#pragma once #include #if defined(__SSE2__) #include #endif #if defined(__SSE4_2__) #include #endif /** find_first_symbols(begin, end): * * Allow to search for next character from the set of 'symbols...' in a string. * It is similar to 'strpbrk', 'strcspn' (and 'strchr', 'memchr' in the case of one symbol and '\0'), * but with the following differencies: * - works with any memory ranges, including containing zero bytes; * - doesn't require terminating zero byte: end of memory range is passed explicitly; * - if not found, returns pointer to end instead of nullptr; * - maximum number of symbols to search is 16. * * Uses SSE 2 in case of small number of symbols for search and SSE 4.2 in the case of large number of symbols, * that have more than 2x performance advantage over trivial loop * in the case of parsing tab-separated dump with (probably escaped) string fields. * In the case of parsing tab separated dump with short strings, there is no performance degradation over trivial loop. * * Note: the optimal threshold to choose between SSE 2 and SSE 4.2 may depend on CPU model. * * find_last_symbols_or_null(begin, end): * * Allow to search for the last matching character in a string. * If no such characters, returns nullptr. */ namespace detail { template inline bool is_in(char x) { return x == s0; } template inline bool is_in(char x) { return x == s0 || is_in(x); } #if defined(__SSE2__) template inline __m128i mm_is_in(__m128i bytes) { __m128i eq0 = _mm_cmpeq_epi8(bytes, _mm_set1_epi8(s0)); return eq0; } template inline __m128i mm_is_in(__m128i bytes) { __m128i eq0 = _mm_cmpeq_epi8(bytes, _mm_set1_epi8(s0)); __m128i eq = mm_is_in(bytes); return _mm_or_si128(eq0, eq); } #endif template bool maybe_negate(bool x) { if constexpr (positive) return x; else return !x; } template uint16_t maybe_negate(uint16_t x) { if constexpr (positive) return x; else return ~x; } enum class ReturnMode { End, Nullptr, }; template inline const char * find_first_symbols_sse2(const char * const begin, const char * const end) { const char * pos = begin; #if defined(__SSE2__) for (; pos + 15 < end; pos += 16) { __m128i bytes = _mm_loadu_si128(reinterpret_cast(pos)); __m128i eq = mm_is_in(bytes); uint16_t bit_mask = maybe_negate(uint16_t(_mm_movemask_epi8(eq))); if (bit_mask) return pos + __builtin_ctz(bit_mask); } #endif for (; pos < end; ++pos) if (maybe_negate(is_in(*pos))) return pos; return return_mode == ReturnMode::End ? end : nullptr; } template inline const char * find_last_symbols_sse2(const char * const begin, const char * const end) { const char * pos = end; #if defined(__SSE2__) for (; pos - 16 >= begin; pos -= 16) /// Assuming the pointer cannot overflow. Assuming we can compare these pointers. { __m128i bytes = _mm_loadu_si128(reinterpret_cast(pos - 16)); __m128i eq = mm_is_in(bytes); uint16_t bit_mask = maybe_negate(uint16_t(_mm_movemask_epi8(eq))); if (bit_mask) return pos - 1 - (__builtin_clz(bit_mask) - 16); /// because __builtin_clz works with mask as uint32. } #endif --pos; for (; pos >= begin; --pos) if (maybe_negate(is_in(*pos))) return pos; return return_mode == ReturnMode::End ? end : nullptr; } template inline const char * find_first_symbols_sse42_impl(const char * const begin, const char * const end) { const char * pos = begin; #if defined(__SSE4_2__) #define MODE (_SIDD_UBYTE_OPS | _SIDD_CMP_EQUAL_ANY | _SIDD_LEAST_SIGNIFICANT) __m128i set = _mm_setr_epi8(c01, c02, c03, c04, c05, c06, c07, c08, c09, c10, c11, c12, c13, c14, c15, c16); for (; pos + 15 < end; pos += 16) { __m128i bytes = _mm_loadu_si128(reinterpret_cast(pos)); if constexpr (positive) { if (_mm_cmpestrc(set, num_chars, bytes, 16, MODE)) return pos + _mm_cmpestri(set, num_chars, bytes, 16, MODE); } else { if (_mm_cmpestrc(set, num_chars, bytes, 16, MODE | _SIDD_NEGATIVE_POLARITY)) return pos + _mm_cmpestri(set, num_chars, bytes, 16, MODE | _SIDD_NEGATIVE_POLARITY); } } #undef MODE #endif for (; pos < end; ++pos) if ( (num_chars >= 1 && maybe_negate(*pos == c01)) || (num_chars >= 2 && maybe_negate(*pos == c02)) || (num_chars >= 3 && maybe_negate(*pos == c03)) || (num_chars >= 4 && maybe_negate(*pos == c04)) || (num_chars >= 5 && maybe_negate(*pos == c05)) || (num_chars >= 6 && maybe_negate(*pos == c06)) || (num_chars >= 7 && maybe_negate(*pos == c07)) || (num_chars >= 8 && maybe_negate(*pos == c08)) || (num_chars >= 9 && maybe_negate(*pos == c09)) || (num_chars >= 10 && maybe_negate(*pos == c10)) || (num_chars >= 11 && maybe_negate(*pos == c11)) || (num_chars >= 12 && maybe_negate(*pos == c12)) || (num_chars >= 13 && maybe_negate(*pos == c13)) || (num_chars >= 14 && maybe_negate(*pos == c14)) || (num_chars >= 15 && maybe_negate(*pos == c15)) || (num_chars >= 16 && maybe_negate(*pos == c16))) return pos; return return_mode == ReturnMode::End ? end : nullptr; } template inline const char * find_first_symbols_sse42(const char * begin, const char * end) { return find_first_symbols_sse42_impl(begin, end); } /// NOTE No SSE 4.2 implementation for find_last_symbols_or_null. Not worth to do. template inline const char * find_first_symbols_dispatch(const char * begin, const char * end) { #if defined(__SSE4_2__) if (sizeof...(symbols) >= 5) return find_first_symbols_sse42(begin, end); else #endif return find_first_symbols_sse2(begin, end); } } template inline const char * find_first_symbols(const char * begin, const char * end) { return detail::find_first_symbols_dispatch(begin, end); } /// Returning non const result for non const arguments. /// It is convenient when you are using this function to iterate through non-const buffer. template inline char * find_first_symbols(char * begin, char * end) { return const_cast(detail::find_first_symbols_dispatch(begin, end)); } template inline const char * find_first_not_symbols(const char * begin, const char * end) { return detail::find_first_symbols_dispatch(begin, end); } template inline char * find_first_not_symbols(char * begin, char * end) { return const_cast(detail::find_first_symbols_dispatch(begin, end)); } template inline const char * find_first_symbols_or_null(const char * begin, const char * end) { return detail::find_first_symbols_dispatch(begin, end); } template inline char * find_first_symbols_or_null(char * begin, char * end) { return const_cast(detail::find_first_symbols_dispatch(begin, end)); } template inline const char * find_first_not_symbols_or_null(const char * begin, const char * end) { return detail::find_first_symbols_dispatch(begin, end); } template inline char * find_first_not_symbols_or_null(char * begin, char * end) { return const_cast(detail::find_first_symbols_dispatch(begin, end)); } template inline const char * find_last_symbols_or_null(const char * begin, const char * end) { return detail::find_last_symbols_sse2(begin, end); } template inline char * find_last_symbols_or_null(char * begin, char * end) { return const_cast(detail::find_last_symbols_sse2(begin, end)); } template inline const char * find_last_not_symbols_or_null(const char * begin, const char * end) { return detail::find_last_symbols_sse2(begin, end); } template inline char * find_last_not_symbols_or_null(char * begin, char * end) { return const_cast(detail::find_last_symbols_sse2(begin, end)); } /// Slightly resembles boost::split. The drawback of boost::split is that it fires a false positive in clang static analyzer. /// See https://github.com/boostorg/algorithm/issues/63 /// And https://bugs.llvm.org/show_bug.cgi?id=41141 template inline void splitInto(To & to, const std::string & what, bool token_compress = false) { const char * pos = what.data(); const char * end = pos + what.size(); while (pos < end) { const char * delimiter_or_end = find_first_symbols<','>(pos, end); if (!token_compress || pos + 1 < delimiter_or_end) to.emplace_back(pos, delimiter_or_end); if (delimiter_or_end < end) pos = delimiter_or_end + 1; else pos = delimiter_or_end; } }