reformatted and C99 - ified by Joshua Haberman .
* Note - This code makes a few assumptions about how your machine behaves -
* 1. We can read a 4 - byte value from any address without crashing
* 2. sizeof ( int ) = = 4 ( in upb this limitation is removed by using uint32_t
* And it has a few limitations -
* 1. It will not work incrementally .
* 2. It will not produce the same results on little - endian and big - endian
* machines . */
uint32_t upb_murmur_hash2 ( const void * key , size_t len , uint32_t seed ) {
/* 'm' and 'r' are mixing constants generated offline.
* They ' re not really ' magic ' , they just happen to work well . */
const uint32_t m = 0x5bd1e995 ;
const int32_t r = 24 ;
/* Initialize the hash to a 'random' value */
uint32_t h = seed ^ len ;
/* Mix 4 bytes at a time into the hash */
const uint8_t * data = ( const uint8_t * ) key ;
while ( len > = 4 ) {
uint32_t k ;
memcpy ( & k , data , sizeof ( k ) ) ;
k * = m ;
k ^ = k > > r ;
k * = m ;
h * = m ;
h ^ = k ;
data + = 4 ;
len - = 4 ;
}
/* Handle the last few bytes of the input array */
switch ( len ) {
case 3 : h ^ = data [ 2 ] < < 16 ;
case 2 : h ^ = data [ 1 ] < < 8 ;
case 1 : h ^ = data [ 0 ] ; h * = m ;
} ;
/* Do a few final mixes of the hash to ensure the last few
* bytes are well - incorporated . */
h ^ = h > > 13 ;
h * = m ;
h ^ = h > > 15 ;
return h ;
}
# else /* !UPB_UNALIGNED_READS_OK */
/* -----------------------------------------------------------------------------
* MurmurHashAligned2 , by Austin Appleby
* Same algorithm as MurmurHash2 , but only does aligned reads - should be safer
* on certain platforms .
* Performance will be lower than MurmurHash2 */
# define MIX(h,k,m) { k *= m; k ^= k >> r; k *= m; h *= m; h ^= k; }
uint32_t upb_murmur_hash2 ( const void * key , size_t len , uint32_t seed ) {
const uint32_t m = 0x5bd1e995 ;
const int32_t r = 24 ;
const uint8_t * data = ( const uint8_t * ) key ;
uint32_t h = ( uint32_t ) ( seed ^ len ) ;
uint8_t align = ( uintptr_t ) data & 3 ;
if ( align & & ( len > = 4 ) ) {
/* Pre-load the temp registers */
uint32_t t = 0 , d = 0 ;
int32_t sl ;
int32_t sr ;
switch ( align ) {
case 1 : t | = data [ 2 ] < < 16 ; /* fallthrough */
case 2 : t | = data [ 1 ] < < 8 ; /* fallthrough */
case 3 : t | = data [ 0 ] ;
}
t < < = ( 8 * align ) ;
data + = 4 - align ;
len - = 4 - align ;
sl = 8 * ( 4 - align ) ;
sr = 8 * align ;
/* Mix */
while ( len > = 4 ) {
uint32_t k ;
d = * ( uint32_t * ) data ;
t = ( t > > sr ) | ( d < < sl ) ;
k = t ;
MIX ( h , k , m ) ;
t = d ;
data + = 4 ;
len - = 4 ;
}
/* Handle leftover data in temp registers */
d = 0 ;
if ( len > = align ) {
uint32_t k ;
switch ( align ) {
case 3 : d | = data [ 2 ] < < 16 ; /* fallthrough */
case 2 : d | = data [ 1 ] < < 8 ; /* fallthrough */
case 1 : d | = data [ 0 ] ; /* fallthrough */
}
k = ( t > > sr ) | ( d < < sl ) ;
MIX ( h , k , m ) ;
data + = align ;
len - = align ;
/* ----------
* Handle tail bytes */
switch ( len ) {
case 3 : h ^ = data [ 2 ] < < 16 ; /* fallthrough */
case 2 : h ^ = data [ 1 ] < < 8 ; /* fallthrough */
case 1 : h ^ = data [ 0 ] ; h * = m ; /* fallthrough */
} ;
} else {
switch ( len ) {
case 3 : d | = data [ 2 ] < < 16 ; /* fallthrough */
case 2 : d | = data [ 1 ] < < 8 ; /* fallthrough */
case 1 : d | = data [ 0 ] ; /* fallthrough */
case 0 : h ^ = ( t > > sr ) | ( d < < sl ) ; h * = m ;
}
}
h ^ = h > > 13 ;
h * = m ;
h ^ = h > > 15 ;
return h ;
} else {
while ( len > = 4 ) {
uint32_t k = * ( uint32_t * ) data ;
MIX ( h , k , m ) ;
data + = 4 ;
len - = 4 ;
}
/* ----------
* Handle tail bytes */
switch ( len ) {
case 3 : h ^ = data [ 2 ] < < 16 ; /* fallthrough */
case 2 : h ^ = data [ 1 ] < < 8 ; /* fallthrough */
case 1 : h ^ = data [ 0 ] ; h * = m ;
} ;
h ^ = h > > 13 ;
h * = m ;
h ^ = h > > 15 ;
return h ;
}
}
# undef MIX
# endif /* UPB_UNALIGNED_READS_OK */
Joshua Haberman
4 years ago