diff options
Diffstat (limited to 'source/blender/blenlib/BLI_hash.hh')
-rw-r--r-- | source/blender/blenlib/BLI_hash.hh | 101 |
1 files changed, 52 insertions, 49 deletions
diff --git a/source/blender/blenlib/BLI_hash.hh b/source/blender/blenlib/BLI_hash.hh index 57d5f7f9d8a..b14a4ca933c 100644 --- a/source/blender/blenlib/BLI_hash.hh +++ b/source/blender/blenlib/BLI_hash.hh @@ -23,7 +23,7 @@ * A specialization of `blender::DefaultHash<T>` provides a hash function for values of type T. * This hash function is used by default in hash table implementations in blenlib. * - * The actual hash function is in the `operator()` method of DefaultHash<T>. The following code + * The actual hash function is in the `operator()` method of `DefaultHash<T>`. The following code * computes the hash of some value using DefaultHash. * * T value = ...; @@ -32,43 +32,43 @@ * * Hash table implementations like blender::Set support heterogeneous key lookups. That means that * one can do a lookup with a key of type A in a hash table that stores keys of type B. This is - * commonly done when B is std::string, because the conversion from e.g. a StringRef to std::string - * can be costly and is unnecessary. To make this work, values of type A and B that compare equal - * have to have the same hash value. This is achieved by defining potentially multiple `operator()` - * in a specialization of DefaultHash. All those methods have to compute the same hash for values - * that compare equal. + * commonly done when B is std::string, because the conversion from e.g. a #StringRef to + * std::string can be costly and is unnecessary. To make this work, values of type A and B that + * compare equal have to have the same hash value. This is achieved by defining potentially + * multiple `operator()` in a specialization of #DefaultHash. All those methods have to compute the + * same hash for values that compare equal. * - * The computed hash is an unsigned 32 bit integer. Ideally, the hash function would generate + * The computed hash is an unsigned 64 bit integer. Ideally, the hash function would generate * uniformly random hash values for a set of keys. However, in many cases trivial hash functions * are faster and produce a good enough distribution. In general it is better when more information * is in the lower bits of the hash. By choosing a good probing strategy, the effects of a bad hash - * function are less noticable though. In this context a good probing strategy is one that takes + * function are less noticeable though. In this context a good probing strategy is one that takes * all bits of the hash into account eventually. One has to check on a case by case basis to see if * a better but more expensive or trivial hash function works better. * * There are three main ways to provide a hash table implementation with a custom hash function. * * - When you want to provide a default hash function for your own custom type: Add a `hash` - * member function to it. The function should return `uint32_t` and take no arguments. This - * method will be called by the default implementation of DefaultHash. It will automatically be + * member function to it. The function should return `uint64_t` and take no arguments. This + * method will be called by the default implementation of #DefaultHash. It will automatically be * used by hash table implementations. * * - When you want to provide a default hash function for a type that you cannot modify: Add a new - * specialization to the DefaultHash struct. This can be done by writing code like below in + * specialization to the #DefaultHash struct. This can be done by writing code like below in * either global or BLI namespace. * * template<> struct blender::DefaultHash<TheType> { - * uint32_t operator()(const TheType &value) const { + * uint64_t operator()(const TheType &value) const { * return ...; * } * }; * * - When you want to provide a different hash function for a type that already has a default hash * function: Implement a struct like the one below and pass it as template parameter to the hash - * table explicitely. + * table explicitly. * * struct MyCustomHash { - * uint32_t operator()(const TheType &value) const { + * uint64_t operator()(const TheType &value) const { * return ...; * } * }; @@ -86,22 +86,32 @@ namespace blender { /** - * If there is no other specialization of DefaultHash for a given type, try to call `hash()` on the - * value. If there is no such method, this will result in a compiler error. Usually that means that - * you have to implement a hash function using one of three strategies listed above. + * If there is no other specialization of #DefaultHash for a given type, try to call `hash()` on + * the value. If there is no such method, this will result in a compiler error. Usually that means + * that you have to implement a hash function using one of three strategies listed above. */ template<typename T> struct DefaultHash { - uint32_t operator()(const T &value) const + uint64_t operator()(const T &value) const { return value.hash(); } }; +/** + * Use the same hash function for const and non const variants of a type. + */ +template<typename T> struct DefaultHash<const T> { + uint64_t operator()(const T &value) const + { + return DefaultHash<T>{}(value); + } +}; + #define TRIVIAL_DEFAULT_INT_HASH(TYPE) \ template<> struct DefaultHash<TYPE> { \ - uint32_t operator()(TYPE value) const \ + uint64_t operator()(TYPE value) const \ { \ - return (uint32_t)value; \ + return (uint64_t)value; \ } \ } @@ -117,36 +127,29 @@ TRIVIAL_DEFAULT_INT_HASH(int16_t); TRIVIAL_DEFAULT_INT_HASH(uint16_t); TRIVIAL_DEFAULT_INT_HASH(int32_t); TRIVIAL_DEFAULT_INT_HASH(uint32_t); - -template<> struct DefaultHash<uint64_t> { - uint32_t operator()(uint64_t value) const - { - uint32_t low = (uint32_t)value; - uint32_t high = (uint32_t)(value >> 32); - return low ^ (high * 0x45d9f3b); - } -}; - -template<> struct DefaultHash<int64_t> { - uint32_t operator()(uint64_t value) const - { - return DefaultHash<uint64_t>{}((uint64_t)value); - } -}; +TRIVIAL_DEFAULT_INT_HASH(int64_t); +TRIVIAL_DEFAULT_INT_HASH(uint64_t); /** * One should try to avoid using floats as keys in hash tables, but sometimes it is convenient. */ template<> struct DefaultHash<float> { - uint32_t operator()(float value) const + uint64_t operator()(float value) const { return *(uint32_t *)&value; } }; -inline uint32_t hash_string(StringRef str) +template<> struct DefaultHash<bool> { + uint64_t operator()(bool value) const + { + return (uint64_t)(value != false) * 1298191; + } +}; + +inline uint64_t hash_string(StringRef str) { - uint32_t hash = 5381; + uint64_t hash = 5381; for (char c : str) { hash = hash * 33 + c; } @@ -155,24 +158,24 @@ inline uint32_t hash_string(StringRef str) template<> struct DefaultHash<std::string> { /** - * Take a StringRef as parameter to support heterogeneous lookups in hash table implementations + * Take a #StringRef as parameter to support heterogeneous lookups in hash table implementations * when std::string is used as key. */ - uint32_t operator()(StringRef value) const + uint64_t operator()(StringRef value) const { return hash_string(value); } }; template<> struct DefaultHash<StringRef> { - uint32_t operator()(StringRef value) const + uint64_t operator()(StringRef value) const { return hash_string(value); } }; template<> struct DefaultHash<StringRefNull> { - uint32_t operator()(StringRef value) const + uint64_t operator()(StringRef value) const { return hash_string(value); } @@ -182,26 +185,26 @@ template<> struct DefaultHash<StringRefNull> { * While we cannot guarantee that the lower 4 bits of a pointer are zero, it is often the case. */ template<typename T> struct DefaultHash<T *> { - uint32_t operator()(const T *value) const + uint64_t operator()(const T *value) const { uintptr_t ptr = (uintptr_t)value; - uint32_t hash = (uint32_t)(ptr >> 4); + uint64_t hash = (uint64_t)(ptr >> 4); return hash; } }; template<typename T> struct DefaultHash<std::unique_ptr<T>> { - uint32_t operator()(const std::unique_ptr<T> &value) const + uint64_t operator()(const std::unique_ptr<T> &value) const { return DefaultHash<T *>{}(value.get()); } }; template<typename T1, typename T2> struct DefaultHash<std::pair<T1, T2>> { - uint32_t operator()(const std::pair<T1, T2> &value) const + uint64_t operator()(const std::pair<T1, T2> &value) const { - uint32_t hash1 = DefaultHash<T1>{}(value.first); - uint32_t hash2 = DefaultHash<T2>{}(value.second); + uint64_t hash1 = DefaultHash<T1>{}(value.first); + uint64_t hash2 = DefaultHash<T2>{}(value.second); return hash1 ^ (hash2 * 33); } }; |