#define PERL_SEEN_HV_FUNC_H
#if !( 0 \
+ || defined(PERL_HASH_FUNC_SIPHASH) \
|| defined(PERL_HASH_FUNC_SDBM) \
|| defined(PERL_HASH_FUNC_DJB2) \
|| defined(PERL_HASH_FUNC_SUPERFAST) \
|| defined(PERL_HASH_FUNC_ONE_AT_A_TIME) \
|| defined(PERL_HASH_FUNC_ONE_AT_A_TIME_HARD) \
|| defined(PERL_HASH_FUNC_ONE_AT_A_TIME_OLD) \
+ || defined(PERL_HASH_FUNC_MURMUR_HASH_64A) \
+ || defined(PERL_HASH_FUNC_MURMUR_HASH_64B) \
)
-#ifdef HAS_QUAD
-#define PERL_HASH_FUNC_SIPHASH
-#else
#define PERL_HASH_FUNC_ONE_AT_A_TIME_HARD
#endif
-#endif
#if defined(PERL_HASH_FUNC_SIPHASH)
# define PERL_HASH_FUNC "SIPHASH_2_4"
# define PERL_HASH_SEED_BYTES 16
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_siphash_2_4(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_siphash_2_4((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_SUPERFAST)
# define PERL_HASH_FUNC "SUPERFAST"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_superfast(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_superfast((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_MURMUR3)
# define PERL_HASH_FUNC "MURMUR3"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_murmur3(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_murmur3((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_DJB2)
# define PERL_HASH_FUNC "DJB2"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_djb2(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_djb2((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_SDBM)
# define PERL_HASH_FUNC "SDBM"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_sdbm(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_sdbm((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME_HARD)
# define PERL_HASH_FUNC "ONE_AT_A_TIME_HARD"
# define PERL_HASH_SEED_BYTES 8
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_one_at_a_time_hard(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_one_at_a_time_hard((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME)
# define PERL_HASH_FUNC "ONE_AT_A_TIME"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_one_at_a_time(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_one_at_a_time((seed),(U8*)(str),(len))
#elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME_OLD)
# define PERL_HASH_FUNC "ONE_AT_A_TIME_OLD"
# define PERL_HASH_SEED_BYTES 4
-# define PERL_HASH(hash,str,len) (hash)= S_perl_hash_old_one_at_a_time(PERL_HASH_SEED,(U8*)(str),(len))
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_old_one_at_a_time((seed),(U8*)(str),(len))
+#elif defined(PERL_HASH_FUNC_MURMUR_HASH_64A)
+# define PERL_HASH_FUNC "MURMUR_HASH_64A"
+# define PERL_HASH_SEED_BYTES 8
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_murmur_hash_64a((seed),(U8*)(str),(len))
+#elif defined(PERL_HASH_FUNC_MURMUR_HASH_64B)
+# define PERL_HASH_FUNC "MURMUR_HASH_64B"
+# define PERL_HASH_SEED_BYTES 8
+# define PERL_HASH_WITH_SEED(seed,hash,str,len) (hash)= S_perl_hash_murmur_hash_64b((seed),(U8*)(str),(len))
#endif
-#ifndef PERL_HASH
+#ifndef PERL_HASH_WITH_SEED
#error "No hash function defined!"
#endif
#ifndef PERL_HASH_SEED_BYTES
# endif
#endif
+#define PERL_HASH(hash,str,len) PERL_HASH_WITH_SEED(PERL_HASH_SEED,hash,str,len)
+
/*-----------------------------------------------------------------------------
* Endianess, misalignment capabilities and util macros
*
*/
#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
- || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
+ || defined(_MSC_VER) || defined (__TURBOC__)
#define U8TO16_LE(d) (*((const U16 *) (d)))
#endif
#define UNALIGNED_SAFE
#endif
-/* Find best way to ROTL32 */
+#ifdef HAS_QUAD
+#ifndef U64TYPE
+/* This probably isn't going to work, but failing with a compiler error due to
+ lack of uint64_t is no worse than failing right now with an #error. */
+#define U64TYPE uint64_t
+#endif
+#endif
+
+/* Find best way to ROTL32/ROTL64 */
#if defined(_MSC_VER)
#include <stdlib.h> /* Microsoft put _rotl declaration in here */
#define ROTL32(x,r) _rotl(x,r)
+ #ifdef HAS_QUAD
+ #define ROTL64(x,r) _rotl64(x,r)
+ #endif
#else
/* gcc recognises this code and generates a rotate instruction for CPUs with one */
#define ROTL32(x,r) (((U32)x << r) | ((U32)x >> (32 - r)))
+ #ifdef HAS_QUAD
+ #define ROTL64(x,r) (((U64TYPE)x << r) | ((U64TYPE)x >> (64 - r)))
+ #endif
#endif
+#ifdef UV_IS_QUAD
+#define ROTL_UV(x,r) ROTL64(x,r)
+#else
+#define ROTL_UV(x,r) ROTL32(x,r)
+#endif
+
/* This is SipHash by Jean-Philippe Aumasson and Daniel J. Bernstein.
* The authors claim it is relatively secure compared to the alternatives
* and that performance wise it is a suitable hash for languages like Perl.
#ifdef HAS_QUAD
-#ifndef U64TYPE
-/* This probably isn't going to work, but failing with a compiler error due to
- lack of uint64_t is no worse than failing right now with an #error. */
-#define U64TYPE uint64_t
-#endif
-
-
-#define ROTL64(x,b) (U64TYPE)( ((x) << (b)) | ( (x) >> (64 - (b))) )
-
#define U8TO64_LE(p) \
(((U64TYPE)((p)[0]) ) | \
((U64TYPE)((p)[1]) << 8) | \
PERL_STATIC_INLINE U32
S_perl_hash_siphash_2_4(const unsigned char * const seed, const unsigned char *in, const STRLEN inlen) {
/* "somepseudorandomlygeneratedbytes" */
- U64TYPE v0 = 0x736f6d6570736575ULL;
- U64TYPE v1 = 0x646f72616e646f6dULL;
- U64TYPE v2 = 0x6c7967656e657261ULL;
- U64TYPE v3 = 0x7465646279746573ULL;
+ U64TYPE v0 = UINT64_C(0x736f6d6570736575);
+ U64TYPE v1 = UINT64_C(0x646f72616e646f6d);
+ U64TYPE v2 = UINT64_C(0x6c7967656e657261);
+ U64TYPE v3 = UINT64_C(0x7465646279746573);
U64TYPE b;
U64TYPE k0 = ((U64TYPE*)seed)[0];
/* This CPU does not handle unaligned word access */
/* Consume enough so that the next data byte is word aligned */
- int i = -(long)ptr & 3;
+ STRLEN i = -PTR2IV(ptr) & 3;
if(i && i <= len) {
MURMUR_DOBYTES(i, h1, carry, bytes_in_carry, ptr, len);
}
PERL_STATIC_INLINE U32
S_perl_hash_djb2(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
const unsigned char * const end = (const unsigned char *)str + len;
- U32 hash = *((U32*)seed + len);
+ U32 hash = *((U32*)seed) + len;
while (str < end) {
hash = ((hash << 5) + hash) + *str++;
}
PERL_STATIC_INLINE U32
S_perl_hash_sdbm(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
const unsigned char * const end = (const unsigned char *)str + len;
- U32 hash = *((U32*)seed + len);
+ U32 hash = *((U32*)seed) + len;
while (str < end) {
hash = (hash << 6) + (hash << 16) - hash + *str++;
}
return hash;
}
+/* - ONE_AT_A_TIME_HARD is the 5.17+ recommend ONE_AT_A_TIME algorithm
+ * - ONE_AT_A_TIME_OLD is the unmodified 5.16 and older algorithm
+ * - ONE_AT_A_TIME is a 5.17+ tweak of ONE_AT_A_TIME_OLD to
+ * prevent strings of only \0 but different lengths from colliding
+ *
+ * Security-wise, from best to worst,
+ * ONE_AT_A_TIME_HARD > ONE_AT_A_TIME > ONE_AT_A_TIME_OLD
+ * There is a big drop-off in security between ONE_AT_A_TIME_HARD and
+ * ONE_AT_A_TIME
+ * */
/* This is the "One-at-a-Time" algorithm by Bob Jenkins
* from requirements by Colin Plumb.
return (hash + (hash << 15));
}
+#ifdef PERL_HASH_FUNC_MURMUR_HASH_64A
+/* This code is from Austin Appleby and is in the public domain.
+ Altered by Yves Orton to match Perl's hash interface, and to
+ return a 32 bit hash.
+
+ Note uses unaligned 64 bit loads - will NOT work on machines with
+ strict alginment requirements.
+
+ Also this code may not be suitable for big-endian machines.
+*/
+
+/* a 64 bit hash where we only use the low 32 bits */
+PERL_STATIC_INLINE U32
+S_perl_hash_murmur_hash_64a (const unsigned char * const seed, const unsigned char *str, const STRLEN len)
+{
+ const U64TYPE m = 0xc6a4a7935bd1e995;
+ const int r = 47;
+ U64TYPE h = *((U64TYPE*)seed) ^ len;
+ const U64TYPE * data = (const U64TYPE *)str;
+ const U64TYPE * end = data + (len/8);
+ const unsigned char * data2;
+
+ while(data != end)
+ {
+ U64TYPE k = *data++;
+
+ k *= m;
+ k ^= k >> r;
+ k *= m;
+
+ h ^= k;
+ h *= m;
+ }
+
+ data2 = (const unsigned char *)data;
+
+ switch(len & 7)
+ {
+ case 7: h ^= (U64TYPE)(data2[6]) << 48; /* fallthrough */
+ case 6: h ^= (U64TYPE)(data2[5]) << 40; /* fallthrough */
+ case 5: h ^= (U64TYPE)(data2[4]) << 32; /* fallthrough */
+ case 4: h ^= (U64TYPE)(data2[3]) << 24; /* fallthrough */
+ case 3: h ^= (U64TYPE)(data2[2]) << 16; /* fallthrough */
+ case 2: h ^= (U64TYPE)(data2[1]) << 8; /* fallthrough */
+ case 1: h ^= (U64TYPE)(data2[0]); /* fallthrough */
+ h *= m;
+ };
+
+ h ^= h >> r;
+ h *= m;
+ h ^= h >> r;
+
+ /* was: return h; */
+ return h & 0xFFFFFFFF;
+}
+
+#endif
+
+#ifdef PERL_HASH_FUNC_MURMUR_HASH_64B
+/* This code is from Austin Appleby and is in the public domain.
+ Altered by Yves Orton to match Perl's hash interface and return
+ a 32 bit value
+
+ Note uses unaligned 32 bit loads - will NOT work on machines with
+ strict alginment requirements.
+
+ Also this code may not be suitable for big-endian machines.
+*/
+
+/* a 64-bit hash for 32-bit platforms where we only use the low 32 bits */
+PERL_STATIC_INLINE U32
+S_perl_hash_murmur_hash_64b (const unsigned char * const seed, const unsigned char *str, STRLEN len)
+{
+ const U32 m = 0x5bd1e995;
+ const int r = 24;
+
+ U32 h1 = ((U32 *)seed)[0] ^ len;
+ U32 h2 = ((U32 *)seed)[1];
+
+ const U32 * data = (const U32 *)str;
+
+ while(len >= 8)
+ {
+ U32 k1, k2;
+ k1 = *data++;
+ k1 *= m; k1 ^= k1 >> r; k1 *= m;
+ h1 *= m; h1 ^= k1;
+ len -= 4;
+
+ k2 = *data++;
+ k2 *= m; k2 ^= k2 >> r; k2 *= m;
+ h2 *= m; h2 ^= k2;
+ len -= 4;
+ }
+
+ if(len >= 4)
+ {
+ U32 k1 = *data++;
+ k1 *= m; k1 ^= k1 >> r; k1 *= m;
+ h1 *= m; h1 ^= k1;
+ len -= 4;
+ }
+
+ switch(len)
+ {
+ case 3: h2 ^= ((unsigned char*)data)[2] << 16; /* fallthrough */
+ case 2: h2 ^= ((unsigned char*)data)[1] << 8; /* fallthrough */
+ case 1: h2 ^= ((unsigned char*)data)[0]; /* fallthrough */
+ h2 *= m;
+ };
+
+ h1 ^= h2 >> 18; h1 *= m;
+ h2 ^= h1 >> 22; h2 *= m;
+ /*
+ The following code has been removed as it is unused
+ when only the low 32 bits are used. -- Yves
+
+ h1 ^= h2 >> 17; h1 *= m;
+
+ U64TYPE h = h1;
+
+ h = (h << 32) | h2;
+ */
+
+ return h2;
+}
+#endif
+
/* legacy - only mod_perl should be doing this. */
#ifdef PERL_HASH_INTERNAL_ACCESS
#define PERL_HASH_INTERNAL(hash,str,len) PERL_HASH(hash,str,len)