#pragma once #include #include // for std::logic_error #include #include #include #include #include #include #include #if defined(__SSE2__) #include #endif #if defined(__SSE4_2__) #include #include #endif /** * The std::string_view-like container to avoid creating strings to find substrings in the hash table. */ struct StringRef { const char * data = nullptr; size_t size = 0; /// Non-constexpr due to reinterpret_cast. template > StringRef(const CharT * data_, size_t size_) : data(reinterpret_cast(data_)), size(size_) { /// Sanity check for overflowed values. assert(size < 0x8000000000000000ULL); } constexpr StringRef(const char * data_, size_t size_) : data(data_), size(size_) {} StringRef(const std::string & s) : data(s.data()), size(s.size()) {} constexpr explicit StringRef(std::string_view s) : data(s.data()), size(s.size()) {} constexpr StringRef(const char * data_) : StringRef(std::string_view{data_}) {} constexpr StringRef() = default; std::string toString() const { return std::string(data, size); } explicit operator std::string() const { return toString(); } constexpr explicit operator std::string_view() const { return {data, size}; } }; /// Here constexpr doesn't implicate inline, see https://www.viva64.com/en/w/v1043/ /// nullptr can't be used because the StringRef values are used in SipHash's pointer arithmetic /// and the UBSan thinks that something like nullptr + 8 is UB. constexpr const inline char empty_string_ref_addr{}; constexpr const inline StringRef EMPTY_STRING_REF{&empty_string_ref_addr, 0}; using StringRefs = std::vector; #if defined(__SSE2__) /** Compare strings for equality. * The approach is controversial and does not win in all cases. * For more information, see hash_map_string_2.cpp */ inline bool compareSSE2(const char * p1, const char * p2) { return 0xFFFF == _mm_movemask_epi8(_mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(p1)), _mm_loadu_si128(reinterpret_cast(p2)))); } inline bool compareSSE2x4(const char * p1, const char * p2) { return 0xFFFF == _mm_movemask_epi8( _mm_and_si128( _mm_and_si128( _mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(p1)), _mm_loadu_si128(reinterpret_cast(p2))), _mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(p1) + 1), _mm_loadu_si128(reinterpret_cast(p2) + 1))), _mm_and_si128( _mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(p1) + 2), _mm_loadu_si128(reinterpret_cast(p2) + 2)), _mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(p1) + 3), _mm_loadu_si128(reinterpret_cast(p2) + 3))))); } inline bool memequalSSE2Wide(const char * p1, const char * p2, size_t size) { /** The order of branches and the trick with overlapping comparisons * are the same as in memcpy implementation. * See the comments in base/glibc-compatibility/memcpy/memcpy.h */ if (size <= 16) { if (size >= 8) { /// Chunks of 8..16 bytes. return unalignedLoad(p1) == unalignedLoad(p2) && unalignedLoad(p1 + size - 8) == unalignedLoad(p2 + size - 8); } else if (size >= 4) { /// Chunks of 4..7 bytes. return unalignedLoad(p1) == unalignedLoad(p2) && unalignedLoad(p1 + size - 4) == unalignedLoad(p2 + size - 4); } else if (size >= 2) { /// Chunks of 2..3 bytes. return unalignedLoad(p1) == unalignedLoad(p2) && unalignedLoad(p1 + size - 2) == unalignedLoad(p2 + size - 2); } else if (size >= 1) { /// A single byte. return *p1 == *p2; } return true; } while (size >= 64) { if (compareSSE2x4(p1, p2)) { p1 += 64; p2 += 64; size -= 64; } else return false; } switch (size / 16) { case 3: if (!compareSSE2(p1 + 32, p2 + 32)) return false; [[fallthrough]]; case 2: if (!compareSSE2(p1 + 16, p2 + 16)) return false; [[fallthrough]]; case 1: if (!compareSSE2(p1, p2)) return false; } return compareSSE2(p1 + size - 16, p2 + size - 16); } #endif inline bool operator== (StringRef lhs, StringRef rhs) { if (lhs.size != rhs.size) return false; if (lhs.size == 0) return true; #if defined(__SSE2__) return memequalSSE2Wide(lhs.data, rhs.data, lhs.size); #else return 0 == memcmp(lhs.data, rhs.data, lhs.size); #endif } inline bool operator!= (StringRef lhs, StringRef rhs) { return !(lhs == rhs); } inline bool operator< (StringRef lhs, StringRef rhs) { int cmp = memcmp(lhs.data, rhs.data, std::min(lhs.size, rhs.size)); return cmp < 0 || (cmp == 0 && lhs.size < rhs.size); } inline bool operator> (StringRef lhs, StringRef rhs) { int cmp = memcmp(lhs.data, rhs.data, std::min(lhs.size, rhs.size)); return cmp > 0 || (cmp == 0 && lhs.size > rhs.size); } /** Hash functions. * You can use either CityHash64, * or a function based on the crc32 statement, * which is obviously less qualitative, but on real data sets, * when used in a hash table, works much faster. * For more information, see hash_map_string_3.cpp */ struct StringRefHash64 { size_t operator() (StringRef x) const { return CityHash_v1_0_2::CityHash64(x.data, x.size); } }; #if defined(__SSE4_2__) /// Parts are taken from CityHash. inline UInt64 hashLen16(UInt64 u, UInt64 v) { return CityHash_v1_0_2::Hash128to64(CityHash_v1_0_2::uint128(u, v)); } inline UInt64 shiftMix(UInt64 val) { return val ^ (val >> 47); } inline UInt64 rotateByAtLeast1(UInt64 val, int shift) { return (val >> shift) | (val << (64 - shift)); } inline size_t hashLessThan8(const char * data, size_t size) { static constexpr UInt64 k2 = 0x9ae16a3b2f90404fULL; static constexpr UInt64 k3 = 0xc949d7c7509e6557ULL; if (size >= 4) { UInt64 a = unalignedLoad(data); return hashLen16(size + (a << 3), unalignedLoad(data + size - 4)); } if (size > 0) { uint8_t a = data[0]; uint8_t b = data[size >> 1]; uint8_t c = data[size - 1]; uint32_t y = static_cast(a) + (static_cast(b) << 8); uint32_t z = size + (static_cast(c) << 2); return shiftMix(y * k2 ^ z * k3) * k2; } return k2; } inline size_t hashLessThan16(const char * data, size_t size) { if (size > 8) { UInt64 a = unalignedLoad(data); UInt64 b = unalignedLoad(data + size - 8); return hashLen16(a, rotateByAtLeast1(b + size, size)) ^ b; } return hashLessThan8(data, size); } struct CRC32Hash { size_t operator() (StringRef x) const { const char * pos = x.data; size_t size = x.size; if (size == 0) return 0; if (size < 8) { return hashLessThan8(x.data, x.size); } const char * end = pos + size; size_t res = -1ULL; do { UInt64 word = unalignedLoad(pos); res = _mm_crc32_u64(res, word); pos += 8; } while (pos + 8 < end); UInt64 word = unalignedLoad(end - 8); /// I'm not sure if this is normal. res = _mm_crc32_u64(res, word); return res; } }; struct StringRefHash : CRC32Hash {}; #else struct CRC32Hash { size_t operator() (StringRef /* x */) const { throw std::logic_error{"Not implemented CRC32Hash without SSE"}; } }; struct StringRefHash : StringRefHash64 {}; #endif namespace std { template <> struct hash : public StringRefHash {}; } namespace ZeroTraits { inline bool check(const StringRef & x) { return 0 == x.size; } inline void set(StringRef & x) { x.size = 0; } } std::ostream & operator<<(std::ostream & os, const StringRef & str);