microsoft/onnxruntime-extensions
Publicmirrored fromhttps://github.com/microsoft/onnxruntime-extensionsAvailable
base/string_utils.cc
232lines · modecode
| 1 | #ifdef ENABLE_TF_STRING |
| 2 | #include "farmhash.h" |
| 3 | #endif |
| 4 | |
| 5 | #include "string_utils.h" |
| 6 | |
| 7 | std::vector<std::string_view> SplitString(const std::string_view& str, const std::string_view& seps, bool remove_empty_entries) { |
| 8 | std::vector<std::string_view> result; |
| 9 | std::string ::size_type pre_pos = 0; |
| 10 | |
| 11 | //TODO: bug fix |
| 12 | while (true) { |
| 13 | auto next_pos = str.find_first_of(seps, pre_pos); |
| 14 | |
| 15 | if (next_pos == std::string::npos) { |
| 16 | auto sub_str = str.substr(pre_pos, next_pos); |
| 17 | // sub_str is empty means the last sep reach the end of string |
| 18 | if (!sub_str.empty()) { |
| 19 | result.push_back(sub_str); |
| 20 | } |
| 21 | |
| 22 | break; |
| 23 | } |
| 24 | |
| 25 | if (pre_pos != next_pos || !remove_empty_entries) { |
| 26 | auto sub_str = str.substr(pre_pos, next_pos - pre_pos); |
| 27 | result.push_back(sub_str); |
| 28 | } |
| 29 | |
| 30 | pre_pos = next_pos + 1; |
| 31 | } |
| 32 | |
| 33 | return result; |
| 34 | } |
| 35 | |
| 36 | bool IsCJK(char32_t c) { |
| 37 | return (c >= 0x4E00 && c <= 0x9FFF) |
| 38 | || (c >= 0x3400 && c <= 0x4DBF) |
| 39 | || (c >= 0x20000 && c <= 0x2A6DF) |
| 40 | || (c >= 0x2A700 && c <= 0x2B73F) |
| 41 | || (c >= 0x2B740 && c <= 0x2B81F) |
| 42 | || (c >= 0x2B820 && c <= 0x2CEAF) |
| 43 | || (c >= 0xF900 && c <= 0xFAFF) |
| 44 | || (c >= 0x2F800 && c <= 0x2FA1F); |
| 45 | } |
| 46 | |
| 47 | // Generated by tools/generate_unicode_category_table.py |
| 48 | bool IsSpace(char32_t c) { |
| 49 | if (c == 13||c == 32||c == 160||c == 8239||c == 8287||c == 12288) { |
| 50 | return true; |
| 51 | } |
| 52 | |
| 53 | if ((c >= 9 && c <= 10)||(c >= 8192 && c <= 8202)) { |
| 54 | return true; |
| 55 | } |
| 56 | |
| 57 | return false; |
| 58 | } |
| 59 | |
| 60 | // Generated by tools/generate_unicode_category_table.py |
| 61 | bool IsPunct(char32_t c) { |
| 62 | if (c == 161||c == 167||c == 171||c == 187||c == 191||c == 894||c == 903||c == 12336||c == 12349) { |
| 63 | return true; |
| 64 | } |
| 65 | |
| 66 | if ((c >= 33 && c <= 47)||(c >= 58 && c <= 64)||(c >= 91 && c <= 96)||(c >= 123 && c <= 126) |
| 67 | ||(c >= 182 && c <= 183)||(c >= 8208 && c <= 8231)||(c >= 8240 && c <= 8259) |
| 68 | || (c >= 8261 && c <= 8273)||(c >= 8275 && c <= 8286)||(c >= 12289 && c <= 12291) |
| 69 | ||(c >= 12296 && c <= 12305)||(c >= 12308 && c <= 12319)) { |
| 70 | return true; |
| 71 | } |
| 72 | |
| 73 | return false; |
| 74 | } |
| 75 | |
| 76 | // Generated by tools/generate_unicode_category_table.py |
| 77 | bool IsControl(char32_t c) { |
| 78 | if (c == 173||c == 907||c == 909||c == 930||c == 11930||c == 173790||c == 195102 |
| 79 | ) { |
| 80 | return true; |
| 81 | } |
| 82 | |
| 83 | if ((c >= 0 && c <= 8)||(c >= 11 && c <= 12)||(c >= 14 && c <= 31)||(c >= 128 && c <= 159) |
| 84 | ||(c >= 888 && c <= 889)||(c >= 896 && c <= 899)||(c >= 8203 && c <= 8207) |
| 85 | ||(c >= 8234 && c <= 8238)||(c >= 8288 && c <= 8302)||(c >= 12020 && c <= 12030) |
| 86 | ||(c >= 40957 && c <= 40958)||(c >= 64110 && c <= 64111)||(c >= 64218 && c <= 64254) |
| 87 | ||(c >= 177973 && c <= 177982)||(c >= 178206 && c <= 178207)||(c >= 183970 && c <= 183982)) { |
| 88 | return true; |
| 89 | } |
| 90 | |
| 91 | return false; |
| 92 | } |
| 93 | |
| 94 | bool IsAccent(char32_t c) |
| 95 | { |
| 96 | // only support part of accent |
| 97 | // [TODO] support more accent |
| 98 | return c >= 0x300 && c <= 0x36F; |
| 99 | } |
| 100 | |
| 101 | char32_t ToLower(char32_t c) { |
| 102 | // Basic ASCII uppercase to lowercase |
| 103 | if (c >= 'A' && c <= 'Z') { |
| 104 | return c + 32; |
| 105 | } |
| 106 | |
| 107 | // Latin-1 Supplement (À-ß) |
| 108 | if (c >= U'À' && c <= U'Þ') { |
| 109 | // Special case for × |
| 110 | if(c == U'×') { |
| 111 | return c; |
| 112 | } |
| 113 | return c + 32; |
| 114 | } |
| 115 | |
| 116 | // // Latin Extended-A (Ā-ſ) |
| 117 | // if (c >= U'Ā' && c <= U'Ʒ') { |
| 118 | // return c + 1; |
| 119 | // } |
| 120 | |
| 121 | // Greek and Coptic |
| 122 | if (c >= U'Ἀ' && c <= U'Ὠ') { |
| 123 | return c + 32; |
| 124 | } |
| 125 | |
| 126 | // Cyrillic |
| 127 | if (c >= U'А' && c <= U'Я') { |
| 128 | return c + 32; |
| 129 | } |
| 130 | |
| 131 | // Special cases |
| 132 | switch (c) { |
| 133 | case U'Ι': |
| 134 | return U'ι'; |
| 135 | case U'Κ': |
| 136 | return U'κ'; |
| 137 | case U'Λ': |
| 138 | return U'λ'; |
| 139 | case U'Μ': |
| 140 | return U'μ'; |
| 141 | case U'Ν': |
| 142 | return U'ν'; |
| 143 | case U'Ο': |
| 144 | return U'ο'; |
| 145 | case U'Π': |
| 146 | return U'π'; |
| 147 | } |
| 148 | |
| 149 | // If no conversion rule applies, return the original character |
| 150 | return c; |
| 151 | } |
| 152 | |
| 153 | char32_t StripAccent(char32_t c) |
| 154 | { |
| 155 | // "ÀÁÂÃÄÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖרÙÚÛÜÝÞßàáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿ" |
| 156 | const char32_t * tr = U"AAAAAAÆCEEEEIIIIÐNOOOOOרUUUUYÞßaaaaaaæceeeeiiiiðnooooo÷øuuuuyþy"; |
| 157 | if (c < 192 || c > 255) { |
| 158 | return c; |
| 159 | } |
| 160 | |
| 161 | return tr[c - 192]; |
| 162 | } |
| 163 | |
| 164 | #ifdef ENABLE_TF_STRING |
| 165 | // Source: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/platform/hash.cc#L28 |
| 166 | static inline uint64_t ByteAs64(char c) { return static_cast<uint64_t>(c) & 0xff; } |
| 167 | |
| 168 | // Source: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/platform/raw_coding.h#L41 |
| 169 | uint64_t DecodeFixed32(const char* ptr) { |
| 170 | return ((static_cast<uint64_t>(static_cast<unsigned char>(ptr[0]))) | |
| 171 | (static_cast<uint64_t>(static_cast<unsigned char>(ptr[1])) << 8) | |
| 172 | (static_cast<uint64_t>(static_cast<unsigned char>(ptr[2])) << 16) | |
| 173 | (static_cast<uint64_t>(static_cast<unsigned char>(ptr[3])) << 24)); |
| 174 | } |
| 175 | |
| 176 | // Source: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/platform/raw_coding.h#L55 |
| 177 | static uint64_t DecodeFixed64(const char* ptr) { |
| 178 | uint64_t lo = DecodeFixed32(ptr); |
| 179 | uint64_t hi = DecodeFixed32(ptr + 4); |
| 180 | return (hi << 32) | lo; |
| 181 | } |
| 182 | |
| 183 | // Source: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/platform/hash.cc#L79 |
| 184 | uint64_t Hash64(const char* data, size_t n, uint64_t seed) { |
| 185 | const uint64_t m = 0xc6a4a7935bd1e995; |
| 186 | const int r = 47; |
| 187 | |
| 188 | uint64_t h = seed ^ (n * m); |
| 189 | |
| 190 | while (n >= 8) { |
| 191 | uint64_t k = DecodeFixed64(data); |
| 192 | data += 8; |
| 193 | n -= 8; |
| 194 | |
| 195 | k *= m; |
| 196 | k ^= k >> r; |
| 197 | k *= m; |
| 198 | |
| 199 | h ^= k; |
| 200 | h *= m; |
| 201 | } |
| 202 | |
| 203 | switch (n) { |
| 204 | case 7: |
| 205 | h ^= ByteAs64(data[6]) << 48; |
| 206 | case 6: |
| 207 | h ^= ByteAs64(data[5]) << 40; |
| 208 | case 5: |
| 209 | h ^= ByteAs64(data[4]) << 32; |
| 210 | case 4: |
| 211 | h ^= ByteAs64(data[3]) << 24; |
| 212 | case 3: |
| 213 | h ^= ByteAs64(data[2]) << 16; |
| 214 | case 2: |
| 215 | h ^= ByteAs64(data[1]) << 8; |
| 216 | default: // case 1: make some code analyzer be happier. |
| 217 | h ^= ByteAs64(data[0]); |
| 218 | h *= m; |
| 219 | } |
| 220 | |
| 221 | h ^= h >> r; |
| 222 | h *= m; |
| 223 | h ^= h >> r; |
| 224 | |
| 225 | return h; |
| 226 | } |
| 227 | |
| 228 | uint64_t Hash64Fast(const char* data, size_t n) { |
| 229 | return static_cast<int64_t>(util::Fingerprint64(data, n)); |
| 230 | } |
| 231 | |
| 232 | #endif // ENABLE_TF_STRING |
| 233 | |