mirror of
https://github.com/yuzu-emu/ext-boost.git
synced 2024-12-22 18:55:42 +00:00
820 lines
26 KiB
C++
820 lines
26 KiB
C++
#ifndef BOOST_NUMERIC_CHECKED_INTEGER_HPP
|
|
#define BOOST_NUMERIC_CHECKED_INTEGER_HPP
|
|
|
|
// Copyright (c) 2012 Robert Ramey
|
|
//
|
|
// Distributed under the Boost Software License, Version 1.0. (See
|
|
// accompanying file LICENSE_1_0.txt or copy at
|
|
// http://www.boost.org/LICENSE_1_0.txt)
|
|
|
|
// contains operations for doing checked aritmetic on NATIVE
|
|
// C++ types.
|
|
|
|
#include <limits>
|
|
#include <type_traits> // is_integral, make_unsigned, enable_if
|
|
#include <algorithm> // std::max
|
|
|
|
#include "checked_result.hpp"
|
|
#include "checked_default.hpp"
|
|
#include "safe_compare.hpp"
|
|
#include "utility.hpp"
|
|
#include "exception.hpp"
|
|
|
|
namespace boost {
|
|
namespace safe_numerics {
|
|
|
|
// utility
|
|
|
|
template<bool tf>
|
|
using bool_type = typename std::conditional<tf, std::true_type, std::false_type>::type;
|
|
|
|
////////////////////////////////////////////////////
|
|
// layer 0 - implement safe operations for intrinsic integers
|
|
// Note presumption of twos complement integer arithmetic
|
|
|
|
// convert an integral value to some other integral type
|
|
template<
|
|
typename R,
|
|
R Min,
|
|
R Max,
|
|
typename T,
|
|
class F
|
|
>
|
|
struct heterogeneous_checked_operation<
|
|
R,
|
|
Min,
|
|
Max,
|
|
T,
|
|
F,
|
|
typename std::enable_if<
|
|
std::is_integral<R>::value
|
|
&& std::is_integral<T>::value
|
|
>::type
|
|
>{
|
|
////////////////////////////////////////////////////
|
|
// safe casting on primitive types
|
|
|
|
struct cast_impl_detail {
|
|
constexpr static checked_result<R>
|
|
cast_impl(
|
|
const T & t,
|
|
std::true_type, // R is signed
|
|
std::true_type // T is signed
|
|
){
|
|
// INT32-C Ensure that operations on signed
|
|
// integers do not overflow
|
|
return
|
|
boost::safe_numerics::safe_compare::greater_than(
|
|
t,
|
|
Max
|
|
) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"converted signed value too large"
|
|
)
|
|
:
|
|
boost::safe_numerics::safe_compare::less_than(
|
|
t,
|
|
Min
|
|
) ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"converted signed value too small"
|
|
)
|
|
:
|
|
checked_result<R>(static_cast<R>(t))
|
|
;
|
|
}
|
|
constexpr static checked_result<R>
|
|
cast_impl(
|
|
const T & t,
|
|
std::true_type, // R is signed
|
|
std::false_type // T is unsigned
|
|
){
|
|
// INT30-C Ensure that unsigned integer operations
|
|
// do not wrap
|
|
return
|
|
boost::safe_numerics::safe_compare::greater_than(
|
|
t,
|
|
Max
|
|
) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"converted unsigned value too large"
|
|
)
|
|
:
|
|
checked_result<R>(static_cast<R>(t))
|
|
;
|
|
}
|
|
constexpr static checked_result<R>
|
|
cast_impl(
|
|
const T & t,
|
|
std::false_type, // R is unsigned
|
|
std::false_type // T is unsigned
|
|
){
|
|
// INT32-C Ensure that operations on unsigned
|
|
// integers do not overflow
|
|
return
|
|
boost::safe_numerics::safe_compare::greater_than(
|
|
t,
|
|
Max
|
|
) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"converted unsigned value too large"
|
|
)
|
|
:
|
|
checked_result<R>(static_cast<R>(t))
|
|
;
|
|
}
|
|
constexpr static checked_result<R>
|
|
cast_impl(
|
|
const T & t,
|
|
std::false_type, // R is unsigned
|
|
std::true_type // T is signed
|
|
){
|
|
return
|
|
boost::safe_numerics::safe_compare::less_than(t, 0) ?
|
|
F::template invoke<safe_numerics_error::domain_error>(
|
|
"converted negative value to unsigned"
|
|
)
|
|
:
|
|
boost::safe_numerics::safe_compare::greater_than(
|
|
t,
|
|
Max
|
|
) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"converted signed value too large"
|
|
)
|
|
:
|
|
checked_result<R>(static_cast<R>(t))
|
|
;
|
|
}
|
|
}; // cast_impl_detail
|
|
|
|
constexpr static checked_result<R>
|
|
cast(const T & t){
|
|
return
|
|
cast_impl_detail::cast_impl(
|
|
t,
|
|
std::is_signed<R>(),
|
|
std::is_signed<T>()
|
|
);
|
|
}
|
|
};
|
|
|
|
// converting floating point value to integral type
|
|
template<
|
|
typename R,
|
|
R Min,
|
|
R Max,
|
|
typename T,
|
|
class F
|
|
>
|
|
struct heterogeneous_checked_operation<
|
|
R,
|
|
Min,
|
|
Max,
|
|
T,
|
|
F,
|
|
typename std::enable_if<
|
|
std::is_integral<R>::value
|
|
&& std::is_floating_point<T>::value
|
|
>::type
|
|
>{
|
|
constexpr static checked_result<R>
|
|
cast(const T & t){
|
|
return static_cast<R>(t);
|
|
}
|
|
};
|
|
|
|
// converting integral value to floating point type
|
|
|
|
// INT35-C. Use correct integer precisions
|
|
template<
|
|
typename R,
|
|
R Min,
|
|
R Max,
|
|
typename T,
|
|
class F
|
|
>
|
|
struct heterogeneous_checked_operation<
|
|
R,
|
|
Min,
|
|
Max,
|
|
T,
|
|
F,
|
|
typename std::enable_if<
|
|
std::is_floating_point<R>::value
|
|
&& std::is_integral<T>::value
|
|
>::type
|
|
>{
|
|
constexpr static checked_result<R>
|
|
cast(const T & t){
|
|
if(std::numeric_limits<R>::digits < std::numeric_limits<T>::digits){
|
|
if(utility::significant_bits(t) > std::numeric_limits<R>::digits){
|
|
return F::invoke(
|
|
safe_numerics_error::precision_overflow_error,
|
|
"keep precision"
|
|
);
|
|
}
|
|
}
|
|
return t;
|
|
}
|
|
};
|
|
|
|
// binary operations on primitive integer types
|
|
|
|
template<
|
|
typename R,
|
|
class F
|
|
>
|
|
struct checked_operation<R, F,
|
|
typename std::enable_if<
|
|
std::is_integral<R>::value
|
|
>::type
|
|
>{
|
|
////////////////////////////////////////////////////
|
|
// safe addition on primitive types
|
|
|
|
struct add_impl_detail {
|
|
// result unsigned
|
|
constexpr static checked_result<R> add(
|
|
const R t,
|
|
const R u,
|
|
std::false_type // R unsigned
|
|
){
|
|
return
|
|
// INT30-C. Ensure that unsigned integer operations do not wrap
|
|
std::numeric_limits<R>::max() - u < t ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"addition result too large"
|
|
)
|
|
:
|
|
checked_result<R>(t + u)
|
|
;
|
|
}
|
|
|
|
// result signed
|
|
constexpr static checked_result<R> add(
|
|
const R t,
|
|
const R u,
|
|
std::true_type // R signed
|
|
){
|
|
// INT32-C. Ensure that operations on signed integers do not result in overflow
|
|
return
|
|
// INT32-C. Ensure that operations on signed integers do not result in overflow
|
|
((u > 0) && (t > (std::numeric_limits<R>::max() - u))) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"addition result too large"
|
|
)
|
|
:
|
|
((u < 0) && (t < (std::numeric_limits<R>::min() - u))) ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"addition result too low"
|
|
)
|
|
:
|
|
checked_result<R>(t + u)
|
|
;
|
|
}
|
|
}; // add_impl_detail
|
|
|
|
constexpr static checked_result<R>
|
|
add(const R & t, const R & u){
|
|
return add_impl_detail::add(t, u, std::is_signed<R>());
|
|
}
|
|
|
|
////////////////////////////////////////////////////
|
|
// safe subtraction on primitive types
|
|
struct subtract_impl_detail {
|
|
|
|
// result unsigned
|
|
constexpr static checked_result<R> subtract(
|
|
const R t,
|
|
const R u,
|
|
std::false_type // R is unsigned
|
|
){
|
|
// INT30-C. Ensure that unsigned integer operations do not wrap
|
|
return
|
|
t < u ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"subtraction result cannot be negative"
|
|
)
|
|
:
|
|
checked_result<R>(t - u)
|
|
;
|
|
}
|
|
|
|
// result signed
|
|
constexpr static checked_result<R> subtract(
|
|
const R t,
|
|
const R u,
|
|
std::true_type // R is signed
|
|
){ // INT32-C
|
|
return
|
|
// INT32-C. Ensure that operations on signed integers do not result in overflow
|
|
((u > 0) && (t < (std::numeric_limits<R>::min() + u))) ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"subtraction result overflows result type"
|
|
)
|
|
:
|
|
((u < 0) && (t > (std::numeric_limits<R>::max() + u))) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"subtraction result overflows result type"
|
|
)
|
|
:
|
|
checked_result<R>(t - u)
|
|
;
|
|
}
|
|
|
|
}; // subtract_impl_detail
|
|
|
|
constexpr static checked_result<R> subtract(const R & t, const R & u){
|
|
return subtract_impl_detail::subtract(t, u, std::is_signed<R>());
|
|
}
|
|
|
|
////////////////////////////////////////////////////
|
|
// safe minus on primitive types
|
|
struct minus_impl_detail {
|
|
|
|
// result unsigned
|
|
constexpr static checked_result<R> minus(
|
|
const R t,
|
|
std::false_type // R is unsigned
|
|
){
|
|
return t > 0 ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"minus unsigned would be negative"
|
|
)
|
|
:
|
|
// t == 0
|
|
checked_result<R>(0)
|
|
;
|
|
}
|
|
|
|
// result signed
|
|
constexpr static checked_result<R> minus(
|
|
const R t,
|
|
std::true_type // R is signed
|
|
){ // INT32-C
|
|
return t == std::numeric_limits<R>::min() ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"subtraction result overflows result type"
|
|
)
|
|
:
|
|
checked_result<R>(-t)
|
|
;
|
|
}
|
|
|
|
}; // minus_impl_detail
|
|
|
|
constexpr static checked_result<R> minus(const R & t){
|
|
return minus_impl_detail::minus(t, std::is_signed<R>());
|
|
}
|
|
|
|
////////////////////////////////////////////////////
|
|
// safe multiplication on primitive types
|
|
|
|
struct multiply_impl_detail {
|
|
|
|
// result unsigned
|
|
constexpr static checked_result<R> multiply(
|
|
const R t,
|
|
const R u,
|
|
std::false_type, // R is unsigned
|
|
std::false_type // !(sizeochecked_result<R>R) > sizeochecked_result<R>std::uintmax_t) / 2)
|
|
|
|
){
|
|
// INT30-C
|
|
// fast method using intermediate result guaranteed not to overflow
|
|
// todo - replace std::uintmax_t with a size double the size of R
|
|
using i_type = std::uintmax_t;
|
|
return
|
|
static_cast<i_type>(t) * static_cast<i_type>(u)
|
|
> std::numeric_limits<R>::max() ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
;
|
|
}
|
|
constexpr static checked_result<R> multiply(
|
|
const R t,
|
|
const R u,
|
|
std::false_type, // R is unsigned
|
|
std::true_type // (sizeochecked_result<R>R) > sizeochecked_result<R>std::uintmax_t) / 2)
|
|
|
|
){
|
|
// INT30-C
|
|
return
|
|
u > 0 && t > std::numeric_limits<R>::max() / u ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
;
|
|
}
|
|
|
|
// result signed
|
|
constexpr static checked_result<R> multiply(
|
|
const R t,
|
|
const R u,
|
|
std::true_type, // R is signed
|
|
std::false_type // ! (sizeochecked_result<R>R) > (sizeochecked_result<R>std::intmax_t) / 2))
|
|
|
|
){
|
|
// INT30-C
|
|
// fast method using intermediate result guaranteed not to overflow
|
|
// todo - replace std::intmax_t with a size double the size of R
|
|
using i_type = std::intmax_t;
|
|
return
|
|
(
|
|
static_cast<i_type>(t) * static_cast<i_type>(u)
|
|
> static_cast<i_type>(std::numeric_limits<R>::max())
|
|
) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
(
|
|
static_cast<i_type>(t) * static_cast<i_type>(u)
|
|
< static_cast<i_type>(std::numeric_limits<R>::min())
|
|
) ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
;
|
|
}
|
|
constexpr static checked_result<R> multiply(
|
|
const R t,
|
|
const R u,
|
|
std::true_type, // R is signed
|
|
std::true_type // (sizeochecked_result<R>R) > (sizeochecked_result<R>std::intmax_t) / 2))
|
|
){ // INT32-C
|
|
return t > 0 ?
|
|
u > 0 ?
|
|
t > std::numeric_limits<R>::max() / u ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
: // u <= 0
|
|
u < std::numeric_limits<R>::min() / t ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
: // t <= 0
|
|
u > 0 ?
|
|
t < std::numeric_limits<R>::min() / u ?
|
|
F::template invoke<safe_numerics_error::negative_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
: // u <= 0
|
|
t != 0 && u < std::numeric_limits<R>::max() / t ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"multiplication overflow"
|
|
)
|
|
:
|
|
checked_result<R>(t * u)
|
|
;
|
|
}
|
|
}; // multiply_impl_detail
|
|
|
|
constexpr static checked_result<R> multiply(const R & t, const R & u){
|
|
return multiply_impl_detail::multiply(
|
|
t,
|
|
u,
|
|
std::is_signed<R>(),
|
|
std::integral_constant<
|
|
bool,
|
|
(sizeof(R) > sizeof(std::uintmax_t) / 2)
|
|
>()
|
|
);
|
|
}
|
|
|
|
////////////////////////////////
|
|
// safe division on unsafe types
|
|
|
|
struct divide_impl_detail {
|
|
constexpr static checked_result<R> divide(
|
|
const R & t,
|
|
const R & u,
|
|
std::false_type // R is unsigned
|
|
){
|
|
return t / u;
|
|
}
|
|
|
|
constexpr static checked_result<R> divide(
|
|
const R & t,
|
|
const R & u,
|
|
std::true_type // R is signed
|
|
){
|
|
return
|
|
(u == -1 && t == std::numeric_limits<R>::min()) ?
|
|
F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"result cannot be represented"
|
|
)
|
|
:
|
|
checked_result<R>(t / u)
|
|
;
|
|
}
|
|
}; // divide_impl_detail
|
|
|
|
// note that we presume that the size of R >= size of T
|
|
constexpr static checked_result<R> divide(const R & t, const R & u){
|
|
if(u == 0){
|
|
return F::template invoke<safe_numerics_error::domain_error>(
|
|
"divide by zero"
|
|
);
|
|
}
|
|
return divide_impl_detail::divide(t, u, std::is_signed<R>());
|
|
}
|
|
|
|
////////////////////////////////
|
|
// safe modulus on unsafe types
|
|
|
|
struct modulus_impl_detail {
|
|
constexpr static checked_result<R> modulus(
|
|
const R & t,
|
|
const R & u,
|
|
std::false_type // R is unsigned
|
|
){
|
|
return t % u;
|
|
}
|
|
|
|
constexpr static checked_result<R> modulus(
|
|
const R & t,
|
|
const R & u,
|
|
std::true_type // R is signed
|
|
){
|
|
if(u >= 0)
|
|
return t % u;
|
|
checked_result<R> ux = checked::minus(u);
|
|
if(ux.exception())
|
|
return t;
|
|
return t % static_cast<R>(ux);
|
|
}
|
|
}; // modulus_impl_detail
|
|
|
|
constexpr static checked_result<R> modulus(const R & t, const R & u){
|
|
if(0 == u)
|
|
return F::template invoke<safe_numerics_error::domain_error>(
|
|
"denominator is zero"
|
|
);
|
|
|
|
// why to we need abs here? the sign of the modulus is the sign of the
|
|
// dividend. Consider -128 % -1 The result of this operation should be -1
|
|
// but if I use t % u the x86 hardware uses the divide instruction
|
|
// capturing the modulus as a side effect. When it does this, it
|
|
// invokes the operation -128 / -1 -> 128 which overflows a signed type
|
|
// and provokes a hardware exception. We can fix this using abs()
|
|
// since -128 % -1 = -128 % 1 = 0
|
|
return modulus_impl_detail::modulus(t, u, typename std::is_signed<R>::type());
|
|
}
|
|
|
|
///////////////////////////////////
|
|
// shift operations
|
|
|
|
struct left_shift_integer_detail {
|
|
|
|
#if 0
|
|
// todo - optimize for gcc to exploit builtin
|
|
/* for gcc compilers
|
|
int __builtin_clz (unsigned int x)
|
|
Returns the number of leading 0-bits in x, starting at the
|
|
most significant bit position. If x is 0, the result is undefined.
|
|
*/
|
|
|
|
#ifndef __has_feature // Optional of course.
|
|
#define __has_feature(x) 0 // Compatibility with non-clang compilers.
|
|
#endif
|
|
|
|
template<typename T>
|
|
constexpr unsigned int leading_zeros(const T & t){
|
|
if(0 == t)
|
|
return 0;
|
|
#if __has_feature(builtin_clz)
|
|
return __builtin_clz(t);
|
|
#else
|
|
#endif
|
|
}
|
|
#endif
|
|
|
|
// INT34-C C++
|
|
|
|
// standard paragraph 5.8 / 2
|
|
// The value of E1 << E2 is E1 left-shifted E2 bit positions;
|
|
// vacated bits are zero-filled.
|
|
constexpr static checked_result<R> left_shift(
|
|
const R & t,
|
|
const R & u,
|
|
std::false_type // R is unsigned
|
|
){
|
|
// the value of the result is E1 x 2^E2, reduced modulo one more than
|
|
// the maximum value representable in the result type.
|
|
|
|
// see 5.8 & 1
|
|
// if right operand is
|
|
// greater than or equal to the length in bits of the promoted left operand.
|
|
if(
|
|
safe_compare::greater_than(
|
|
u,
|
|
std::numeric_limits<R>::digits - utility::significant_bits(t)
|
|
)
|
|
){
|
|
// behavior is undefined
|
|
return F::template invoke<safe_numerics_error::shift_too_large>(
|
|
"shifting left more bits than available is undefined behavior"
|
|
);
|
|
}
|
|
return t << u;
|
|
}
|
|
|
|
constexpr static checked_result<R> left_shift(
|
|
const R & t,
|
|
const R & u,
|
|
std::true_type // R is signed
|
|
){
|
|
// and [E1] has a non-negative value
|
|
if(t >= 0){
|
|
// and E1 x 2^E2 is representable in the corresponding
|
|
// unsigned type of the result type,
|
|
|
|
// see 5.8 & 1
|
|
// if right operand is
|
|
// greater than or equal to the length in bits of the promoted left operand.
|
|
if(
|
|
safe_compare::greater_than(
|
|
u,
|
|
std::numeric_limits<R>::digits - utility::significant_bits(t)
|
|
)
|
|
){
|
|
// behavior is undefined
|
|
return F::template invoke<safe_numerics_error::shift_too_large>(
|
|
"shifting left more bits than available"
|
|
);
|
|
}
|
|
else{
|
|
return t << u;
|
|
}
|
|
}
|
|
// otherwise, the behavior is undefined.
|
|
return F::template invoke<safe_numerics_error::negative_shift>(
|
|
"shifting a negative value"
|
|
);
|
|
}
|
|
|
|
}; // left_shift_integer_detail
|
|
|
|
constexpr static checked_result<R> left_shift(
|
|
const R & t,
|
|
const R & u
|
|
){
|
|
// INT34-C - Do not shift an expression by a negative number of bits
|
|
|
|
// standard paragraph 5.8 & 1
|
|
// if the right operand is negative
|
|
if(u == 0){
|
|
return t;
|
|
}
|
|
if(u < 0){
|
|
return F::template invoke<safe_numerics_error::negative_shift>(
|
|
"shifting negative amount"
|
|
);
|
|
}
|
|
if(u > std::numeric_limits<R>::digits){
|
|
// behavior is undefined
|
|
return F::template invoke<safe_numerics_error::shift_too_large>(
|
|
"shifting more bits than available"
|
|
);
|
|
}
|
|
return left_shift_integer_detail::left_shift(t, u, std::is_signed<R>());
|
|
}
|
|
|
|
// right shift
|
|
|
|
struct right_shift_integer_detail {
|
|
|
|
// INT34-C C++
|
|
|
|
// standard paragraph 5.8 / 3
|
|
// The value of E1 << E2 is E1 left-shifted E2 bit positions;
|
|
// vacated bits are zero-filled.
|
|
constexpr static checked_result<R> right_shift(
|
|
const R & t,
|
|
const R & u,
|
|
std::false_type // T is unsigned
|
|
){
|
|
// the value of the result is the integral part of the
|
|
// quotient of E1/2E2
|
|
return t >> u;
|
|
}
|
|
|
|
constexpr static checked_result<R> right_shift(
|
|
const R & t,
|
|
const R & u,
|
|
std::true_type // T is signed;
|
|
){
|
|
if(t < 0){
|
|
// note that the C++ standard considers this case is "implemenation
|
|
// defined" rather than "undefined".
|
|
return F::template invoke<safe_numerics_error::negative_value_shift>(
|
|
"shifting a negative value"
|
|
);
|
|
}
|
|
|
|
// the value is the integral part of E1 / 2^E2,
|
|
return t >> u;
|
|
}
|
|
}; // right_shift_integer_detail
|
|
|
|
constexpr static checked_result<R> right_shift(
|
|
const R & t,
|
|
const R & u
|
|
){
|
|
// INT34-C - Do not shift an expression by a negative number of bits
|
|
|
|
// standard paragraph 5.8 & 1
|
|
// if the right operand is negative
|
|
if(u < 0){
|
|
return F::template invoke<safe_numerics_error::negative_shift>(
|
|
"shifting negative amount"
|
|
);
|
|
}
|
|
if(u > std::numeric_limits<R>::digits){
|
|
// behavior is undefined
|
|
return F::template invoke<safe_numerics_error::shift_too_large>(
|
|
"shifting more bits than available"
|
|
);
|
|
}
|
|
return right_shift_integer_detail::right_shift(t, u ,std::is_signed<R>());
|
|
}
|
|
|
|
///////////////////////////////////
|
|
// bitwise operations
|
|
|
|
// INT13-C Note: We don't enforce recommendation as acually written
|
|
// as it would break too many programs. Specifically, we permit signed
|
|
// integer operands.
|
|
|
|
constexpr static checked_result<R> bitwise_or(const R & t, const R & u){
|
|
using namespace boost::safe_numerics::utility;
|
|
const unsigned int result_size
|
|
= std::max(significant_bits(t), significant_bits(u));
|
|
|
|
if(result_size > bits_type<R>::value){
|
|
return F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"result type too small to hold bitwise or"
|
|
);
|
|
}
|
|
return t | u;
|
|
}
|
|
|
|
constexpr static checked_result<R> bitwise_xor(const R & t, const R & u){
|
|
using namespace boost::safe_numerics::utility;
|
|
const unsigned int result_size
|
|
= std::max(significant_bits(t), significant_bits(u));
|
|
|
|
if(result_size > bits_type<R>::value){
|
|
return F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"result type too small to hold bitwise or"
|
|
);
|
|
}
|
|
return t ^ u;
|
|
}
|
|
|
|
constexpr static checked_result<R> bitwise_and(const R & t, const R & u){
|
|
using namespace boost::safe_numerics::utility;
|
|
const unsigned int result_size
|
|
= std::min(significant_bits(t), significant_bits(u));
|
|
|
|
if(result_size > bits_type<R>::value){
|
|
return F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"result type too small to hold bitwise and"
|
|
);
|
|
}
|
|
return t & u;
|
|
}
|
|
|
|
constexpr static checked_result<R> bitwise_not(const R & t){
|
|
using namespace boost::safe_numerics::utility;
|
|
|
|
if(significant_bits(t) > bits_type<R>::value){
|
|
return F::template invoke<safe_numerics_error::positive_overflow_error>(
|
|
"result type too small to hold bitwise inverse"
|
|
);
|
|
}
|
|
return ~t;
|
|
}
|
|
|
|
}; // checked_operation
|
|
} // safe_numerics
|
|
} // boost
|
|
|
|
#endif // BOOST_NUMERIC_CHECKED_INTEGER_HPP
|