Browse Source

switched to Pearson B. Hashing (#570)

* switched to Pearson B. Hashing

* switched to Pearson B. Hashing
pull/572/head
Logan oos Even 4 years ago
committed by GitHub
parent
commit
808920089d
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
  1. 11
      include/pearson.h
  2. 506
      src/pearson.c

11
include/pearson.h

@ -1,5 +1,5 @@
/** /**
* (C) 2007-20 - ntop.org and contributors * (C) 2007-21 - ntop.org and contributors
* *
* This program is free software; you can redistribute it and/or modify * This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by * it under the terms of the GNU General Public License as published by
@ -20,14 +20,7 @@
#include <stddef.h> #include <stddef.h>
#include <stdint.h> #include <stdint.h>
#include "portable_endian.h"
#if defined (__SSSE3__) && defined (__AES__) // AES-NI & SSSE3 ----------------------------------------------------
#include <immintrin.h>
#endif // AES-NI & SSSE3 ------------------------------------------------------------------------------------------
void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len); void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len);

506
src/pearson.c

@ -1,5 +1,5 @@
/** /**
* (C) 2007-20 - ntop.org and contributors * (C) 2007-21 - ntop.org and contributors
* *
* This program is free software; you can redistribute it and/or modify * This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by * it under the terms of the GNU General Public License as published by
@ -17,418 +17,208 @@
*/ */
// taken from https://github.com/Logan007/pearson // taken from https://github.com/Logan007/pearsonB
// This is free and unencumbered software released into the public domain. // this is free and unencumbered software released into the public domain
#include "pearson.h" #include "pearson.h"
// AES S-Box table -- allows for eventually supported hardware accelerated look-up // Christopher Wellons' triple32 from https://github.com/skeeto/hash-prospector
static const uint8_t t[256] = { // published under The Unlicense
0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76, #define permute32(in) \
0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0, in ^= in >> 17; \
0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15, in *= 0xed5ad4bb; \
0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75, in ^= in >> 11; \
0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84, in *= 0xac4c1b51; \
0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf, in ^= in >> 15; \
0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8, in *= 0x31848bab; \
0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2, in ^= in >> 14
0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16 };
/*
// table as in original paper "Fast Hashing of Variable-Length Text Strings" by Peter K. Pearson
// as published in The Communications of the ACM Vol.33, No. 6 (June 1990), pp. 677-680.
static const uint8_t t[256] = {
0x01, 0x57, 0x31, 0x0c, 0xb0, 0xb2, 0x66, 0xa6, 0x79, 0xc1, 0x06, 0x54, 0xf9, 0xe6, 0x2c, 0xa3,
0x0e, 0xc5, 0xd5, 0xb5, 0xa1, 0x55, 0xda, 0x50, 0x40, 0xef, 0x18, 0xe2, 0xec, 0x8e, 0x26, 0xc8,
0x6e, 0xb1, 0x68, 0x67, 0x8d, 0xfd, 0xff, 0x32, 0x4d, 0x65, 0x51, 0x12, 0x2d, 0x60, 0x1f, 0xde,
0x19, 0x6b, 0xbe, 0x46, 0x56, 0xed, 0xf0, 0x22, 0x48, 0xf2, 0x14, 0xd6, 0xf4, 0xe3, 0x95, 0xeb,
0x61, 0xea, 0x39, 0x16, 0x3c, 0xfa, 0x52, 0xaf, 0xd0, 0x05, 0x7f, 0xc7, 0x6f, 0x3e, 0x87, 0xf8,
0xae, 0xa9, 0xd3, 0x3a, 0x42, 0x9a, 0x6a, 0xc3, 0xf5, 0xab, 0x11, 0xbb, 0xb6, 0xb3, 0x00, 0xf3,
0x84, 0x38, 0x94, 0x4b, 0x80, 0x85, 0x9e, 0x64, 0x82, 0x7e, 0x5b, 0x0d, 0x99, 0xf6, 0xd8, 0xdb,
0x77, 0x44, 0xdf, 0x4e, 0x53, 0x58, 0xc9, 0x63, 0x7a, 0x0b, 0x5c, 0x20, 0x88, 0x72, 0x34, 0x0a,
0x8a, 0x1e, 0x30, 0xb7, 0x9c, 0x23, 0x3d, 0x1a, 0x8f, 0x4a, 0xfb, 0x5e, 0x81, 0xa2, 0x3f, 0x98,
0xaa, 0x07, 0x73, 0xa7, 0xf1, 0xce, 0x03, 0x96, 0x37, 0x3b, 0x97, 0xdc, 0x5a, 0x35, 0x17, 0x83,
0x7d, 0xad, 0x0f, 0xee, 0x4f, 0x5f, 0x59, 0x10, 0x69, 0x89, 0xe1, 0xe0, 0xd9, 0xa0, 0x25, 0x7b,
0x76, 0x49, 0x02, 0x9d, 0x2e, 0x74, 0x09, 0x91, 0x86, 0xe4, 0xcf, 0xd4, 0xca, 0xd7, 0x45, 0xe5,
0x1b, 0xbc, 0x43, 0x7c, 0xa8, 0xfc, 0x2a, 0x04, 0x1d, 0x6c, 0x15, 0xf7, 0x13, 0xcd, 0x27, 0xcb,
0xe9, 0x28, 0xba, 0x93, 0xc6, 0xc0, 0x9b, 0x21, 0xa4, 0xbf, 0x62, 0xcc, 0xa5, 0xb4, 0x75, 0x4c,
0x8c, 0x24, 0xd2, 0xac, 0x29, 0x36, 0x9f, 0x08, 0xb9, 0xe8, 0x71, 0xc4, 0xe7, 0x2f, 0x92, 0x78,
0x33, 0x41, 0x1c, 0x90, 0xfe, 0xdd, 0x5d, 0xbd, 0xc2, 0x8b, 0x70, 0x2b, 0x47, 0x6d, 0xb8, 0xd1 };
*/
#if defined (__AES__) && defined (__SSSE3__) // AES-NI & SSSE3 ----------------------------------------------------
// David Stafford's Mix13 from http://zimbry.blogspot.com/2011/09/better-bit-mixing-improving-on.html
// the author clarified via eMail that this of his work is released to the public domain
#define permute64(in) \
in ^= (in >> 30); \
in *= 0xbf58476d1ce4e5b9; \
in ^= (in >> 27); \
in *= 0x94d049bb133111eb; \
in ^= (in >> 31)
void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) { #define dec1(in) \
in--
size_t i;
uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 };
uint64_t upper_hash_mask = *(uint64_t*)&upper;
uint64_t lower_hash_mask = *(uint64_t*)&lower;
__m128i tmp = _mm_set1_epi8(0x10);
__m128i hash_mask = _mm_set_epi64((__m64)lower_hash_mask, (__m64)upper_hash_mask);
__m128i high_hash_mask = _mm_xor_si128(tmp, hash_mask);
__m128i hash= _mm_setzero_si128();
__m128i high_hash= _mm_setzero_si128();
__m128i ZERO = _mm_setzero_si128();
__m128i ISOLATE_SBOX_MASK = _mm_set_epi32(0x0306090C, 0x0F020508, 0x0B0E0104, 0x070A0D00);
for(i = 0; i < len; i++) {
// broadcast the character
__m128i cc = _mm_set1_epi8(in[i]);
// xor into hash
hash = _mm_xor_si128(hash, cc);
high_hash = _mm_xor_si128(high_hash, cc);
// make them different permutations
hash = _mm_xor_si128(hash, hash_mask);
high_hash = _mm_xor_si128(high_hash, high_hash_mask);
// table lookup
hash = _mm_shuffle_epi8(hash, ISOLATE_SBOX_MASK); // re-order along AES round
high_hash = _mm_shuffle_epi8(high_hash, ISOLATE_SBOX_MASK); // re-order along AES round
hash = _mm_aesenclast_si128(hash, ZERO);
high_hash = _mm_aesenclast_si128(high_hash, ZERO);
}
// store output
_mm_store_si128((__m128i*)out , high_hash);
_mm_store_si128((__m128i*)&out[16] , hash);
}
void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) { #define dec2(in) \
dec1(in); \
dec1(in)
size_t i; #define dec3(in) \
dec2(in); \
dec1(in)
uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 }; #define dec4(in) \
uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 }; dec3(in); \
dec1(in)
uint64_t upper_hash_mask = *(uint64_t*)&upper; #define hash_round(hash, in, part) \
uint64_t lower_hash_mask = *(uint64_t*)&lower; hash##part ^= in; \
dec##part(hash##part); \
permute64(hash##part)
__m128i hash_mask = _mm_set_epi64((__m64)lower_hash_mask, (__m64)upper_hash_mask);
__m128i hash = _mm_setzero_si128();
__m128i ZERO = _mm_setzero_si128(); void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
__m128i ISOLATE_SBOX_MASK = _mm_set_epi32(0x0306090C, 0x0F020508, 0x0B0E0104, 0x070A0D00);
for(i = 0; i < len; i++) { uint64_t *current;
// broadcast the character current = (uint64_t*)in;
__m128i cc = _mm_set1_epi8(in[i]); uint64_t org_len = len;
// xor into hash uint64_t hash1 = 0;
hash = _mm_xor_si128(hash, cc); uint64_t hash2 = 0;
// make them different permutations uint64_t hash3 = 0;
hash = _mm_xor_si128(hash, hash_mask); uint64_t hash4 = 0;
// table lookup
hash = _mm_shuffle_epi8(hash, ISOLATE_SBOX_MASK); // re-order along AES round while (len > 7) {
hash = _mm_aesenclast_si128(hash, ZERO); // digest words little endian first
hash_round(hash, le64toh(*current), 1);
hash_round(hash, le64toh(*current), 2);
hash_round(hash, le64toh(*current), 3);
hash_round(hash, le64toh(*current), 4);
current++;
len-=8;
} }
// store output // handle the rest
_mm_store_si128((__m128i*)out , hash); hash1 = ~hash1;
} hash2 = ~hash2;
hash3 = ~hash3;
hash4 = ~hash4;
uint64_t pearson_hash_64 (const uint8_t *in, size_t len) {
while(len) {
size_t i; // byte-wise, no endianess
hash_round(hash, *(uint8_t*)current, 1);
__m128i hash_mask = _mm_cvtsi64_si128(0x0706050403020100); hash_round(hash, *(uint8_t*)current, 2);
hash_round(hash, *(uint8_t*)current, 3);
__m128i hash = _mm_setzero_si128(); hash_round(hash, *(uint8_t*)current, 4);
__m128i ZERO = _mm_setzero_si128(); current = (uint64_t*)((uint8_t*)current + 1);
__m128i ISOLATE_SBOX_MASK = _mm_set_epi32(0x0306090C, 0x0F020508, 0x0B0E0104, 0x070A0D00); len--;
for(i = 0; i < len; i++) {
// broadcast the character
__m128i cc = _mm_set1_epi8(in[i]);
// xor into hash
hash = _mm_xor_si128(hash, cc);
// make them different permutations
hash = _mm_xor_si128(hash, hash_mask);
// table lookup
hash = _mm_shuffle_epi8(hash, ISOLATE_SBOX_MASK); // re-order along AES round
hash = _mm_aesenclast_si128(hash, ZERO);
} }
// return lower 64 bits // digest length
return _mm_cvtsi128_si64(hash); hash1 = ~hash1;
} hash2 = ~hash2;
hash3 = ~hash3;
hash4 = ~hash4;
uint32_t pearson_hash_32 (const uint8_t *in, size_t len) { hash_round(hash, org_len, 1);
hash_round(hash, org_len, 2);
hash_round(hash, org_len, 3);
hash_round(hash, org_len, 4);
// return lower 32 bits (type casted) // hash string is stored big endian, the natural way to read
return pearson_hash_64(in, len); uint64_t *o;
o = (uint64_t*)out;
*o = htobe64(hash4);
o++;
*o = htobe64(hash3);
o++;
*o = htobe64(hash2);
o++;
*o = htobe64(hash1);
} }
uint16_t pearson_hash_16 (const uint8_t *in, size_t len) { void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
// return lower 16 bits (type casted)
return pearson_hash_64(in, len);
}
uint64_t *current;
current = (uint64_t*)in;
uint64_t org_len = len;
uint64_t hash1 = 0;
uint64_t hash2 = 0;
#else // plain C -------------------------------------------------------------------------------------------------- while (len > 7) {
// digest words little endian first
hash_round(hash, le64toh(*current), 1);
hash_round(hash, le64toh(*current), 2);
current++;
len-=8;
}
static uint16_t t16[65536]; // 16-bit look-up table // handle the rest
hash1 = ~hash1;
hash2 = ~hash2;
#define ROR64(x,r) (((x)>>(r))|((x)<<(64-(r)))) while(len) {
#define ROR32(x,r) (((x)>>(r))|((x)<<(32-(r)))) // byte-wise, no endianess
hash_round(hash, *(uint8_t*)current, 1);
hash_round(hash, *(uint8_t*)current, 2);
current = (uint64_t*)((uint8_t*)current + 1);
len--;
}
void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) { // digest length
hash1 = ~hash1;
hash2 = ~hash2;
size_t i; hash_round(hash, org_len, 1);
/* initial values - astonishingly, assembling using SHIFTs and ORs (in register) hash_round(hash, org_len, 2);
* works faster on well pipelined CPUs than loading the 64-bit value from memory.
* however, there is one advantage to loading from memory: as we also store back to
* memory at the end, we do not need to care about endianess! */
uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 };
uint64_t upper_hash_mask = *(uint64_t*)&upper;
uint64_t lower_hash_mask = *(uint64_t*)&lower;
uint64_t high_upper_hash_mask = upper_hash_mask + 0x1010101010101010;
uint64_t high_lower_hash_mask = lower_hash_mask + 0x1010101010101010;
uint64_t upper_hash = 0;
uint64_t lower_hash = 0;
uint64_t high_upper_hash = 0;
uint64_t high_lower_hash = 0;
for(i = 0; i < len; i++) {
// broadcast the character
uint64_t c = (uint8_t)in[i];
c |= c << 8;
c |= c << 16;
c |= c << 32;
// xor into hash & make them different permutations
upper_hash ^= c ^ upper_hash_mask;
lower_hash ^= c ^ lower_hash_mask;
high_upper_hash ^= c ^ high_upper_hash_mask;
high_lower_hash ^= c ^ high_lower_hash_mask;
// table lookup
uint64_t h = 0;
uint16_t x;
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
upper_hash = h;
h = 0;
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
lower_hash = h;
h = 0;
x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
high_upper_hash = h;
h = 0;
x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
high_lower_hash = h;
}
// store output // hash string is stored big endian, the natural way to read
uint64_t *o; uint64_t *o;
o = (uint64_t*)&out[0]; o = (uint64_t*)out;
*o = high_upper_hash; *o = htobe64(hash2);
o = (uint64_t*)&out[8]; o++;
*o = high_lower_hash; *o = htobe64(hash1);
o = (uint64_t*)&out[16];
*o = upper_hash;
o = (uint64_t*)&out[24];
*o = lower_hash;
} }
void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) { uint64_t pearson_hash_64 (const uint8_t *in, size_t len) {
size_t i; uint64_t *current;
/* initial values - astonishingly, assembling using SHIFTs and ORs (in register) current = (uint64_t*)in;
* works faster on well pipelined CPUs than loading the 64-bit value from memory. uint64_t org_len = len;
* however, there is one advantage to loading from memory: as we also store back to uint64_t hash1 = 0;
* memory at the end, we do not need to care about endianess! */
uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 };
uint64_t upper_hash_mask = *(uint64_t*)&upper;
uint64_t lower_hash_mask = *(uint64_t*)&lower;
uint64_t upper_hash = 0;
uint64_t lower_hash = 0;
for(i = 0; i < len; i++) {
// broadcast the character
uint64_t c = (uint8_t)in[i];
c |= c << 8;
c |= c << 16;
c |= c << 32;
// xor into hash, make them different permutations
upper_hash ^= c ^ upper_hash_mask;
lower_hash ^= c ^ lower_hash_mask;
// table lookup
uint64_t h = 0;
uint16_t x;
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
upper_hash = h;
h = 0;
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
lower_hash = h;
}
// store output while(len > 7) {
uint64_t *o; // digest words little endian first
o = (uint64_t*)&out[0]; hash_round(hash, le64toh(*current), 1);
*o = upper_hash;
o = (uint64_t*)&out[8];
*o = lower_hash;
}
current++;
len-=8;
}
// 64-bit hash: the return value has to be interpreted as uint64_t and // handle the rest
// follows machine-specific endianess in memory hash1 = ~hash1;
uint64_t pearson_hash_64 (const uint8_t *in, size_t len) { while(len) {
// byte-wise, no endianess
hash_round(hash, *(uint8_t*)current, 1);
size_t i; current = (uint64_t*)((uint8_t*)current + 1);
uint64_t hash_mask = 0x0706050403020100; len--;
uint64_t hash = 0;
uint32_t h1 = 0;
uint32_t h2 = 0;
uint32_t hash1 = hash;
uint32_t hash2 = (hash >> 32);
uint8_t x = 0;
uint8_t y = 0;
for(i = 0; i < len; i++) {
// broadcast the character
uint64_t c = (uint8_t)in[i];
c |= c << 8;
c |= c << 16;
c |= c << 32;
// into hash, make them different permutations
hash ^= c ^ hash_mask;
// table lookup
x = hash1; x = t[x]; hash1 >>= 8; h1 = x; h1 = ROR32 (h1, 8);
x = hash1; x = t[x]; hash1 >>= 8; h1 |= x; h1 = ROR32 (h1, 8);
x = hash1; x = t[x]; hash1 >>= 8; h1 |= x; h1 = ROR32 (h1, 8);
x = hash1; x = t[x]; ; h1 |= x; h1 = ROR32 (h1, 8);
hash1 = h1;
y = hash2; y = t[y]; hash2 >>= 8; h2 = y; h2 = ROR32 (h2, 8);
y = hash2; y = t[y]; hash2 >>= 8; h2 |= y; h2 = ROR32 (h2, 8);
y = hash2; y = t[y]; hash2 >>= 8; h2 |= y; h2 = ROR32 (h2, 8);
y = hash2; y = t[y]; ; h2 |= y; h2 = ROR32 (h2, 8);
hash2 = h2;
hash = h1 | ((uint64_t)h2 << 32);
} }
// return value // digest length
return hash; hash1 = ~hash1;
hash_round(hash, org_len, 1);
// caller is responsible for storing it big endian to memory (if ever)
return hash1;
} }
// 32-bit hash: the return value has to be interpreted as uint32_t and
// follows machine-specific endianess in memory
uint32_t pearson_hash_32 (const uint8_t *in, size_t len) { uint32_t pearson_hash_32 (const uint8_t *in, size_t len) {
size_t i; return pearson_hash_64(in, len);
uint32_t hash = 0;
uint32_t hash_mask = 0x03020100;
for(i = 0; i < len; i++) {
// broadcast the character
uint32_t c = (uint8_t)in[i];
c |= c << 8;
c |= c << 16;
// xor into hash, make them different permutations
hash ^= c ^ hash_mask;
// table lookup
uint32_t h = 0;
uint8_t x;
x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
hash = h;
}
// return value
return hash;
} }
// 16-bit hash: the return value has to be interpreted as uint16_t and
// follows machine-specific endianess in memory
uint16_t pearson_hash_16 (const uint8_t *in, size_t len) { uint16_t pearson_hash_16 (const uint8_t *in, size_t len) {
size_t i; return pearson_hash_64(in, len);
uint16_t hash = 0;
uint16_t hash_mask = 0x0100;
for(i = 0; i < len; i++) {
// broadcast the character
uint16_t c = (uint8_t)in[i];
c |= c << 8;
// xor into hash, make them different permutations
hash ^= c ^ hash_mask;
// table lookup
hash = t[(uint8_t)hash] + (t[hash >> 8] << 8);
}
// return value
return hash;
} }
#endif // AES-NI & SSSE3, plain C --------------------------------------------------------- void pearson_hash_init(void) {
void pearson_hash_init () {
#if defined (__AES__) && (__SSSE3__)
// no initialization required for SSSE/AES-NI
#else
size_t i;
// lookup table for 16-bit lookups
for(i = 0; i < 65536; i++)
t16[i] = (t[i >> 8] << 8) + t[(uint8_t)i];
#endif
} }

Loading…
Cancel
Save