Arithmetic & Logic Operations

Header aarith/float/float_operations.hpp

namespace aarith

Functions

template<size_t E, size_t M, class Function_add, class Function_sub>
auto add_(const floating_point<E, M> lhs, const floating_point<E, M> rhs, Function_add fun_add, Function_sub fun_sub) -> floating_point<E, M>

Generic addition of two floating_point values.

This method computes the sum of two floating-point values using the provided functions fun_add and fun_sub to compute the new mantissa. This generic function allows to easily implement own adders, e.g. to develop new hardware implementations.

Note

As an end-user of aarith, you will, most likely, never need to call this function.

Template Parameters
  • E – Exponent width

  • M – Mantissa width

  • Function_add – Function object type for performing an addition

  • Function_sub – Function object fype for performing a subtraction

Parameters
  • lhs – Left-hand side argument of the usm

  • rhs – Right-hand side argument of the sum

  • fun_add – Function performing the addition of the mantissae

  • fun_sub – Function performing the subtraction of the mantissae

Returns

The sum of lhs + rhs using the provided functions

template<size_t E, size_t M, class Function_add, class Function_sub>
auto sub_(const floating_point<E, M> lhs, const floating_point<E, M> rhs, Function_add fun_add, Function_sub fun_sub) -> floating_point<E, M>

Generic subtraction of two floating_point values.

This method computes the difference of two floating-point values using the provided functions fun_add and fun_sub to compute the new mantissa. This generic function allows to easily implement own adders, e.g. to develop new hardware implementations.*

Note

As an end-user of aarith, you will, most likely, never need to call this function.

Template Parameters
  • E – Exponent width

  • M – Mantissa width

  • Function_add – Function object type for performing an addition

  • Function_sub – Function object fype for performing a subtraction

Parameters
  • lhs – Left-hand side argument of the usm

  • rhs – Right-hand side argument of the sum

  • fun_add – Function performing the addition of the mantissae

  • fun_sub – Function performing the subtraction of the mantissae

Returns

The sum of lhs + rhs using the provided functions

template<size_t E, size_t M>
auto add(const floating_point<E, M> lhs, const floating_point<E, M> rhs) -> floating_point<E, M>

Adds two floating_point values.

Parameters
  • lhs – The first number that is to be summed up

  • rhs – The second number that is to be summed up

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa including the leading 1

Returns

The sum

template<size_t E, size_t M>
auto sub(const floating_point<E, M> lhs, const floating_point<E, M> rhs) -> floating_point<E, M>

Subtract two floating_point values.

Parameters
  • lhs – The minuend

  • rhs – The subtrahend

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa including the leading 1

Returns

The difference lhs-rhs

template<size_t E, size_t M, typename WordType>
auto mul(const floating_point<E, M, WordType> lhs, const floating_point<E, M, WordType> rhs) -> floating_point<E, M, WordType>

Multiplies two floating_point numbers.

Parameters
  • lhs – The multiplicand

  • rhs – The multiplicator

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa including the leading 1

Returns

The product lhs*rhs

template<size_t E, size_t M, typename WordType>
auto div(const floating_point<E, M, WordType> lhs, const floating_point<E, M, WordType> rhs) -> floating_point<E, M, WordType>

Division with floating_points: lhs/rhs.

Parameters
  • lhs – The dividend

  • rhs – The divisor

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa including the leading 1

  • WordType – The word type used to internally store the data

Returns

The quotient lhs/rhs

template<size_t E, size_t M, typename WordType = uint64_t>
constexpr floating_point<E, M, WordType> negate(const floating_point<E, M, WordType> &x)

Computes the negative value of the floating-point number.

Quoting the standard: copies a floating-point operand x to a destination in the same format, reversing the sign bit. negate(x) is not the same as subtraction(0, x)

Note

This method ignores NaN values in the sense that they are also copied and the sign bit flipped.

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa

  • WordType – The word type used to internally store the data

Returns

The negated value of the provided number

template<size_t E, size_t M, typename WordType = uint64_t>
constexpr floating_point<E, M, WordType> copy(const floating_point<E, M, WordType> &x)

Copies the floating-point number.

Quoting the standard: copies a floating-point operand x to a destination in the same format, with no change to the sign bit.

Note

This method ignores NaN values in the sense that they are also copied not signalling any error.

Note

This is a rather useless method that only exists to be more compliant with the IEEE 754 (2019) standard.

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa

  • WordType – The word type used to internally store the data

Returns

The copied value

template<size_t E, size_t M, typename WordType = uint64_t>
constexpr floating_point<E, M, WordType> copySign(const floating_point<E, M, WordType> &x, const floating_point<E, M, WordType> &y)

Copies a floating-point number using the sign of another number.

Quoting the standard: copies a floating-point operand x to a destination in the same format as x, but with the sign bit of y.

Note

This method ignores NaN values in the sense that they are also copied not signalling any error.

Template Parameters
  • E – Width of exponent

  • M – Width of mantissa

  • WordType – The word type used to internally store the data

Returns

The copied value

template<size_t Start, size_t End, size_t E, size_t M, typename WordType>
constexpr word_array<(Start - End) + 1, WordType> bit_range(const floating_point<E, M, WordType> &f)

Extracts a bitstring range from the bit representation of the float.

Note that the indexing is done

  • zero based starting from the LSB

  • is inclusive (i.e. the start and end point are part of the range)

Template Parameters
  • Start – Starting index (inclusive, from left to right)

  • Eend – Ending index (inclusive, from left to right)

  • E – Width of the exponent

  • M – Width of the mantissa

Parameters

f – Float from which the range is taken from

Returns

Range float[End,Start], inclusive

namespace float_operators

This additional nesting of a namespace allows to include aarith without having the usual operator names imported as well.

The use case for this is to allow explicitly replace the conventional arithmetic operations with sepcialized ones. This can, e.g., be used when evaluating approximate operations in the context of neural networks. The name lookup of C++ makes it necessary not to see the operators earlier.

Functions

template<size_t E, size_t M, typename WordType>
auto operator+(const floating_point<E, M, WordType> &lhs, const floating_point<E, M, WordType> &rhs) -> floating_point<E, M, WordType>
template<size_t E, size_t M, typename WordType>
auto operator-(const floating_point<E, M, WordType> &lhs, const floating_point<E, M, WordType> &rhs) -> floating_point<E, M, WordType>
template<size_t E, size_t M, typename WordType>
auto operator*(const floating_point<E, M, WordType> &lhs, const floating_point<E, M, WordType> &rhs) -> floating_point<E, M, WordType>
template<size_t E, size_t M, typename WordType>
auto operator/(const floating_point<E, M, WordType> &lhs, const floating_point<E, M, WordType> &rhs) -> floating_point<E, M, WordType>
template<size_t E, size_t M, typename WordType>
auto operator-(const floating_point<E, M, WordType> &x) -> floating_point<E, M, WordType>