#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_HARD) \
|| defined(PERL_HASH_FUNC_ONE_AT_A_TIME_OLD) \
)
-#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"
*/
#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;
- if(i && (STRLEN)i <= len) {
+ 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.