Thrust

◆ inclusive_scan() [1/4]

template<typename DerivedPolicy , typename InputIterator , typename OutputIterator >
__host__ __device__ OutputIterator thrust::inclusive_scan ( const thrust::detail::execution_policy_base< DerivedPolicy > &  exec,
InputIterator  first,
InputIterator  last,
OutputIterator  result 
)

inclusive_scan computes an inclusive prefix sum operation. The term 'inclusive' means that each result includes the corresponding input operand in the partial sum. More precisely, *first is assigned to *result and the sum of *first and *(first + 1) is assigned to *(result + 1), and so on. This version of inclusive_scan assumes plus as the associative operator. When the input and output sequences are the same, the scan is performed in-place.

inclusive_scan is similar to std::partial_sum in the STL. The primary difference between the two functions is that std::partial_sum guarantees a serial summation order, while inclusive_scan requires associativity of the binary operation to parallelize the prefix sum.

Results are not deterministic for pseudo-associative operators (e.g., addition of floating-point types). Results for pseudo-associative operators may vary from run to run.

The algorithm's execution is parallelized as determined by exec.

Parameters
execThe execution policy to use for parallelization.
firstThe beginning of the input sequence.
lastThe end of the input sequence.
resultThe beginning of the output sequence.
Returns
The end of the output sequence.
Template Parameters
DerivedPolicyThe name of the derived execution policy.
InputIteratoris a model of Input Iterator and InputIterator's value_type is convertible to OutputIterator's value_type.
OutputIteratoris a model of Output Iterator, and if x and y are objects of OutputIterator's value_type, then x + y is defined. If T is OutputIterator's value_type, then T(0) is defined.
Precondition
first may equal result but the range [first, last) and the range [result, result + (last - first)) shall not overlap otherwise.

The following code snippet demonstrates how to use inclusive_scan to compute an in-place prefix sum using the thrust::host execution policy for parallelization:

#include <thrust/scan.h>
...
int data[6] = {1, 0, 2, 2, 1, 3};
thrust::inclusive_scan(thrust::host, data, data + 6, data); // in-place scan
// data is now {1, 1, 3, 5, 6, 9}
See also
https://en.cppreference.com/w/cpp/algorithm/partial_sum