From 43cdc2e037320354f67129215ff731b269153c2a Mon Sep 17 00:00:00 2001 From: Fabio Alessandrelli Date: Wed, 24 Aug 2022 14:28:44 +0200 Subject: [PATCH] Update hashfuncs, add some missing math funcs. --- include/godot_cpp/core/math.hpp | 16 ++ include/godot_cpp/templates/hashfuncs.hpp | 336 +++++++++++++++++----- 2 files changed, 281 insertions(+), 71 deletions(-) diff --git a/include/godot_cpp/core/math.hpp b/include/godot_cpp/core/math.hpp index 84f67b02..e0d93dbe 100644 --- a/include/godot_cpp/core/math.hpp +++ b/include/godot_cpp/core/math.hpp @@ -352,6 +352,22 @@ inline float range_lerp(float p_value, float p_istart, float p_istop, float p_os return Math::lerp(p_ostart, p_ostop, Math::inverse_lerp(p_istart, p_istop, p_value)); } +inline bool is_nan(float p_val) { + return std::isnan(p_val); +} + +inline bool is_nan(double p_val) { + return std::isnan(p_val); +} + +inline bool is_inf(float p_val) { + return std::isinf(p_val); +} + +inline bool is_inf(double p_val) { + return std::isinf(p_val); +} + inline bool is_equal_approx(real_t a, real_t b) { // Check for exact equality first, required to handle "infinity" values. if (a == b) { diff --git a/include/godot_cpp/templates/hashfuncs.hpp b/include/godot_cpp/templates/hashfuncs.hpp index 086b84ba..19dd9a6b 100644 --- a/include/godot_cpp/templates/hashfuncs.hpp +++ b/include/godot_cpp/templates/hashfuncs.hpp @@ -31,6 +31,11 @@ #ifndef HASHFUNCS_HPP #define HASHFUNCS_HPP +// Needed for fastmod. +#if defined(_MSC_VER) +#include +#endif + #include #include #include @@ -59,30 +64,30 @@ namespace godot { * @param C String * @return 32-bits hashcode */ -static inline uint32_t hash_djb2(const char *p_cstr) { +static _FORCE_INLINE_ uint32_t hash_djb2(const char *p_cstr) { const unsigned char *chr = (const unsigned char *)p_cstr; uint32_t hash = 5381; uint32_t c; while ((c = *chr++)) { - hash = ((hash << 5) + hash) + c; /* hash * 33 + c */ + hash = ((hash << 5) + hash) ^ c; /* hash * 33 ^ c */ } return hash; } -static inline uint32_t hash_djb2_buffer(const uint8_t *p_buff, int p_len, uint32_t p_prev = 5381) { +static _FORCE_INLINE_ uint32_t hash_djb2_buffer(const uint8_t *p_buff, int p_len, uint32_t p_prev = 5381) { uint32_t hash = p_prev; for (int i = 0; i < p_len; i++) { - hash = ((hash << 5) + hash) + p_buff[i]; /* hash * 33 + c */ + hash = ((hash << 5) + hash) ^ p_buff[i]; /* hash * 33 + c */ } return hash; } -static inline uint32_t hash_djb2_one_32(uint32_t p_in, uint32_t p_prev = 5381) { - return ((p_prev << 5) + p_prev) + p_in; +static _FORCE_INLINE_ uint32_t hash_djb2_one_32(uint32_t p_in, uint32_t p_prev = 5381) { + return ((p_prev << 5) + p_prev) ^ p_in; } /** @@ -92,7 +97,7 @@ static inline uint32_t hash_djb2_one_32(uint32_t p_in, uint32_t p_prev = 5381) { * @param p_int - 64-bit unsigned integer key to be hashed * @return unsigned 32-bit value representing hashcode */ -static inline uint32_t hash_one_uint64(const uint64_t p_int) { +static _FORCE_INLINE_ uint32_t hash_one_uint64(const uint64_t p_int) { uint64_t v = p_int; v = (~v) + (v << 18); // v = (v << 18) - v - 1; v = v ^ (v >> 31); @@ -103,7 +108,46 @@ static inline uint32_t hash_one_uint64(const uint64_t p_int) { return uint32_t(v); } -static inline uint32_t hash_djb2_one_float(double p_in, uint32_t p_prev = 5381) { +#define HASH_MURMUR3_SEED 0x7F07C65 +// Murmurhash3 32-bit version. +// All MurmurHash versions are public domain software, and the author disclaims all copyright to their code. + +static _FORCE_INLINE_ uint32_t hash_murmur3_one_32(uint32_t p_in, uint32_t p_seed = HASH_MURMUR3_SEED) { + p_in *= 0xcc9e2d51; + p_in = (p_in << 15) | (p_in >> 17); + p_in *= 0x1b873593; + + p_seed ^= p_in; + p_seed = (p_seed << 13) | (p_seed >> 19); + p_seed = p_seed * 5 + 0xe6546b64; + + return p_seed; +} + +static _FORCE_INLINE_ uint32_t hash_murmur3_one_float(float p_in, uint32_t p_seed = HASH_MURMUR3_SEED) { + union { + float f; + uint32_t i; + } u; + + // Normalize +/- 0.0 and NaN values so they hash the same. + if (p_in == 0.0f) { + u.f = 0.0; + } else if (Math::is_nan(p_in)) { + u.f = NAN; + } else { + u.f = p_in; + } + + return hash_murmur3_one_32(u.i, p_seed); +} + +static _FORCE_INLINE_ uint32_t hash_murmur3_one_64(uint64_t p_in, uint32_t p_seed = HASH_MURMUR3_SEED) { + p_seed = hash_murmur3_one_32(p_in & 0xFFFFFFFF, p_seed); + return hash_murmur3_one_32(p_in >> 32, p_seed); +} + +static _FORCE_INLINE_ uint32_t hash_murmur3_one_double(double p_in, uint32_t p_seed = HASH_MURMUR3_SEED) { union { double d; uint64_t i; @@ -112,7 +156,95 @@ static inline uint32_t hash_djb2_one_float(double p_in, uint32_t p_prev = 5381) // Normalize +/- 0.0 and NaN values so they hash the same. if (p_in == 0.0f) { u.d = 0.0; - } else if (std::isnan(p_in)) { + } else if (Math::is_nan(p_in)) { + u.d = NAN; + } else { + u.d = p_in; + } + + return hash_murmur3_one_64(u.i, p_seed); +} + +static _FORCE_INLINE_ uint32_t hash_murmur3_one_real(real_t p_in, uint32_t p_seed = HASH_MURMUR3_SEED) { +#ifdef REAL_T_IS_DOUBLE + return hash_murmur3_one_double(p_in, p_seed); +#else + return hash_murmur3_one_float(p_in, p_seed); +#endif +} + +static _FORCE_INLINE_ uint32_t hash_rotl32(uint32_t x, int8_t r) { + return (x << r) | (x >> (32 - r)); +} + +static _FORCE_INLINE_ uint32_t hash_fmix32(uint32_t h) { + h ^= h >> 16; + h *= 0x85ebca6b; + h ^= h >> 13; + h *= 0xc2b2ae35; + h ^= h >> 16; + + return h; +} + +static _FORCE_INLINE_ uint32_t hash_murmur3_buffer(const void *key, int length, const uint32_t seed = HASH_MURMUR3_SEED) { + // Although not required, this is a random prime number. + const uint8_t *data = (const uint8_t *)key; + const int nblocks = length / 4; + + uint32_t h1 = seed; + + const uint32_t c1 = 0xcc9e2d51; + const uint32_t c2 = 0x1b873593; + + const uint32_t *blocks = (const uint32_t *)(data + nblocks * 4); + + for (int i = -nblocks; i; i++) { + uint32_t k1 = blocks[i]; + + k1 *= c1; + k1 = hash_rotl32(k1, 15); + k1 *= c2; + + h1 ^= k1; + h1 = hash_rotl32(h1, 13); + h1 = h1 * 5 + 0xe6546b64; + } + + const uint8_t *tail = (const uint8_t *)(data + nblocks * 4); + + uint32_t k1 = 0; + + switch (length & 3) { + case 3: + k1 ^= tail[2] << 16; + [[fallthrough]]; + case 2: + k1 ^= tail[1] << 8; + [[fallthrough]]; + case 1: + k1 ^= tail[0]; + k1 *= c1; + k1 = hash_rotl32(k1, 15); + k1 *= c2; + h1 ^= k1; + }; + + // Finalize with additional bit mixing. + h1 ^= length; + return hash_fmix32(h1); +} + +static _FORCE_INLINE_ uint32_t hash_djb2_one_float(double p_in, uint32_t p_prev = 5381) { + union { + double d; + uint64_t i; + } u; + + // Normalize +/- 0.0 and NaN values so they hash the same. + if (p_in == 0.0f) { + u.d = 0.0; + } else if (Math::is_nan(p_in)) { u.d = NAN; } else { u.d = p_in; @@ -122,7 +254,7 @@ static inline uint32_t hash_djb2_one_float(double p_in, uint32_t p_prev = 5381) } template -static inline uint32_t make_uint32_t(T p_in) { +static _FORCE_INLINE_ uint32_t hash_make_uint32_t(T p_in) { union { T t; uint32_t _u32; @@ -132,7 +264,7 @@ static inline uint32_t make_uint32_t(T p_in) { return _u._u32; } -static inline uint64_t hash_djb2_one_float_64(double p_in, uint64_t p_prev = 5381) { +static _FORCE_INLINE_ uint64_t hash_djb2_one_float_64(double p_in, uint64_t p_prev = 5381) { union { double d; uint64_t i; @@ -141,7 +273,7 @@ static inline uint64_t hash_djb2_one_float_64(double p_in, uint64_t p_prev = 538 // Normalize +/- 0.0 and NaN values so they hash the same. if (p_in == 0.0f) { u.d = 0.0; - } else if (std::isnan(p_in)) { + } else if (Math::is_nan(p_in)) { u.d = NAN; } else { u.d = p_in; @@ -150,12 +282,12 @@ static inline uint64_t hash_djb2_one_float_64(double p_in, uint64_t p_prev = 538 return ((p_prev << 5) + p_prev) + u.i; } -static inline uint64_t hash_djb2_one_64(uint64_t p_in, uint64_t p_prev = 5381) { - return ((p_prev << 5) + p_prev) + p_in; +static _FORCE_INLINE_ uint64_t hash_djb2_one_64(uint64_t p_in, uint64_t p_prev = 5381) { + return ((p_prev << 5) + p_prev) ^ p_in; } template -static inline uint64_t make_uint64_t(T p_in) { +static _FORCE_INLINE_ uint64_t hash_make_uint64_t(T p_in) { union { T t; uint64_t _u64; @@ -170,42 +302,43 @@ template class Ref; struct HashMapHasherDefault { - static _FORCE_INLINE_ uint32_t hash(const String &p_string) { return p_string.hash(); } - static _FORCE_INLINE_ uint32_t hash(const char *p_cstr) { return hash_djb2(p_cstr); } - static _FORCE_INLINE_ uint32_t hash(const uint64_t p_int) { return hash_one_uint64(p_int); } - static _FORCE_INLINE_ uint32_t hash(const ObjectID &p_id) { return hash_one_uint64(p_id); } - - static _FORCE_INLINE_ uint32_t hash(const int64_t p_int) { return hash(uint64_t(p_int)); } - static _FORCE_INLINE_ uint32_t hash(const float p_float) { return hash_djb2_one_float(p_float); } - static _FORCE_INLINE_ uint32_t hash(const double p_double) { return hash_djb2_one_float(p_double); } - static _FORCE_INLINE_ uint32_t hash(const uint32_t p_int) { return p_int; } - static _FORCE_INLINE_ uint32_t hash(const int32_t p_int) { return (uint32_t)p_int; } - static _FORCE_INLINE_ uint32_t hash(const uint16_t p_int) { return p_int; } - static _FORCE_INLINE_ uint32_t hash(const int16_t p_int) { return (uint32_t)p_int; } - static _FORCE_INLINE_ uint32_t hash(const uint8_t p_int) { return p_int; } - static _FORCE_INLINE_ uint32_t hash(const int8_t p_int) { return (uint32_t)p_int; } - static _FORCE_INLINE_ uint32_t hash(const wchar_t p_wchar) { return (uint32_t)p_wchar; } - static _FORCE_INLINE_ uint32_t hash(const char16_t p_uchar) { return (uint32_t)p_uchar; } - static _FORCE_INLINE_ uint32_t hash(const char32_t p_uchar) { return (uint32_t)p_uchar; } - static _FORCE_INLINE_ uint32_t hash(const RID &p_rid) { return hash_one_uint64(p_rid.get_id()); } - - static _FORCE_INLINE_ uint32_t hash(const StringName &p_string_name) { return p_string_name.hash(); } - static _FORCE_INLINE_ uint32_t hash(const NodePath &p_path) { return p_path.hash(); } - + // Generic hash function for any type. template static _FORCE_INLINE_ uint32_t hash(const T *p_pointer) { return hash_one_uint64((uint64_t)p_pointer); } template static _FORCE_INLINE_ uint32_t hash(const Ref &p_ref) { return hash_one_uint64((uint64_t)p_ref.operator->()); } + static _FORCE_INLINE_ uint32_t hash(const String &p_string) { return p_string.hash(); } + static _FORCE_INLINE_ uint32_t hash(const char *p_cstr) { return hash_djb2(p_cstr); } + static _FORCE_INLINE_ uint32_t hash(const wchar_t p_wchar) { return hash_fmix32(p_wchar); } + static _FORCE_INLINE_ uint32_t hash(const char16_t p_uchar) { return hash_fmix32(p_uchar); } + static _FORCE_INLINE_ uint32_t hash(const char32_t p_uchar) { return hash_fmix32(p_uchar); } + static _FORCE_INLINE_ uint32_t hash(const RID &p_rid) { return hash_one_uint64(p_rid.get_id()); } + static _FORCE_INLINE_ uint32_t hash(const StringName &p_string_name) { return p_string_name.hash(); } + static _FORCE_INLINE_ uint32_t hash(const NodePath &p_path) { return p_path.hash(); } + static _FORCE_INLINE_ uint32_t hash(const ObjectID &p_id) { return hash_one_uint64(p_id); } + + static _FORCE_INLINE_ uint32_t hash(const uint64_t p_int) { return hash_one_uint64(p_int); } + static _FORCE_INLINE_ uint32_t hash(const int64_t p_int) { return hash_one_uint64(p_int); } + static _FORCE_INLINE_ uint32_t hash(const float p_float) { return hash_murmur3_one_float(p_float); } + static _FORCE_INLINE_ uint32_t hash(const double p_double) { return hash_murmur3_one_double(p_double); } + static _FORCE_INLINE_ uint32_t hash(const uint32_t p_int) { return hash_fmix32(p_int); } + static _FORCE_INLINE_ uint32_t hash(const int32_t p_int) { return hash_fmix32(p_int); } + static _FORCE_INLINE_ uint32_t hash(const uint16_t p_int) { return hash_fmix32(p_int); } + static _FORCE_INLINE_ uint32_t hash(const int16_t p_int) { return hash_fmix32(p_int); } + static _FORCE_INLINE_ uint32_t hash(const uint8_t p_int) { return hash_fmix32(p_int); } + static _FORCE_INLINE_ uint32_t hash(const int8_t p_int) { return hash_fmix32(p_int); } static _FORCE_INLINE_ uint32_t hash(const Vector2i &p_vec) { - uint32_t h = hash_djb2_one_32(p_vec.x); - return hash_djb2_one_32(p_vec.y, h); + uint32_t h = hash_murmur3_one_32(p_vec.x); + h = hash_murmur3_one_32(p_vec.y, h); + return hash_fmix32(h); } static _FORCE_INLINE_ uint32_t hash(const Vector3i &p_vec) { - uint32_t h = hash_djb2_one_32(p_vec.x); - h = hash_djb2_one_32(p_vec.y, h); - return hash_djb2_one_32(p_vec.z, h); + uint32_t h = hash_murmur3_one_32(p_vec.x); + h = hash_murmur3_one_32(p_vec.y, h); + h = hash_murmur3_one_32(p_vec.z, h); + return hash_fmix32(h); } static _FORCE_INLINE_ uint32_t hash(const Vector4i &p_vec) { uint32_t h = hash_murmur3_one_32(p_vec.x); @@ -214,15 +347,16 @@ struct HashMapHasherDefault { h = hash_murmur3_one_32(p_vec.w, h); return hash_fmix32(h); } - static _FORCE_INLINE_ uint32_t hash(const Vector2 &p_vec) { - uint32_t h = hash_djb2_one_float(p_vec.x); - return hash_djb2_one_float(p_vec.y, h); + uint32_t h = hash_murmur3_one_real(p_vec.x); + h = hash_murmur3_one_real(p_vec.y, h); + return hash_fmix32(h); } static _FORCE_INLINE_ uint32_t hash(const Vector3 &p_vec) { - uint32_t h = hash_djb2_one_float(p_vec.x); - h = hash_djb2_one_float(p_vec.y, h); - return hash_djb2_one_float(p_vec.z, h); + uint32_t h = hash_murmur3_one_real(p_vec.x); + h = hash_murmur3_one_real(p_vec.y, h); + h = hash_murmur3_one_real(p_vec.z, h); + return hash_fmix32(h); } static _FORCE_INLINE_ uint32_t hash(const Vector4 &p_vec) { uint32_t h = hash_murmur3_one_real(p_vec.x); @@ -231,31 +365,29 @@ struct HashMapHasherDefault { h = hash_murmur3_one_real(p_vec.w, h); return hash_fmix32(h); } - static _FORCE_INLINE_ uint32_t hash(const Rect2i &p_rect) { - uint32_t h = hash_djb2_one_32(p_rect.position.x); - h = hash_djb2_one_32(p_rect.position.y, h); - h = hash_djb2_one_32(p_rect.size.x, h); - return hash_djb2_one_32(p_rect.size.y, h); + uint32_t h = hash_murmur3_one_32(p_rect.position.x); + h = hash_murmur3_one_32(p_rect.position.y, h); + h = hash_murmur3_one_32(p_rect.size.x, h); + h = hash_murmur3_one_32(p_rect.size.y, h); + return hash_fmix32(h); } - static _FORCE_INLINE_ uint32_t hash(const Rect2 &p_rect) { - uint32_t h = hash_djb2_one_float(p_rect.position.x); - h = hash_djb2_one_float(p_rect.position.y, h); - h = hash_djb2_one_float(p_rect.size.x, h); - return hash_djb2_one_float(p_rect.size.y, h); + uint32_t h = hash_murmur3_one_real(p_rect.position.x); + h = hash_murmur3_one_real(p_rect.position.y, h); + h = hash_murmur3_one_real(p_rect.size.x, h); + h = hash_murmur3_one_real(p_rect.size.y, h); + return hash_fmix32(h); } - static _FORCE_INLINE_ uint32_t hash(const AABB &p_aabb) { - uint32_t h = hash_djb2_one_float(p_aabb.position.x); - h = hash_djb2_one_float(p_aabb.position.y, h); - h = hash_djb2_one_float(p_aabb.position.z, h); - h = hash_djb2_one_float(p_aabb.size.x, h); - h = hash_djb2_one_float(p_aabb.size.y, h); - return hash_djb2_one_float(p_aabb.size.z, h); + uint32_t h = hash_murmur3_one_real(p_aabb.position.x); + h = hash_murmur3_one_real(p_aabb.position.y, h); + h = hash_murmur3_one_real(p_aabb.position.z, h); + h = hash_murmur3_one_real(p_aabb.size.x, h); + h = hash_murmur3_one_real(p_aabb.size.y, h); + h = hash_murmur3_one_real(p_aabb.size.z, h); + return hash_fmix32(h); } - - // static _FORCE_INLINE_ uint32_t hash(const void* p_ptr) { return uint32_t(uint64_t(p_ptr))*(0x9e3779b1L); } }; template @@ -268,28 +400,28 @@ struct HashMapComparatorDefault { template <> struct HashMapComparatorDefault { static bool compare(const float &p_lhs, const float &p_rhs) { - return (p_lhs == p_rhs) || (std::isnan(p_lhs) && std::isnan(p_rhs)); + return (p_lhs == p_rhs) || (Math::is_nan(p_lhs) && Math::is_nan(p_rhs)); } }; template <> struct HashMapComparatorDefault { static bool compare(const double &p_lhs, const double &p_rhs) { - return (p_lhs == p_rhs) || (std::isnan(p_lhs) && std::isnan(p_rhs)); + return (p_lhs == p_rhs) || (Math::is_nan(p_lhs) && Math::is_nan(p_rhs)); } }; template <> struct HashMapComparatorDefault { static bool compare(const Vector2 &p_lhs, const Vector2 &p_rhs) { - return ((p_lhs.x == p_rhs.x) || (std::isnan(p_lhs.x) && std::isnan(p_rhs.x))) && ((p_lhs.y == p_rhs.y) || (std::isnan(p_lhs.y) && std::isnan(p_rhs.y))); + return ((p_lhs.x == p_rhs.x) || (Math::is_nan(p_lhs.x) && Math::is_nan(p_rhs.x))) && ((p_lhs.y == p_rhs.y) || (Math::is_nan(p_lhs.y) && Math::is_nan(p_rhs.y))); } }; template <> struct HashMapComparatorDefault { static bool compare(const Vector3 &p_lhs, const Vector3 &p_rhs) { - return ((p_lhs.x == p_rhs.x) || (std::isnan(p_lhs.x) && std::isnan(p_rhs.x))) && ((p_lhs.y == p_rhs.y) || (std::isnan(p_lhs.y) && std::isnan(p_rhs.y))) && ((p_lhs.z == p_rhs.z) || (std::isnan(p_lhs.z) && std::isnan(p_rhs.z))); + return ((p_lhs.x == p_rhs.x) || (Math::is_nan(p_lhs.x) && Math::is_nan(p_rhs.x))) && ((p_lhs.y == p_rhs.y) || (Math::is_nan(p_lhs.y) && Math::is_nan(p_rhs.y))) && ((p_lhs.z == p_rhs.z) || (Math::is_nan(p_lhs.z) && Math::is_nan(p_rhs.z))); } }; @@ -327,6 +459,68 @@ const uint32_t hash_table_size_primes[HASH_TABLE_SIZE_MAX] = { 1610612741, }; +// Computed with elem_i = UINT64_C (0 x FFFFFFFF FFFFFFFF ) / d_i + 1, where d_i is the i-th element of the above array. +const uint64_t hash_table_size_primes_inv[HASH_TABLE_SIZE_MAX] = { + 3689348814741910324, + 1418980313362273202, + 802032351030850071, + 392483916461905354, + 190172619316593316, + 95578984837873325, + 47420935922132524, + 23987963684927896, + 11955116055547344, + 5991147799191151, + 2998982941588287, + 1501077717772769, + 750081082979285, + 375261795343686, + 187625172388393, + 93822606204624, + 46909513691883, + 23456218233098, + 11728086747027, + 5864041509391, + 2932024948977, + 1466014921160, + 733007198436, + 366503839517, + 183251896093, + 91625960335, + 45812983922, + 22906489714, + 11453246088 +}; + +/** + * Fastmod computes ( n mod d ) given the precomputed c much faster than n % d. + * The implementation of fastmod is based on the following paper by Daniel Lemire et al. + * Faster Remainder by Direct Computation: Applications to Compilers and Software Libraries + * https://arxiv.org/abs/1902.01961 + */ +static _FORCE_INLINE_ uint32_t fastmod(const uint32_t n, const uint64_t c, const uint32_t d) { +#if defined(_MSC_VER) + // Returns the upper 64 bits of the product of two 64-bit unsigned integers. + // This intrinsic function is required since MSVC does not support unsigned 128-bit integers. +#if defined(_M_X64) || defined(_M_ARM64) + return __umulh(c * n, d); +#else + // Fallback to the slower method for 32-bit platforms. + return n % d; +#endif // _M_X64 || _M_ARM64 +#else +#ifdef __SIZEOF_INT128__ + // Prevent compiler warning, because we know what we are doing. + uint64_t lowbits = c * n; + __extension__ typedef unsigned __int128 uint128; + return static_cast(((uint128)lowbits * d) >> 64); +#else + // Fallback to the slower method if no 128-bit unsigned integer type is available. + return n % d; +#endif // __SIZEOF_INT128__ +#endif // _MSC_VER +} + } // namespace godot #endif // HASHFUNCS_HPP