• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2017 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4 
5 #ifndef BASE_NUMERICS_CHECKED_MATH_IMPL_H_
6 #define BASE_NUMERICS_CHECKED_MATH_IMPL_H_
7 
8 #include <stddef.h>
9 #include <stdint.h>
10 
11 #include <climits>
12 #include <cmath>
13 #include <cstdlib>
14 #include <limits>
15 #include <type_traits>
16 
17 #include "base/numerics/safe_conversions.h"
18 #include "base/numerics/safe_math_shared_impl.h"
19 
20 namespace base {
21 namespace internal {
22 
23 template <typename T>
CheckedAddImpl(T x,T y,T * result)24 constexpr bool CheckedAddImpl(T x, T y, T* result) {
25   static_assert(std::is_integral<T>::value, "Type must be integral");
26   // Since the value of x+y is undefined if we have a signed type, we compute
27   // it using the unsigned type of the same size.
28   using UnsignedDst = typename std::make_unsigned<T>::type;
29   using SignedDst = typename std::make_signed<T>::type;
30   UnsignedDst ux = static_cast<UnsignedDst>(x);
31   UnsignedDst uy = static_cast<UnsignedDst>(y);
32   UnsignedDst uresult = static_cast<UnsignedDst>(ux + uy);
33   *result = static_cast<T>(uresult);
34   // Addition is valid if the sign of (x + y) is equal to either that of x or
35   // that of y.
36   return (std::is_signed<T>::value)
37              ? static_cast<SignedDst>((uresult ^ ux) & (uresult ^ uy)) >= 0
38              : uresult >= uy;  // Unsigned is either valid or underflow.
39 }
40 
41 template <typename T, typename U, class Enable = void>
42 struct CheckedAddOp {};
43 
44 template <typename T, typename U>
45 struct CheckedAddOp<T,
46                     U,
47                     typename std::enable_if<std::is_integral<T>::value &&
48                                             std::is_integral<U>::value>::type> {
49   using result_type = typename MaxExponentPromotion<T, U>::type;
50   template <typename V>
51   static constexpr bool Do(T x, U y, V* result) {
52     // TODO(jschuh) Make this "constexpr if" once we're C++17.
53     if (CheckedAddFastOp<T, U>::is_supported)
54       return CheckedAddFastOp<T, U>::Do(x, y, result);
55 
56     // Double the underlying type up to a full machine word.
57     using FastPromotion = typename FastIntegerArithmeticPromotion<T, U>::type;
58     using Promotion =
59         typename std::conditional<(IntegerBitsPlusSign<FastPromotion>::value >
60                                    IntegerBitsPlusSign<intptr_t>::value),
61                                   typename BigEnoughPromotion<T, U>::type,
62                                   FastPromotion>::type;
63     // Fail if either operand is out of range for the promoted type.
64     // TODO(jschuh): This could be made to work for a broader range of values.
65     if (BASE_NUMERICS_UNLIKELY(!IsValueInRangeForNumericType<Promotion>(x) ||
66                                !IsValueInRangeForNumericType<Promotion>(y))) {
67       return false;
68     }
69 
70     Promotion presult = {};
71     bool is_valid = true;
72     if (IsIntegerArithmeticSafe<Promotion, T, U>::value) {
73       presult = static_cast<Promotion>(x) + static_cast<Promotion>(y);
74     } else {
75       is_valid = CheckedAddImpl(static_cast<Promotion>(x),
76                                 static_cast<Promotion>(y), &presult);
77     }
78     *result = static_cast<V>(presult);
79     return is_valid && IsValueInRangeForNumericType<V>(presult);
80   }
81 };
82 
83 template <typename T>
84 constexpr bool CheckedSubImpl(T x, T y, T* result) {
85   static_assert(std::is_integral<T>::value, "Type must be integral");
86   // Since the value of x+y is undefined if we have a signed type, we compute
87   // it using the unsigned type of the same size.
88   using UnsignedDst = typename std::make_unsigned<T>::type;
89   using SignedDst = typename std::make_signed<T>::type;
90   UnsignedDst ux = static_cast<UnsignedDst>(x);
91   UnsignedDst uy = static_cast<UnsignedDst>(y);
92   UnsignedDst uresult = static_cast<UnsignedDst>(ux - uy);
93   *result = static_cast<T>(uresult);
94   // Subtraction is valid if either x and y have same sign, or (x-y) and x have
95   // the same sign.
96   return (std::is_signed<T>::value)
97              ? static_cast<SignedDst>((uresult ^ ux) & (ux ^ uy)) >= 0
98              : x >= y;
99 }
100 
101 template <typename T, typename U, class Enable = void>
102 struct CheckedSubOp {};
103 
104 template <typename T, typename U>
105 struct CheckedSubOp<T,
106                     U,
107                     typename std::enable_if<std::is_integral<T>::value &&
108                                             std::is_integral<U>::value>::type> {
109   using result_type = typename MaxExponentPromotion<T, U>::type;
110   template <typename V>
111   static constexpr bool Do(T x, U y, V* result) {
112     // TODO(jschuh) Make this "constexpr if" once we're C++17.
113     if (CheckedSubFastOp<T, U>::is_supported)
114       return CheckedSubFastOp<T, U>::Do(x, y, result);
115 
116     // Double the underlying type up to a full machine word.
117     using FastPromotion = typename FastIntegerArithmeticPromotion<T, U>::type;
118     using Promotion =
119         typename std::conditional<(IntegerBitsPlusSign<FastPromotion>::value >
120                                    IntegerBitsPlusSign<intptr_t>::value),
121                                   typename BigEnoughPromotion<T, U>::type,
122                                   FastPromotion>::type;
123     // Fail if either operand is out of range for the promoted type.
124     // TODO(jschuh): This could be made to work for a broader range of values.
125     if (BASE_NUMERICS_UNLIKELY(!IsValueInRangeForNumericType<Promotion>(x) ||
126                                !IsValueInRangeForNumericType<Promotion>(y))) {
127       return false;
128     }
129 
130     Promotion presult = {};
131     bool is_valid = true;
132     if (IsIntegerArithmeticSafe<Promotion, T, U>::value) {
133       presult = static_cast<Promotion>(x) - static_cast<Promotion>(y);
134     } else {
135       is_valid = CheckedSubImpl(static_cast<Promotion>(x),
136                                 static_cast<Promotion>(y), &presult);
137     }
138     *result = static_cast<V>(presult);
139     return is_valid && IsValueInRangeForNumericType<V>(presult);
140   }
141 };
142 
143 template <typename T>
144 constexpr bool CheckedMulImpl(T x, T y, T* result) {
145   static_assert(std::is_integral<T>::value, "Type must be integral");
146   // Since the value of x*y is potentially undefined if we have a signed type,
147   // we compute it using the unsigned type of the same size.
148   using UnsignedDst = typename std::make_unsigned<T>::type;
149   using SignedDst = typename std::make_signed<T>::type;
150   const UnsignedDst ux = SafeUnsignedAbs(x);
151   const UnsignedDst uy = SafeUnsignedAbs(y);
152   UnsignedDst uresult = static_cast<UnsignedDst>(ux * uy);
153   const bool is_negative =
154       std::is_signed<T>::value && static_cast<SignedDst>(x ^ y) < 0;
155   *result = is_negative ? 0 - uresult : uresult;
156   // We have a fast out for unsigned identity or zero on the second operand.
157   // After that it's an unsigned overflow check on the absolute value, with
158   // a +1 bound for a negative result.
159   return uy <= UnsignedDst(!std::is_signed<T>::value || is_negative) ||
160          ux <= (std::numeric_limits<T>::max() + UnsignedDst(is_negative)) / uy;
161 }
162 
163 template <typename T, typename U, class Enable = void>
164 struct CheckedMulOp {};
165 
166 template <typename T, typename U>
167 struct CheckedMulOp<T,
168                     U,
169                     typename std::enable_if<std::is_integral<T>::value &&
170                                             std::is_integral<U>::value>::type> {
171   using result_type = typename MaxExponentPromotion<T, U>::type;
172   template <typename V>
173   static constexpr bool Do(T x, U y, V* result) {
174     // TODO(jschuh) Make this "constexpr if" once we're C++17.
175     if (CheckedMulFastOp<T, U>::is_supported)
176       return CheckedMulFastOp<T, U>::Do(x, y, result);
177 
178     using Promotion = typename FastIntegerArithmeticPromotion<T, U>::type;
179     // Verify the destination type can hold the result (always true for 0).
180     if (BASE_NUMERICS_UNLIKELY((!IsValueInRangeForNumericType<Promotion>(x) ||
181                                 !IsValueInRangeForNumericType<Promotion>(y)) &&
182                                x && y)) {
183       return false;
184     }
185 
186     Promotion presult = {};
187     bool is_valid = true;
188     if (CheckedMulFastOp<Promotion, Promotion>::is_supported) {
189       // The fast op may be available with the promoted type.
190       is_valid = CheckedMulFastOp<Promotion, Promotion>::Do(x, y, &presult);
191     } else if (IsIntegerArithmeticSafe<Promotion, T, U>::value) {
192       presult = static_cast<Promotion>(x) * static_cast<Promotion>(y);
193     } else {
194       is_valid = CheckedMulImpl(static_cast<Promotion>(x),
195                                 static_cast<Promotion>(y), &presult);
196     }
197     *result = static_cast<V>(presult);
198     return is_valid && IsValueInRangeForNumericType<V>(presult);
199   }
200 };
201 
202 // Division just requires a check for a zero denominator or an invalid negation
203 // on signed min/-1.
204 template <typename T, typename U, class Enable = void>
205 struct CheckedDivOp {};
206 
207 template <typename T, typename U>
208 struct CheckedDivOp<T,
209                     U,
210                     typename std::enable_if<std::is_integral<T>::value &&
211                                             std::is_integral<U>::value>::type> {
212   using result_type = typename MaxExponentPromotion<T, U>::type;
213   template <typename V>
214   static constexpr bool Do(T x, U y, V* result) {
215     if (BASE_NUMERICS_UNLIKELY(!y))
216       return false;
217 
218     // The overflow check can be compiled away if we don't have the exact
219     // combination of types needed to trigger this case.
220     using Promotion = typename BigEnoughPromotion<T, U>::type;
221     if (BASE_NUMERICS_UNLIKELY(
222             (std::is_signed<T>::value && std::is_signed<U>::value &&
223              IsTypeInRangeForNumericType<T, Promotion>::value &&
224              static_cast<Promotion>(x) ==
225                  std::numeric_limits<Promotion>::lowest() &&
226              y == static_cast<U>(-1)))) {
227       return false;
228     }
229 
230     // This branch always compiles away if the above branch wasn't removed.
231     if (BASE_NUMERICS_UNLIKELY((!IsValueInRangeForNumericType<Promotion>(x) ||
232                                 !IsValueInRangeForNumericType<Promotion>(y)) &&
233                                x)) {
234       return false;
235     }
236 
237     Promotion presult = Promotion(x) / Promotion(y);
238     *result = static_cast<V>(presult);
239     return IsValueInRangeForNumericType<V>(presult);
240   }
241 };
242 
243 template <typename T, typename U, class Enable = void>
244 struct CheckedModOp {};
245 
246 template <typename T, typename U>
247 struct CheckedModOp<T,
248                     U,
249                     typename std::enable_if<std::is_integral<T>::value &&
250                                             std::is_integral<U>::value>::type> {
251   using result_type = typename MaxExponentPromotion<T, U>::type;
252   template <typename V>
253   static constexpr bool Do(T x, U y, V* result) {
254     using Promotion = typename BigEnoughPromotion<T, U>::type;
255     if (BASE_NUMERICS_LIKELY(y)) {
256       Promotion presult = static_cast<Promotion>(x) % static_cast<Promotion>(y);
257       *result = static_cast<Promotion>(presult);
258       return IsValueInRangeForNumericType<V>(presult);
259     }
260     return false;
261   }
262 };
263 
264 template <typename T, typename U, class Enable = void>
265 struct CheckedLshOp {};
266 
267 // Left shift. Shifts less than 0 or greater than or equal to the number
268 // of bits in the promoted type are undefined. Shifts of negative values
269 // are undefined. Otherwise it is defined when the result fits.
270 template <typename T, typename U>
271 struct CheckedLshOp<T,
272                     U,
273                     typename std::enable_if<std::is_integral<T>::value &&
274                                             std::is_integral<U>::value>::type> {
275   using result_type = T;
276   template <typename V>
277   static constexpr bool Do(T x, U shift, V* result) {
278     // Disallow negative numbers and verify the shift is in bounds.
279     if (BASE_NUMERICS_LIKELY(!IsValueNegative(x) &&
280                              as_unsigned(shift) <
281                                  as_unsigned(std::numeric_limits<T>::digits))) {
282       // Shift as unsigned to avoid undefined behavior.
283       *result = static_cast<V>(as_unsigned(x) << shift);
284       // If the shift can be reversed, we know it was valid.
285       return *result >> shift == x;
286     }
287 
288     // Handle the legal corner-case of a full-width signed shift of zero.
289     return std::is_signed<T>::value && !x &&
290            as_unsigned(shift) == as_unsigned(std::numeric_limits<T>::digits);
291   }
292 };
293 
294 template <typename T, typename U, class Enable = void>
295 struct CheckedRshOp {};
296 
297 // Right shift. Shifts less than 0 or greater than or equal to the number
298 // of bits in the promoted type are undefined. Otherwise, it is always defined,
299 // but a right shift of a negative value is implementation-dependent.
300 template <typename T, typename U>
301 struct CheckedRshOp<T,
302                     U,
303                     typename std::enable_if<std::is_integral<T>::value &&
304                                             std::is_integral<U>::value>::type> {
305   using result_type = T;
306   template <typename V>
307   static bool Do(T x, U shift, V* result) {
308     // Use the type conversion push negative values out of range.
309     if (BASE_NUMERICS_LIKELY(as_unsigned(shift) <
310                              IntegerBitsPlusSign<T>::value)) {
311       T tmp = x >> shift;
312       *result = static_cast<V>(tmp);
313       return IsValueInRangeForNumericType<V>(tmp);
314     }
315     return false;
316   }
317 };
318 
319 template <typename T, typename U, class Enable = void>
320 struct CheckedAndOp {};
321 
322 // For simplicity we support only unsigned integer results.
323 template <typename T, typename U>
324 struct CheckedAndOp<T,
325                     U,
326                     typename std::enable_if<std::is_integral<T>::value &&
327                                             std::is_integral<U>::value>::type> {
328   using result_type = typename std::make_unsigned<
329       typename MaxExponentPromotion<T, U>::type>::type;
330   template <typename V>
331   static constexpr bool Do(T x, U y, V* result) {
332     result_type tmp = static_cast<result_type>(x) & static_cast<result_type>(y);
333     *result = static_cast<V>(tmp);
334     return IsValueInRangeForNumericType<V>(tmp);
335   }
336 };
337 
338 template <typename T, typename U, class Enable = void>
339 struct CheckedOrOp {};
340 
341 // For simplicity we support only unsigned integers.
342 template <typename T, typename U>
343 struct CheckedOrOp<T,
344                    U,
345                    typename std::enable_if<std::is_integral<T>::value &&
346                                            std::is_integral<U>::value>::type> {
347   using result_type = typename std::make_unsigned<
348       typename MaxExponentPromotion<T, U>::type>::type;
349   template <typename V>
350   static constexpr bool Do(T x, U y, V* result) {
351     result_type tmp = static_cast<result_type>(x) | static_cast<result_type>(y);
352     *result = static_cast<V>(tmp);
353     return IsValueInRangeForNumericType<V>(tmp);
354   }
355 };
356 
357 template <typename T, typename U, class Enable = void>
358 struct CheckedXorOp {};
359 
360 // For simplicity we support only unsigned integers.
361 template <typename T, typename U>
362 struct CheckedXorOp<T,
363                     U,
364                     typename std::enable_if<std::is_integral<T>::value &&
365                                             std::is_integral<U>::value>::type> {
366   using result_type = typename std::make_unsigned<
367       typename MaxExponentPromotion<T, U>::type>::type;
368   template <typename V>
369   static constexpr bool Do(T x, U y, V* result) {
370     result_type tmp = static_cast<result_type>(x) ^ static_cast<result_type>(y);
371     *result = static_cast<V>(tmp);
372     return IsValueInRangeForNumericType<V>(tmp);
373   }
374 };
375 
376 // Max doesn't really need to be implemented this way because it can't fail,
377 // but it makes the code much cleaner to use the MathOp wrappers.
378 template <typename T, typename U, class Enable = void>
379 struct CheckedMaxOp {};
380 
381 template <typename T, typename U>
382 struct CheckedMaxOp<
383     T,
384     U,
385     typename std::enable_if<std::is_arithmetic<T>::value &&
386                             std::is_arithmetic<U>::value>::type> {
387   using result_type = typename MaxExponentPromotion<T, U>::type;
388   template <typename V>
389   static constexpr bool Do(T x, U y, V* result) {
390     result_type tmp = IsGreater<T, U>::Test(x, y) ? static_cast<result_type>(x)
391                                                   : static_cast<result_type>(y);
392     *result = static_cast<V>(tmp);
393     return IsValueInRangeForNumericType<V>(tmp);
394   }
395 };
396 
397 // Min doesn't really need to be implemented this way because it can't fail,
398 // but it makes the code much cleaner to use the MathOp wrappers.
399 template <typename T, typename U, class Enable = void>
400 struct CheckedMinOp {};
401 
402 template <typename T, typename U>
403 struct CheckedMinOp<
404     T,
405     U,
406     typename std::enable_if<std::is_arithmetic<T>::value &&
407                             std::is_arithmetic<U>::value>::type> {
408   using result_type = typename LowestValuePromotion<T, U>::type;
409   template <typename V>
410   static constexpr bool Do(T x, U y, V* result) {
411     result_type tmp = IsLess<T, U>::Test(x, y) ? static_cast<result_type>(x)
412                                                : static_cast<result_type>(y);
413     *result = static_cast<V>(tmp);
414     return IsValueInRangeForNumericType<V>(tmp);
415   }
416 };
417 
418 // This is just boilerplate that wraps the standard floating point arithmetic.
419 // A macro isn't the nicest solution, but it beats rewriting these repeatedly.
420 #define BASE_FLOAT_ARITHMETIC_OPS(NAME, OP)                              \
421   template <typename T, typename U>                                      \
422   struct Checked##NAME##Op<                                              \
423       T, U,                                                              \
424       typename std::enable_if<std::is_floating_point<T>::value ||        \
425                               std::is_floating_point<U>::value>::type> { \
426     using result_type = typename MaxExponentPromotion<T, U>::type;       \
427     template <typename V>                                                \
428     static constexpr bool Do(T x, U y, V* result) {                      \
429       using Promotion = typename MaxExponentPromotion<T, U>::type;       \
430       Promotion presult = x OP y;                                        \
431       *result = static_cast<V>(presult);                                 \
432       return IsValueInRangeForNumericType<V>(presult);                   \
433     }                                                                    \
434   };
435 
436 BASE_FLOAT_ARITHMETIC_OPS(Add, +)
437 BASE_FLOAT_ARITHMETIC_OPS(Sub, -)
438 BASE_FLOAT_ARITHMETIC_OPS(Mul, *)
439 BASE_FLOAT_ARITHMETIC_OPS(Div, /)
440 
441 #undef BASE_FLOAT_ARITHMETIC_OPS
442 
443 // Floats carry around their validity state with them, but integers do not. So,
444 // we wrap the underlying value in a specialization in order to hide that detail
445 // and expose an interface via accessors.
446 enum NumericRepresentation {
447   NUMERIC_INTEGER,
448   NUMERIC_FLOATING,
449   NUMERIC_UNKNOWN
450 };
451 
452 template <typename NumericType>
453 struct GetNumericRepresentation {
454   static const NumericRepresentation value =
455       std::is_integral<NumericType>::value
456           ? NUMERIC_INTEGER
457           : (std::is_floating_point<NumericType>::value ? NUMERIC_FLOATING
458                                                         : NUMERIC_UNKNOWN);
459 };
460 
461 template <typename T,
462           NumericRepresentation type = GetNumericRepresentation<T>::value>
463 class CheckedNumericState {};
464 
465 // Integrals require quite a bit of additional housekeeping to manage state.
466 template <typename T>
467 class CheckedNumericState<T, NUMERIC_INTEGER> {
468  private:
469   // is_valid_ precedes value_ because member initializers in the constructors
470   // are evaluated in field order, and is_valid_ must be read when initializing
471   // value_.
472   bool is_valid_;
473   T value_;
474 
475   // Ensures that a type conversion does not trigger undefined behavior.
476   template <typename Src>
477   static constexpr T WellDefinedConversionOrZero(const Src value,
478                                                  const bool is_valid) {
479     using SrcType = typename internal::UnderlyingType<Src>::type;
480     return (std::is_integral<SrcType>::value || is_valid)
481                ? static_cast<T>(value)
482                : static_cast<T>(0);
483   }
484 
485  public:
486   template <typename Src, NumericRepresentation type>
487   friend class CheckedNumericState;
488 
489   constexpr CheckedNumericState() : is_valid_(true), value_(0) {}
490 
491   template <typename Src>
492   constexpr CheckedNumericState(Src value, bool is_valid)
493       : is_valid_(is_valid && IsValueInRangeForNumericType<T>(value)),
494         value_(WellDefinedConversionOrZero(value, is_valid_)) {
495     static_assert(std::is_arithmetic<Src>::value, "Argument must be numeric.");
496   }
497 
498   // Copy constructor.
499   template <typename Src>
500   constexpr CheckedNumericState(const CheckedNumericState<Src>& rhs)
501       : is_valid_(rhs.IsValid()),
502         value_(WellDefinedConversionOrZero(rhs.value(), is_valid_)) {}
503 
504   template <typename Src>
505   constexpr explicit CheckedNumericState(Src value)
506       : is_valid_(IsValueInRangeForNumericType<T>(value)),
507         value_(WellDefinedConversionOrZero(value, is_valid_)) {}
508 
509   constexpr bool is_valid() const { return is_valid_; }
510   constexpr T value() const { return value_; }
511 };
512 
513 // Floating points maintain their own validity, but need translation wrappers.
514 template <typename T>
515 class CheckedNumericState<T, NUMERIC_FLOATING> {
516  private:
517   T value_;
518 
519   // Ensures that a type conversion does not trigger undefined behavior.
520   template <typename Src>
521   static constexpr T WellDefinedConversionOrNaN(const Src value,
522                                                 const bool is_valid) {
523     using SrcType = typename internal::UnderlyingType<Src>::type;
524     return (StaticDstRangeRelationToSrcRange<T, SrcType>::value ==
525                 NUMERIC_RANGE_CONTAINED ||
526             is_valid)
527                ? static_cast<T>(value)
528                : std::numeric_limits<T>::quiet_NaN();
529   }
530 
531  public:
532   template <typename Src, NumericRepresentation type>
533   friend class CheckedNumericState;
534 
535   constexpr CheckedNumericState() : value_(0.0) {}
536 
537   template <typename Src>
538   constexpr CheckedNumericState(Src value, bool is_valid)
539       : value_(WellDefinedConversionOrNaN(value, is_valid)) {}
540 
541   template <typename Src>
542   constexpr explicit CheckedNumericState(Src value)
543       : value_(WellDefinedConversionOrNaN(
544             value,
545             IsValueInRangeForNumericType<T>(value))) {}
546 
547   // Copy constructor.
548   template <typename Src>
549   constexpr CheckedNumericState(const CheckedNumericState<Src>& rhs)
550       : value_(WellDefinedConversionOrNaN(
551             rhs.value(),
552             rhs.is_valid() && IsValueInRangeForNumericType<T>(rhs.value()))) {}
553 
554   constexpr bool is_valid() const {
555     // Written this way because std::isfinite is not reliably constexpr.
556     return MustTreatAsConstexpr(value_)
557                ? value_ <= std::numeric_limits<T>::max() &&
558                      value_ >= std::numeric_limits<T>::lowest()
559                : std::isfinite(value_);
560   }
561   constexpr T value() const { return value_; }
562 };
563 
564 }  // namespace internal
565 }  // namespace base
566 
567 #endif  // BASE_NUMERICS_CHECKED_MATH_IMPL_H_
568