X-Git-Url: http://plrg.eecs.uci.edu/git/?p=folly.git;a=blobdiff_plain;f=folly%2FHash.h;h=a57bbbe05a65a63d2dd9822ce2de47ca12751d27;hp=a52da9c634e6ff0206b17e327b852896281ed786;hb=af7e70660f867873638ef2ce5bb33b19771bdd41;hpb=27494a20393fa45072e7d526d358835f3abe312a

diff --git a/folly/Hash.h b/folly/Hash.h
index a52da9c6..a57bbbe0 100644
--- a/folly/Hash.h
+++ b/folly/Hash.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2012 Facebook, Inc.
+ * Copyright 2014 Facebook, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,9 +17,14 @@
 #ifndef FOLLY_BASE_HASH_H_
 #define FOLLY_BASE_HASH_H_
 
-#include <stdint.h>
 #include <cstring>
+#include <stdint.h>
 #include <string>
+#include <utility>
+#include <tuple>
+
+#include "folly/SpookyHashV1.h"
+#include "folly/SpookyHashV2.h"
 
 /*
  * Various hashing functions.
@@ -27,6 +32,60 @@
 
 namespace folly { namespace hash {
 
+// This is a general-purpose way to create a single hash from multiple
+// hashable objects. hash_combine_generic takes a class Hasher implementing
+// hash<T>; hash_combine uses a default hasher StdHasher that uses std::hash.
+// hash_combine_generic hashes each argument and combines those hashes in
+// an order-dependent way to yield a new hash.
+
+
+// This is the Hash128to64 function from Google's cityhash (available
+// under the MIT License).  We use it to reduce multiple 64 bit hashes
+// into a single hash.
+inline size_t hash_128_to_64(const size_t upper, const size_t lower) {
+  // Murmur-inspired hashing.
+  const size_t kMul = 0x9ddfea08eb382d69ULL;
+  size_t a = (lower ^ upper) * kMul;
+  a ^= (a >> 47);
+  size_t b = (upper ^ a) * kMul;
+  b ^= (b >> 47);
+  b *= kMul;
+  return b;
+}
+
+// Never used, but gcc demands it.
+template <class Hasher>
+inline size_t hash_combine_generic() {
+  return 0;
+}
+
+template <class Hasher, typename T, typename... Ts>
+size_t hash_combine_generic(const T& t, const Ts&... ts) {
+  size_t seed = Hasher::hash(t);
+  if (sizeof...(ts) == 0) {
+    return seed;
+  }
+  size_t remainder = hash_combine_generic<Hasher>(ts...);
+  return hash_128_to_64(seed, remainder);
+}
+
+// Simply uses std::hash to hash.  Note that std::hash is not guaranteed
+// to be a very good hash function; provided std::hash doesn't collide on
+// the individual inputs, you are fine, but that won't be true for, say,
+// strings or pairs
+class StdHasher {
+ public:
+  template <typename T>
+  static size_t hash(const T& t) {
+    return std::hash<T>()(t);
+  }
+};
+
+template <typename T, typename... Ts>
+size_t hash_combine(const T& t, const Ts&... ts) {
+  return hash_combine_generic<StdHasher>(t, ts...);
+}
+
 //////////////////////////////////////////////////////////////////////
 
 /*
@@ -34,13 +93,32 @@ namespace folly { namespace hash {
  */
 
 inline uint64_t twang_mix64(uint64_t key) {
-  key = (~key) + (key << 21);
+  key = (~key) + (key << 21);  // key *= (1 << 21) - 1; key -= 1;
   key = key ^ (key >> 24);
-  key = (key + (key << 3)) + (key << 8);
+  key = key + (key << 3) + (key << 8);  // key *= 1 + (1 << 3) + (1 << 8)
   key = key ^ (key >> 14);
-  key = (key + (key << 2)) + (key << 4);
+  key = key + (key << 2) + (key << 4);  // key *= 1 + (1 << 2) + (1 << 4)
   key = key ^ (key >> 28);
-  key = key + (key << 31);
+  key = key + (key << 31);  // key *= 1 + (1 << 31)
+  return key;
+}
+
+/*
+ * Inverse of twang_mix64
+ *
+ * Note that twang_unmix64 is significantly slower than twang_mix64.
+ */
+
+inline uint64_t twang_unmix64(uint64_t key) {
+  // See the comments in jenkins_rev_unmix32 for an explanation as to how this
+  // was generated
+  key *= 4611686016279904257U;
+  key ^= (key >> 28) ^ (key >> 56);
+  key *= 14933078535860113213U;
+  key ^= (key >> 14) ^ (key >> 28) ^ (key >> 42) ^ (key >> 56);
+  key *= 15244667743933553977U;
+  key ^= (key >> 24) ^ (key >> 48);
+  key = (key + 1) * 9223367638806167551U;
   return key;
 }
 
@@ -63,23 +141,56 @@ inline uint32_t twang_32from64(uint64_t key) {
  */
 
 inline uint32_t jenkins_rev_mix32(uint32_t key) {
-  key += (key << 12);
+  key += (key << 12);  // key *= (1 + (1 << 12))
   key ^= (key >> 22);
-  key += (key << 4);
+  key += (key << 4);   // key *= (1 + (1 << 4))
   key ^= (key >> 9);
-  key += (key << 10);
+  key += (key << 10);  // key *= (1 + (1 << 10))
   key ^= (key >> 2);
+  // key *= (1 + (1 << 7)) * (1 + (1 << 12))
   key += (key << 7);
   key += (key << 12);
   return key;
 }
 
+/*
+ * Inverse of jenkins_rev_mix32
+ *
+ * Note that jenkinks_rev_unmix32 is significantly slower than
+ * jenkins_rev_mix32.
+ */
+
+inline uint32_t jenkins_rev_unmix32(uint32_t key) {
+  // These are the modular multiplicative inverses (in Z_2^32) of the
+  // multiplication factors in jenkins_rev_mix32, in reverse order.  They were
+  // computed using the Extended Euclidean algorithm, see
+  // http://en.wikipedia.org/wiki/Modular_multiplicative_inverse
+  key *= 2364026753U;
+
+  // The inverse of a ^= (a >> n) is
+  // b = a
+  // for (int i = n; i < 32; i += n) {
+  //   b ^= (a >> i);
+  // }
+  key ^=
+    (key >> 2) ^ (key >> 4) ^ (key >> 6) ^ (key >> 8) ^
+    (key >> 10) ^ (key >> 12) ^ (key >> 14) ^ (key >> 16) ^
+    (key >> 18) ^ (key >> 20) ^ (key >> 22) ^ (key >> 24) ^
+    (key >> 26) ^ (key >> 28) ^ (key >> 30);
+  key *= 3222273025U;
+  key ^= (key >> 9) ^ (key >> 18) ^ (key >> 27);
+  key *= 4042322161U;
+  key ^= (key >> 22);
+  key *= 16773121U;
+  return key;
+}
+
 /*
  * Fowler / Noll / Vo (FNV) Hash
  *     http://www.isthe.com/chongo/tech/comp/fnv/
  */
 
-const uint32_t FNV_32_HASH_START = 216613626UL;
+const uint32_t FNV_32_HASH_START = 2166136261UL;
 const uint64_t FNV_64_HASH_START = 14695981039346656037ULL;
 
 inline uint32_t fnv32(const char* s,
@@ -107,7 +218,7 @@ inline uint32_t fnv32_buf(const void* buf,
 }
 
 inline uint32_t fnv32(const std::string& str,
-                      uint64_t hash = FNV_32_HASH_START) {
+                      uint32_t hash = FNV_32_HASH_START) {
   return fnv32_buf(str.data(), str.size(), hash);
 }
 
@@ -238,6 +349,51 @@ template<> struct hasher<uint64_t> {
   }
 };
 
+// recursion
+template <size_t index, typename... Ts>
+struct TupleHasher {
+  size_t operator()(std::tuple<Ts...> const& key) const {
+    return hash::hash_combine(
+      TupleHasher<index - 1, Ts...>()(key),
+      std::get<index>(key));
+  }
+};
+
+// base
+template <typename... Ts>
+struct TupleHasher<0, Ts...> {
+  size_t operator()(std::tuple<Ts...> const& key) const {
+    // we could do std::hash here directly, but hash_combine hides all the
+    // ugly templating implicitly
+    return hash::hash_combine(std::get<0>(key));
+  }
+};
+
 } // namespace folly
 
+// Custom hash functions.
+namespace std {
+  // Hash function for pairs. Requires default hash functions for both
+  // items in the pair.
+  template <typename T1, typename T2>
+  class hash<std::pair<T1, T2> > {
+  public:
+    size_t operator()(const std::pair<T1, T2>& x) const {
+      return folly::hash::hash_combine(x.first, x.second);
+    }
+  };
+
+  // Hash function for tuples. Requires default hash functions for all types.
+  template <typename... Ts>
+  struct hash<std::tuple<Ts...>> {
+    size_t operator()(std::tuple<Ts...> const& key) const {
+      folly::TupleHasher<
+        std::tuple_size<std::tuple<Ts...>>::value - 1, // start index
+        Ts...> hasher;
+
+      return hasher(key);
+    }
+  };
+} // namespace std
+
 #endif