Logo ROOT  
Reference Guide
 
Loading...
Searching...
No Matches
ROOT::Math::KahanSum< T, N > Class Template Reference

template<typename T = double, unsigned int N = 1>
class ROOT::Math::KahanSum< T, N >

The Kahan summation is a compensated summation algorithm, which significantly reduces numerical errors when adding a sequence of finite-precision floating point numbers.

This is done by keeping a separate running compensation (a variable to accumulate small errors).

Auto-vectorisable accumulation

This class can internally use multiple accumulators (template parameter N). When filled from a collection that supports index access from a contiguous block of memory, compilers such as gcc, clang and icc can auto-vectorise the accumulation. This happens by cycling through the internal accumulators based on the value of "`index % N`", so N accumulators can be filled from a block of N numbers in a single instruction.

The usage of multiple accumulators might slightly increase the precision in comparison to the single-accumulator version with N = 1. This depends on the order and magnitude of the numbers being accumulated. Therefore, in rare cases, the accumulation result can change in dependence of N, even when the data are identical. The magnitude of such differences is well below the precision of the floating point type, and will therefore mostly show in the compensation sum(see Carry()). Increasing the number of accumulators therefore only makes sense to speed up the accumulation, but not to increase precision.

Parameters
TThe type of the values to be accumulated.
NNumber of accumulators. Defaults to 1. Ideal values are the widths of a vector register on the relevant architecture. Depending on the instruction set, good values are:
  • AVX2-float: 8
  • AVX2-double: 4
  • AVX512-float: 16
  • AVX512-double: 8

Examples

std::vector<double> numbers(1000);
for (std::size_t i=0; i<1000; ++i) {
numbers[i] = rand();
}
k.Add(numbers.begin(), numbers.end());
// or
k.Add(numbers);
The Kahan summation is a compensated summation algorithm, which significantly reduces numerical error...
Definition Util.h:122
void Add(T x)
Single-element accumulation. Will not vectorise.
Definition Util.h:133
double offset = 10.;
auto result = ROOT::Math::KahanSum<double, 4>::Accumulate(numbers.begin(), numbers.end(), offset);
static KahanSum< T, N > Accumulate(Iterator begin, Iterator end, T initialValue=T{})
Iterate over a range and return an instance of a KahanSum.
Definition Util.h:179

Definition at line 122 of file Util.h.

Public Member Functions

 KahanSum (T initialValue=T{})
 Initialise the sum.
 
template<class Container_t >
void Add (const Container_t &inputs)
 Fill from a container that supports index access.
 
template<class Iterator >
void Add (Iterator begin, Iterator end)
 Accumulate from a range denoted by iterators.
 
void Add (T x)
 Single-element accumulation. Will not vectorise.
 
void AddIndexed (T input, std::size_t index)
 Add input to the sum.
 
Carry () const
 
 operator T () const
 Auto-convert to type T.
 
KahanSum< T, N > & operator+= (T arg)
 Add arg into accumulator. Does not vectorise.
 
Result () const
 
Sum () const
 

Static Public Member Functions

template<class Iterator >
static KahanSum< T, NAccumulate (Iterator begin, Iterator end, T initialValue=T{})
 Iterate over a range and return an instance of a KahanSum.
 

Private Attributes

fCarry [N]
 
fSum [N]
 

#include <Math/Util.h>

Constructor & Destructor Documentation

◆ KahanSum()

template<typename T = double, unsigned int N = 1>
ROOT::Math::KahanSum< T, N >::KahanSum ( initialValue = T{})
inline

Initialise the sum.

Parameters
[in]initialValueInitialise with this value. Defaults to 0.

Definition at line 126 of file Util.h.

Member Function Documentation

◆ Accumulate()

template<typename T = double, unsigned int N = 1>
template<class Iterator >
static KahanSum< T, N > ROOT::Math::KahanSum< T, N >::Accumulate ( Iterator  begin,
Iterator  end,
initialValue = T{} 
)
inlinestatic

Iterate over a range and return an instance of a KahanSum.

See Add(Iterator,Iterator) for details.

Parameters
[in]beginBeginning of a range.
[in]endEnd of the range.
[in]initialValueOptional initial value.

Definition at line 179 of file Util.h.

◆ Add() [1/3]

template<typename T = double, unsigned int N = 1>
template<class Container_t >
void ROOT::Math::KahanSum< T, N >::Add ( const Container_t &  inputs)
inline

Fill from a container that supports index access.

Parameters
[in]inputsContainer with index access such as std::vector or array.

Definition at line 163 of file Util.h.

◆ Add() [2/3]

template<typename T = double, unsigned int N = 1>
template<class Iterator >
void ROOT::Math::KahanSum< T, N >::Add ( Iterator  begin,
Iterator  end 
)
inline

Accumulate from a range denoted by iterators.

This function will auto-vectorise with random-access iterators.

Parameters
[in]beginBeginning of a range. Needs to be a random access iterator for automatic vectorisation, because a contiguous block of memory needs to be read.
[in]endEnd of the range.

Definition at line 148 of file Util.h.

◆ Add() [3/3]

template<typename T = double, unsigned int N = 1>
void ROOT::Math::KahanSum< T, N >::Add ( x)
inline

Single-element accumulation. Will not vectorise.

Definition at line 133 of file Util.h.

◆ AddIndexed()

template<typename T = double, unsigned int N = 1>
void ROOT::Math::KahanSum< T, N >::AddIndexed ( input,
std::size_t  index 
)
inline

Add input to the sum.

Particularly helpful when filling from a for loop. This function can be inlined and auto-vectorised if the index parameter is used to enumerate consecutive fills. Use Add() or Accumulate() when no index is available.

Parameters
[in]inputValue to accumulate.
[in]indexIndex of the value. Determines internal accumulator that this value is added to. Make sure that consecutive fills have consecutive indices to make a loop auto-vectorisable. The actual value of the index does not matter, as long as it is consecutive.

Definition at line 199 of file Util.h.

◆ Carry()

template<typename T = double, unsigned int N = 1>
T ROOT::Math::KahanSum< T, N >::Carry ( ) const
inline
Returns
The sum used for compensation.

Definition at line 223 of file Util.h.

◆ operator T()

template<typename T = double, unsigned int N = 1>
ROOT::Math::KahanSum< T, N >::operator T ( ) const
inline

Auto-convert to type T.

Definition at line 218 of file Util.h.

◆ operator+=()

template<typename T = double, unsigned int N = 1>
KahanSum< T, N > & ROOT::Math::KahanSum< T, N >::operator+= ( arg)
inline

Add arg into accumulator. Does not vectorise.

Definition at line 228 of file Util.h.

◆ Result()

template<typename T = double, unsigned int N = 1>
T ROOT::Math::KahanSum< T, N >::Result ( ) const
inline
Returns
Compensated sum.

Definition at line 213 of file Util.h.

◆ Sum()

template<typename T = double, unsigned int N = 1>
T ROOT::Math::KahanSum< T, N >::Sum ( ) const
inline
Returns
Compensated sum.

Definition at line 208 of file Util.h.

Member Data Documentation

◆ fCarry

template<typename T = double, unsigned int N = 1>
T ROOT::Math::KahanSum< T, N >::fCarry[N]
private

Definition at line 235 of file Util.h.

◆ fSum

template<typename T = double, unsigned int N = 1>
T ROOT::Math::KahanSum< T, N >::fSum[N]
private

Definition at line 234 of file Util.h.

  • math/mathcore/inc/Math/Util.h