| 1 | #pragma once |
| 2 | |
| 3 | #include <Core/Types.h> |
| 4 | #include <Common/BitHelpers.h> |
| 5 | |
| 6 | #ifdef __SSE2__ |
| 7 | #include <emmintrin.h> |
| 8 | #endif |
| 9 | |
| 10 | |
| 11 | namespace DB |
| 12 | { |
| 13 | |
| 14 | |
| 15 | namespace UTF8 |
| 16 | { |
| 17 | |
| 18 | static const UInt8 CONTINUATION_OCTET_MASK = 0b11000000u; |
| 19 | static const UInt8 CONTINUATION_OCTET = 0b10000000u; |
| 20 | |
| 21 | /// return true if `octet` binary repr starts with 10 (octet is a UTF-8 sequence continuation) |
| 22 | inline bool isContinuationOctet(const UInt8 octet) |
| 23 | { |
| 24 | return (octet & CONTINUATION_OCTET_MASK) == CONTINUATION_OCTET; |
| 25 | } |
| 26 | |
| 27 | /// moves `s` backward until either first non-continuation octet or begin |
| 28 | inline void syncBackward(const UInt8 * & s, const UInt8 * const begin) |
| 29 | { |
| 30 | while (isContinuationOctet(*s) && s > begin) |
| 31 | --s; |
| 32 | } |
| 33 | |
| 34 | /// moves `s` forward until either first non-continuation octet or string end is met |
| 35 | inline void syncForward(const UInt8 * & s, const UInt8 * const end) |
| 36 | { |
| 37 | while (s < end && isContinuationOctet(*s)) |
| 38 | ++s; |
| 39 | } |
| 40 | |
| 41 | /// returns UTF-8 code point sequence length judging by it's first octet |
| 42 | inline size_t seqLength(const UInt8 first_octet) |
| 43 | { |
| 44 | if (first_octet < 0x80u) |
| 45 | return 1; |
| 46 | |
| 47 | const size_t bits = 8; |
| 48 | const auto first_zero = bitScanReverse(static_cast<UInt8>(~first_octet)); |
| 49 | |
| 50 | return bits - 1 - first_zero; |
| 51 | } |
| 52 | |
| 53 | inline size_t countCodePoints(const UInt8 * data, size_t size) |
| 54 | { |
| 55 | size_t res = 0; |
| 56 | const auto end = data + size; |
| 57 | |
| 58 | #ifdef __SSE2__ |
| 59 | constexpr auto bytes_sse = sizeof(__m128i); |
| 60 | const auto src_end_sse = data + size / bytes_sse * bytes_sse; |
| 61 | |
| 62 | const auto threshold = _mm_set1_epi8(0xBF); |
| 63 | |
| 64 | for (; data < src_end_sse; data += bytes_sse) |
| 65 | res += __builtin_popcount(_mm_movemask_epi8( |
| 66 | _mm_cmpgt_epi8(_mm_loadu_si128(reinterpret_cast<const __m128i *>(data)), threshold))); |
| 67 | #endif |
| 68 | |
| 69 | for (; data < end; ++data) /// Skip UTF-8 continuation bytes. |
| 70 | res += static_cast<Int8>(*data) > static_cast<Int8>(0xBF); |
| 71 | |
| 72 | return res; |
| 73 | } |
| 74 | |
| 75 | /// returns UTF-8 wcswidth. Invalid sequence is treated as zero width character. |
| 76 | /// `prefix` is used to compute the `\t` width which extends the string before |
| 77 | /// and include `\t` to the nearest longer length with multiple of eight. |
| 78 | size_t computeWidth(const UInt8 * data, size_t size, size_t prefix = 0) noexcept; |
| 79 | |
| 80 | } |
| 81 | |
| 82 | |
| 83 | } |
| 84 | |