/root/doris/be/src/runtime/decimalv2_value.h
Line | Count | Source |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | |
18 | | #pragma once |
19 | | |
20 | | #include <glog/logging.h> |
21 | | #include <stdint.h> |
22 | | |
23 | | // IWYU pragma: no_include <bits/std_abs.h> |
24 | | #include <cmath> // IWYU pragma: keep |
25 | | #include <cstdint> |
26 | | #include <cstdlib> |
27 | | #include <iostream> |
28 | | #include <string> |
29 | | #include <string_view> |
30 | | |
31 | | #include "util/hash_util.hpp" |
32 | | #include "vec/core/extended_types.h" |
33 | | |
34 | | namespace doris { |
35 | | #include "common/compile_check_begin.h" |
36 | | |
37 | | typedef __int128_t int128_t; |
38 | | |
39 | | enum DecimalError { |
40 | | E_DEC_OK = 0, |
41 | | E_DEC_TRUNCATED = 1, |
42 | | E_DEC_OVERFLOW = 2, |
43 | | E_DEC_DIV_ZERO = 4, |
44 | | E_DEC_BAD_NUM = 8, |
45 | | E_DEC_OOM = 16, |
46 | | |
47 | | E_DEC_ERROR = 31, |
48 | | E_DEC_FATAL_ERROR = 30 |
49 | | }; |
50 | | |
51 | | enum DecimalRoundMode { HALF_UP = 1, HALF_EVEN = 2, CEILING = 3, FLOOR = 4, TRUNCATE = 5 }; |
52 | | |
53 | | class DecimalV2Value { |
54 | | public: |
55 | | using NativeType = __int128_t; |
56 | | friend DecimalV2Value operator+(const DecimalV2Value& v1, const DecimalV2Value& v2); |
57 | | friend DecimalV2Value operator-(const DecimalV2Value& v1, const DecimalV2Value& v2); |
58 | | friend DecimalV2Value operator*(const DecimalV2Value& v1, const DecimalV2Value& v2); |
59 | | friend DecimalV2Value operator/(const DecimalV2Value& v1, const DecimalV2Value& v2); |
60 | | friend std::istream& operator>>(std::istream& ism, DecimalV2Value& decimal_value); |
61 | | friend DecimalV2Value operator-(const DecimalV2Value& v); |
62 | | |
63 | | static constexpr int32_t PRECISION = 27; |
64 | | static constexpr int32_t SCALE = 9; |
65 | | static constexpr int32_t SCALE_TRIM_ARRAY[SCALE + 1] = { |
66 | | 1000000000, 100000000, 10000000, 1000000, 100000, 10000, 1000, 100, 10, 1}; |
67 | | static constexpr uint32_t ONE_BILLION = 1000000000; |
68 | | static constexpr int64_t MAX_INT_VALUE = 999999999999999999; |
69 | | static constexpr int32_t MAX_FRAC_VALUE = 999999999; |
70 | | static constexpr int64_t MAX_INT64 = 9223372036854775807ll; |
71 | | // In sqrt, the integer part and the decimal part of the square root to be solved separately are |
72 | | // multiplied by the PRECISION/2 power of 10, so that they can be placed in an int128_t variable |
73 | | static const int128_t SQRT_MOLECULAR_MAGNIFICATION; |
74 | | // sqrt(ONE_BILLION) * pow(10, PRECISION/2 - SCALE), it is used to calculate SCALE of the sqrt result |
75 | | static const int128_t SQRT_DENOMINATOR; |
76 | | |
77 | | static const int128_t MAX_DECIMAL_VALUE = |
78 | | static_cast<int128_t>(MAX_INT64) * ONE_BILLION + MAX_FRAC_VALUE; |
79 | | |
80 | | DecimalV2Value() = default; |
81 | 91 | const int128_t& value() const { return _value; } |
82 | 1.35k | int128_t& value() { return _value; } |
83 | | |
84 | 63 | DecimalV2Value(const std::string& decimal_str) { |
85 | 63 | parse_from_str(decimal_str.c_str(), decimal_str.size()); |
86 | 63 | } |
87 | | |
88 | 0 | DecimalV2Value(const std::string_view& decimal_str) { |
89 | 0 | parse_from_str(decimal_str.data(), decimal_str.size()); |
90 | 0 | } |
91 | | // Construct from olap engine |
92 | 38 | DecimalV2Value(int64_t int_value, int64_t frac_value) { |
93 | 38 | from_olap_decimal(int_value, frac_value); |
94 | 38 | } |
95 | | |
96 | 38 | bool from_olap_decimal(int64_t int_value, int64_t frac_value) { |
97 | 38 | bool success = true; |
98 | 38 | bool is_negative = (int_value < 0 || frac_value < 0); |
99 | 38 | if (is_negative) { |
100 | 11 | int_value = std::abs(int_value); |
101 | 11 | frac_value = std::abs(frac_value); |
102 | 11 | } |
103 | | |
104 | | //if (int_value > MAX_INT_VALUE) { |
105 | | // int_value = MAX_INT_VALUE; |
106 | | // success = false; |
107 | | //} |
108 | | |
109 | 38 | if (frac_value > MAX_FRAC_VALUE) { |
110 | 1 | frac_value = MAX_FRAC_VALUE; |
111 | 1 | success = false; |
112 | 1 | } |
113 | | |
114 | 38 | _value = static_cast<int128_t>(int_value) * ONE_BILLION + frac_value; |
115 | 38 | if (is_negative) _value = -_value; |
116 | | |
117 | 38 | return success; |
118 | 38 | } |
119 | | |
120 | 12.6k | explicit DecimalV2Value(int128_t int_value) { _value = int_value; } |
121 | | |
122 | 50 | void set_value(int128_t value) { _value = value; } |
123 | | |
124 | 1 | DecimalV2Value& assign_from_float(const float float_value) { |
125 | 1 | _value = static_cast<int128_t>(float_value * ONE_BILLION); |
126 | 1 | return *this; |
127 | 1 | } |
128 | | |
129 | 51 | DecimalV2Value& assign_from_double(const double double_value) { |
130 | 51 | _value = static_cast<int128_t>(double_value * ONE_BILLION); |
131 | 51 | return *this; |
132 | 51 | } |
133 | | |
134 | | // These cast functions are needed in "functions.cc", which is generated by python script. |
135 | | // e.g. "ComputeFunctions::Cast_DecimalV2Value_double()" |
136 | | // Discard the scale part |
137 | | // ATTN: invoker must make sure no OVERFLOW |
138 | 15.4k | operator int64_t() const { return static_cast<int64_t>(_value / ONE_BILLION); } |
139 | | |
140 | | // These cast functions are needed in "functions.cc", which is generated by python script. |
141 | | // e.g. "ComputeFunctions::Cast_DecimalV2Value_double()" |
142 | | // Discard the scale part |
143 | | // ATTN: invoker must make sure no OVERFLOW |
144 | 0 | operator int128_t() const { return static_cast<int128_t>(_value / ONE_BILLION); } |
145 | | |
146 | 0 | operator wide::Int256() const { |
147 | 0 | wide::Int256 result; |
148 | 0 | wide::Int256::_impl::wide_integer_from_builtin(result, _value); |
149 | 0 | return result; |
150 | 0 | } |
151 | | |
152 | 0 | operator bool() const { return _value != 0; } |
153 | | |
154 | 0 | operator int8_t() const { return static_cast<char>(operator int64_t()); } |
155 | | |
156 | 0 | operator int16_t() const { return static_cast<int16_t>(operator int64_t()); } |
157 | | |
158 | 0 | operator int32_t() const { return static_cast<int32_t>(operator int64_t()); } |
159 | | |
160 | 0 | operator size_t() const { return static_cast<size_t>(operator int64_t()); } |
161 | | |
162 | 0 | operator float() const { return (float)operator double(); } |
163 | | |
164 | 0 | operator double() const { |
165 | 0 | std::string str_buff = to_string(); |
166 | 0 | double result = std::strtod(str_buff.c_str(), nullptr); |
167 | 0 | return result; |
168 | 0 | } |
169 | | |
170 | | DecimalV2Value& operator+=(const DecimalV2Value& other); |
171 | | |
172 | | // To be Compatible with OLAP |
173 | | // ATTN: NO-OVERFLOW should be guaranteed. |
174 | 15.4k | int64_t int_value() const { return operator int64_t(); } |
175 | | |
176 | | // To be Compatible with OLAP |
177 | | // NOTE: return a negative value if decimal is negative. |
178 | | // ATTN: the max length of fraction part in OLAP is 9, so the 'big digits' except the first one |
179 | | // will be truncated. |
180 | 15.4k | int32_t frac_value() const { return static_cast<int32_t>(_value % ONE_BILLION); } |
181 | | |
182 | 29 | bool operator==(const DecimalV2Value& other) const { return _value == other.value(); } |
183 | | |
184 | 35 | auto operator<=>(const DecimalV2Value& other) const { return _value <=> other.value(); } |
185 | | |
186 | | // change to maximum value for given precision and scale |
187 | | // precision/scale - see decimal_bin_size() below |
188 | | // to - decimal where where the result will be stored |
189 | | void to_max_decimal(int precision, int frac); |
190 | 0 | void to_min_decimal(int precision, int frac) { |
191 | 0 | to_max_decimal(precision, frac); |
192 | 0 | if (_value > 0) _value = -_value; |
193 | 0 | } |
194 | | |
195 | | // The maximum of fraction part is "scale". |
196 | | // If the length of fraction part is less than "scale", '0' will be filled. |
197 | | std::string to_string(int scale) const; |
198 | | |
199 | | int32_t to_buffer(char* buffer, int scale) const; |
200 | | |
201 | | // Output actual "scale", remove ending zeroes. |
202 | | std::string to_string() const; |
203 | | |
204 | | // Convert string to decimal |
205 | | // @param from - value to convert. Doesn't have to be \0 terminated! |
206 | | // will stop at the fist non-digit char(nor '.' 'e' 'E'), |
207 | | // or reaches the length |
208 | | // @param length - maximum length |
209 | | // @return error number. |
210 | | // |
211 | | // E_DEC_OK/E_DEC_TRUNCATED/E_DEC_OVERFLOW/E_DEC_BAD_NUM/E_DEC_OOM |
212 | | // In case of E_DEC_FATAL_ERROR *to is set to decimal zero |
213 | | // (to make error handling easier) |
214 | | // |
215 | | // e.g. "1.2" ".2" "1.2e-3" "1.2e3" |
216 | | int parse_from_str(const char* decimal_str, size_t length); |
217 | | |
218 | 0 | std::string get_debug_info() const { return to_string(); } |
219 | | |
220 | 5 | static DecimalV2Value get_min_decimal() { |
221 | 5 | return DecimalV2Value(-MAX_INT_VALUE, MAX_FRAC_VALUE); |
222 | 5 | } |
223 | | |
224 | 7 | static DecimalV2Value get_max_decimal() { |
225 | 7 | return DecimalV2Value(MAX_INT_VALUE, MAX_FRAC_VALUE); |
226 | 7 | } |
227 | | |
228 | 0 | static DecimalV2Value get_min_decimal(int precision, int scale) { |
229 | 0 | DCHECK(precision > 0 && precision <= 27 && scale >= 0 && scale <= 9 && precision >= scale && |
230 | 0 | (precision - scale <= 18)); |
231 | 0 | return DecimalV2Value( |
232 | 0 | -MAX_INT_VALUE % static_cast<int64_t>(get_scale_base(18 - precision + scale)), |
233 | 0 | MAX_FRAC_VALUE / static_cast<int64_t>(get_scale_base(9 - scale)) * |
234 | 0 | static_cast<int64_t>(get_scale_base(9 - scale))); |
235 | 0 | } |
236 | | |
237 | 0 | static DecimalV2Value get_max_decimal(int precision, int scale) { |
238 | 0 | DCHECK(precision > 0 && precision <= 27 && scale >= 0 && scale <= 9 && precision >= scale && |
239 | 0 | (precision - scale <= 18)); |
240 | 0 | return DecimalV2Value( |
241 | 0 | MAX_INT_VALUE % static_cast<int64_t>(get_scale_base(18 - precision + scale)), |
242 | 0 | MAX_FRAC_VALUE / static_cast<int64_t>(get_scale_base(9 - scale)) * |
243 | 0 | static_cast<int64_t>(get_scale_base(9 - scale))); |
244 | 0 | } |
245 | | |
246 | | // Solve Square root for int128 |
247 | | static DecimalV2Value sqrt(const DecimalV2Value& v); |
248 | | |
249 | | // set DecimalV2Value to zero |
250 | 0 | void set_to_zero() { _value = 0; } |
251 | | |
252 | 0 | void to_abs_value() { |
253 | 0 | if (_value < 0) _value = -_value; |
254 | 0 | } |
255 | | |
256 | 18 | uint32_t hash(uint32_t seed) const { return HashUtil::hash(&_value, sizeof(_value), seed); } |
257 | | |
258 | 1 | int32_t precision() const { return PRECISION; } |
259 | | |
260 | 1.34k | int32_t scale() const { return SCALE; } |
261 | | |
262 | | bool greater_than_scale(int scale); |
263 | | |
264 | | int round(DecimalV2Value* to, int scale, DecimalRoundMode mode); |
265 | | |
266 | 52 | inline static int128_t get_scale_base(int scale) { |
267 | 52 | static const int128_t values[] = { |
268 | 52 | static_cast<int128_t>(1ll), |
269 | 52 | static_cast<int128_t>(10ll), |
270 | 52 | static_cast<int128_t>(100ll), |
271 | 52 | static_cast<int128_t>(1000ll), |
272 | 52 | static_cast<int128_t>(10000ll), |
273 | 52 | static_cast<int128_t>(100000ll), |
274 | 52 | static_cast<int128_t>(1000000ll), |
275 | 52 | static_cast<int128_t>(10000000ll), |
276 | 52 | static_cast<int128_t>(100000000ll), |
277 | 52 | static_cast<int128_t>(1000000000ll), |
278 | 52 | static_cast<int128_t>(10000000000ll), |
279 | 52 | static_cast<int128_t>(100000000000ll), |
280 | 52 | static_cast<int128_t>(1000000000000ll), |
281 | 52 | static_cast<int128_t>(10000000000000ll), |
282 | 52 | static_cast<int128_t>(100000000000000ll), |
283 | 52 | static_cast<int128_t>(1000000000000000ll), |
284 | 52 | static_cast<int128_t>(10000000000000000ll), |
285 | 52 | static_cast<int128_t>(100000000000000000ll), |
286 | 52 | static_cast<int128_t>(1000000000000000000ll), |
287 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10ll, |
288 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100ll, |
289 | 52 | static_cast<int128_t>(1000000000000000000ll) * 1000ll, |
290 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10000ll, |
291 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000ll, |
292 | 52 | static_cast<int128_t>(1000000000000000000ll) * 1000000ll, |
293 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10000000ll, |
294 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000ll, |
295 | 52 | static_cast<int128_t>(1000000000000000000ll) * 1000000000ll, |
296 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10000000000ll, |
297 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000ll, |
298 | 52 | static_cast<int128_t>(1000000000000000000ll) * 1000000000000ll, |
299 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10000000000000ll, |
300 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000000ll, |
301 | 52 | static_cast<int128_t>(1000000000000000000ll) * 1000000000000000ll, |
302 | 52 | static_cast<int128_t>(1000000000000000000ll) * 10000000000000000ll, |
303 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000000000ll, |
304 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000000000ll * 10ll, |
305 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000000000ll * 100ll, |
306 | 52 | static_cast<int128_t>(1000000000000000000ll) * 100000000000000000ll * 1000ll}; |
307 | 52 | if (scale >= 0 && scale < 38) return values[scale]; |
308 | 0 | return -1; // Overflow |
309 | 52 | } |
310 | | |
311 | 2 | bool is_zero() const { return _value == 0; } |
312 | | |
313 | | private: |
314 | | int128_t _value; |
315 | | }; |
316 | | |
317 | | DecimalV2Value operator+(const DecimalV2Value& v1, const DecimalV2Value& v2); |
318 | | DecimalV2Value operator-(const DecimalV2Value& v1, const DecimalV2Value& v2); |
319 | | DecimalV2Value operator*(const DecimalV2Value& v1, const DecimalV2Value& v2); |
320 | | DecimalV2Value operator/(const DecimalV2Value& v1, const DecimalV2Value& v2); |
321 | | DecimalV2Value operator%(const DecimalV2Value& v1, const DecimalV2Value& v2); |
322 | | |
323 | | DecimalV2Value operator-(const DecimalV2Value& v); |
324 | | |
325 | | std::ostream& operator<<(std::ostream& os, DecimalV2Value const& decimal_value); |
326 | | std::istream& operator>>(std::istream& ism, DecimalV2Value& decimal_value); |
327 | | |
328 | | std::size_t hash_value(DecimalV2Value const& value); |
329 | | |
330 | | #include "common/compile_check_end.h" |
331 | | } // end namespace doris |
332 | | |
333 | | template <> |
334 | | struct std::hash<doris::DecimalV2Value> { |
335 | 0 | size_t operator()(const doris::DecimalV2Value& v) const { return doris::hash_value(v); } |
336 | | }; |
337 | | |
338 | | template <> |
339 | | struct std::equal_to<doris::DecimalV2Value> { |
340 | 8 | bool operator()(const doris::DecimalV2Value& lhs, const doris::DecimalV2Value& rhs) const { |
341 | 8 | return lhs == rhs; |
342 | 8 | } |
343 | | }; |