Shamrock 2025.10.0
Astrophysical Code
Loading...
Searching...
No Matches
shamalgs::numeric Namespace Reference

namespace containing the numeric algorithms of shamalgs More...

Classes

struct  histogram_result
struct  BinnedCompute
 Structure holding the result of binning values for further computation. More...

Functions

template<class T>
sycl::buffer< T > scan_exclusive (sycl::queue &q, sycl::buffer< T > &buf1, u32 len)
 Computes the exclusive sum of elements in a SYCL buffer.
template<class T>
sham::DeviceBuffer< T > scan_exclusive (sham::DeviceScheduler_ptr sched, sham::DeviceBuffer< T > &buf1, u32 len)
 Compute the exclusive sum of a buffer on the device.
std::tuple< std::optional< sycl::buffer< u32 > >, u32 > stream_compact (sycl::queue &q, sycl::buffer< u32 > &buf_flags, u32 len)
 Stream compaction algorithm.
sham::DeviceBuffer< u32 > stream_compact (const sham::DeviceScheduler_ptr &sched, sham::DeviceBuffer< u32 > &buf_flags, u32 len)
 Stream compaction algorithm.
template<class Tret, class T>
sham::DeviceBuffer< Tret > device_histogram (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the histogram of values between bin_edges.
template<class T>
sham::DeviceBuffer< u64 > device_histogram_u64 (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
template<class T>
sham::DeviceBuffer< u32 > device_histogram_u32 (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
template<class T>
histogram_result< T > device_histogram_full (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the histogram and bin properties (center, width) for a set of values and bin edges.
template<class T>
BinnedCompute< T > binned_init_compute (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
 Prepare binned data for per-bin computation.
template<class Tret, class T, class Fct>
sham::DeviceBuffer< Tret > binned_compute (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len, Fct &&fct)
 Perform a custom reduction or computation over values in each bin.
template<class T>
sham::DeviceBuffer< T > binned_sum (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
 Compute the sum of values in each bin.
template<class T>
sham::DeviceBuffer< T > binned_average (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
 Compute the average of values in each bin.
template<class Tret, class T>
sham::DeviceBuffer< Tret > device_histogram_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the histogram of values between bin_edges across all MPI ranks.
template<class T>
sham::DeviceBuffer< u64 > device_histogram_u64_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the u64 histogram of values between bin_edges across all MPI ranks.
template<class T>
sham::DeviceBuffer< u32 > device_histogram_u32_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the u32 histogram of values between bin_edges across all MPI ranks.
template<class T>
sham::DeviceBuffer< T > binned_sum_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
 Compute the sum of values in each bin across all MPI ranks.
template<class T>
sham::DeviceBuffer< T > binned_average_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len, const sham::DeviceBuffer< u32 > &bin_counts_global)
 Compute the average of values in each bin across all MPI ranks (with pre-computed global counts).
template<class T>
sham::DeviceBuffer< T > binned_average_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
 Compute the average of values in each bin across all MPI ranks.
template<class T>
histogram_result< T > device_histogram_full_mpi (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
 Compute the histogram and bin properties (center, width) for a set of values and bin edges.
template sycl::buffer< u32 > scan_exclusive (sycl::queue &q, sycl::buffer< u32 > &buf1, u32 len)
template sham::DeviceBuffer< u32 > scan_exclusive (sham::DeviceScheduler_ptr sched, sham::DeviceBuffer< u32 > &buf1, u32 len)
template sham::DeviceBuffer< u64 > device_histogram< u64, f64 > (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f64 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f64 > &values, u32 len)
template sham::DeviceBuffer< u64 > device_histogram< u64, f32 > (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f32 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f32 > &values, u32 len)
template sham::DeviceBuffer< u32 > device_histogram< u32, f64 > (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f64 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f64 > &values, u32 len)
template sham::DeviceBuffer< u32 > device_histogram< u32, f32 > (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f32 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f32 > &values, u32 len)
template BinnedCompute< f64 > binned_init_compute (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f64 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f64 > &values, const sham::DeviceBuffer< f64 > &keys, u32 len)
template BinnedCompute< f32 > binned_init_compute (const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< f32 > &bin_edges, u64 nbins, const sham::DeviceBuffer< f32 > &values, const sham::DeviceBuffer< f32 > &keys, u32 len)

Detailed Description

namespace containing the numeric algorithms of shamalgs

Function Documentation

◆ binned_average()

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::binned_average ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len )

Compute the average of values in each bin.

This function calculates the average of all values in each bin, using the keys to assign values to bins. It returns a buffer containing the average for each bin.

Template Parameters
TThe data type of the values and keys.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to be averaged (e.g., f(r)).
keysThe keys used for binning (e.g., r).
lenThe number of elements in values/keys.
Returns
sham::DeviceBuffer<T> Buffer of averages, one per bin.

Example:

auto dev_sched = shamsys::instance::get_compute_scheduler_ptr();
sham::DeviceBuffer<double> bin_edges = ...;
u64 nbins = bin_edges.get_size() - 1;
auto averages = shamalgs::numeric::binned_average(dev_sched, bin_edges, nbins, values,
keys, values.get_size());
std::uint64_t u64
64 bit unsigned integer
A buffer allocated in USM (Unified Shared Memory).
size_t get_size() const
Gets the number of elements in the buffer.
sham::DeviceBuffer< T > binned_average(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
Compute the average of values in each bin.
Definition numeric.hpp:402

Definition at line 402 of file numeric.hpp.

Here is the call graph for this function:

◆ binned_average_mpi() [1/2]

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::binned_average_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len )

Compute the average of values in each bin across all MPI ranks.

This function computes the global binned average across all MPI ranks. It first computes the global histogram counts using the keys, then computes the global binned sum, and finally divides the sums by the counts to obtain the averages. This is a convenience function that automatically computes the global bin counts.

Template Parameters
TThe data type of the values, keys, and bin edges.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins.
valuesThe local values for this rank to be averaged (e.g., f(r)).
keysThe local keys for this rank used for binning (e.g., r).
lenThe number of elements in local values/keys arrays.
Returns
sham::DeviceBuffer<T> Buffer of global averages, one per bin.

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
sham::DeviceBuffer<double> local_keys = ...; // Different on each rank
dev_sched, bin_edges, nbins, local_values, local_keys, local_values.get_size());
sham::DeviceBuffer< T > binned_average_mpi(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len, const sham::DeviceBuffer< u32 > &bin_counts_global)
Compute the average of values in each bin across all MPI ranks (with pre-computed global counts).
Definition numeric.hpp:650

If rank 0 has keys={0.5,1.5}, values={10,30} and rank 1 has keys={0.3,1.7}, values={20,10} with bin_edges={0.0,1.0,2.0}, the result={15,20} ((10+20)/2 in bin 0, (30+10)/2 in bin 1)

Definition at line 714 of file numeric.hpp.

Here is the call graph for this function:

◆ binned_average_mpi() [2/2]

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::binned_average_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len,
const sham::DeviceBuffer< u32 > & bin_counts_global )

Compute the average of values in each bin across all MPI ranks (with pre-computed global counts).

This function computes the global binned average using pre-computed global bin counts. It first computes the global binned sum across all MPI ranks and then divides by the provided global counts to obtain the average. This variant is useful when the global bin counts are already known or computed separately.

Template Parameters
TThe data type of the values, keys, and bin edges.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins.
valuesThe local values for this rank to be averaged (e.g., f(r)).
keysThe local keys for this rank used for binning (e.g., r).
lenThe number of elements in local values/keys arrays.
bin_counts_globalThe global counts for each bin across all ranks.
Returns
sham::DeviceBuffer<T> Buffer of global averages, one per bin.

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
sham::DeviceBuffer<double> local_keys = ...; // Different on each rank
sham::DeviceBuffer<u32> global_counts = ...; // Pre-computed global counts
dev_sched, bin_edges, nbins, local_values, local_keys,
local_values.get_size(), global_counts);

Definition at line 650 of file numeric.hpp.

Here is the call graph for this function:

◆ binned_compute()

template<class Tret, class T, class Fct>
sham::DeviceBuffer< Tret > shamalgs::numeric::binned_compute ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len,
Fct && fct )

Perform a custom reduction or computation over values in each bin.

This function applies a user-provided function to all values in each bin, allowing for flexible per-bin reductions (e.g., sum, mean, min, max, etc.).

Template Parameters
TThe data type of the values and keys.
TretThe return type of the per-bin computation.
FctThe type of the function to apply per bin. The function should have the signature: Tret f(for_each_values, u32 bin_count) where for_each_values is a callable that applies a function to each value in the bin.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to be binned (e.g., f(r)).
keysThe keys used for binning (e.g., r).
lenThe number of elements in values/keys.
fctThe function to apply to each bin's values.
Returns
sham::DeviceBuffer<Tret> Buffer of computed values, one per bin.

Example (per-bin sum):

auto dev_sched = shamsys::instance::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
dev_sched, bin_edges, nbins, values, keys, values.get_size(),
[](auto for_each_values, u32 bin_count) {
double sum = 0;
for_each_values([&](double v) { sum += v; });
return sum;
});
std::uint32_t u32
32 bit unsigned integer
sham::DeviceBuffer< Tret > binned_compute(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len, Fct &&fct)
Perform a custom reduction or computation over values in each bin.
Definition numeric.hpp:287

Definition at line 287 of file numeric.hpp.

Here is the call graph for this function:

◆ binned_init_compute()

template<class T>
BinnedCompute< T > shamalgs::numeric::binned_init_compute ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len )

Prepare binned data for per-bin computation.

Filters and sorts the input values and keys into bins defined by bin_edges, returning the valid values and the offsets for each bin. This is useful for custom per-bin reductions or statistics.

Template Parameters
TThe data type of the values and keys.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to be binned (e.g., f(r)).
keysThe keys used for binning (e.g., r).
lenThe number of elements in values/keys.
Returns
BinnedCompute<T> Structure containing valid values and bin offsets.

Example:

auto dev_sched = shamsys::instance::get_compute_scheduler_ptr();
sham::DeviceBuffer<double> bin_edges = ...;
sham::DeviceBuffer<double> values = ...; // f(r)
sham::DeviceBuffer<double> keys = ...; // r
u64 nbins = bin_edges.get_size() - 1;
auto binned = shamalgs::numeric::binned_init_compute(dev_sched, bin_edges, nbins, values,
keys, values.get_size());
// binned.valid_values, binned.offsets_bins
BinnedCompute< T > binned_init_compute(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
Prepare binned data for per-bin computation.
Definition numeric.cpp:164

Definition at line 164 of file numeric.cpp.

Here is the call graph for this function:

◆ binned_sum()

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::binned_sum ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len )

Compute the sum of values in each bin.

This function computes the sum of all values in each bin, using the keys to assign values to bins.

Template Parameters
TThe data type of the values and keys.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to be summed (e.g., f(r)).
keysThe keys used for binning (e.g., r).
lenThe number of elements in values/keys.
Returns
sham::DeviceBuffer<T> Buffer of sums, one per bin.

Example:

auto dev_sched = shamsys::instance::get_compute_scheduler_ptr();
sham::DeviceBuffer<double> bin_edges = ...;
u64 nbins = bin_edges.get_size() - 1;
auto sums = shamalgs::numeric::binned_sum(dev_sched, bin_edges, nbins, values, keys,
values.get_size());
sham::DeviceBuffer< T > binned_sum(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
Compute the sum of values in each bin.
Definition numeric.hpp:357

Definition at line 357 of file numeric.hpp.

Here is the call graph for this function:

◆ binned_sum_mpi()

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::binned_sum_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
const sham::DeviceBuffer< T > & keys,
u32 len )

Compute the sum of values in each bin across all MPI ranks.

This function computes the local binned sum on each MPI rank and then performs an MPI reduction to sum the results across all ranks, producing the global binned sum. Each rank processes its own subset of keys and values, and the final result contains the combined sums from all processes.

Template Parameters
TThe data type of the values, keys, and bin edges.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins.
valuesThe local values for this rank to be summed (e.g., f(r)).
keysThe local keys for this rank used for binning (e.g., r).
lenThe number of elements in local values/keys arrays.
Returns
sham::DeviceBuffer<T> Buffer of global sums, one per bin.

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
sham::DeviceBuffer<double> local_keys = ...; // Different on each rank
dev_sched, bin_edges, nbins, local_values, local_keys, local_values.get_size());
sham::DeviceBuffer< T > binned_sum_mpi(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, const sham::DeviceBuffer< T > &keys, u32 len)
Compute the sum of values in each bin across all MPI ranks.
Definition numeric.hpp:597

If rank 0 has keys={0.5,1.5}, values={10,20} and rank 1 has keys={0.3,1.7}, values={5,15} with bin_edges={0.0,1.0,2.0}, the result={15,35} (10+5 in bin 0, 20+15 in bin 1)

Definition at line 597 of file numeric.hpp.

Here is the call graph for this function:

◆ device_histogram()

template<class Tret, class T>
sham::DeviceBuffer< Tret > shamalgs::numeric::device_histogram ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )

Compute the histogram of values between bin_edges.

This function computes the histogram of the input values, counting how many values fall into each bin defined by the bin_edges array. Only values within [bin_edges[0], bin_edges[nbins]) are counted; values outside this range are ignored.

Template Parameters
TThe data type of the values and bin edges (e.g., float, double).
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order.
nbinsThe number of bins (must be > 0, nbins = bin_edges.size() - 1).
valuesThe values to compute the histogram on.
lenThe length of the values array.
Returns
sham::DeviceBuffer<u64> The counts in each bin (length == nbins).

Example:

auto dev_sched = shamsys::instance::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
dev_sched, d_bin_edges, nbins, values, values.get_size());
sham::DeviceBuffer< Tret > device_histogram(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
Compute the histogram of values between bin_edges.
Definition numeric.cpp:76

bin_edges = {0.0, 1.0, 2.0, 3.0, 4.0} (4 bins: [0,1), [1,2), [2,3), [3,4)) values = {0.5, 1.5, 2.5, 3.5, 2.1, 1.9, 0.1, 3.9} result = {2, 2, 2, 2}

Definition at line 76 of file numeric.cpp.

Here is the call graph for this function:

◆ device_histogram_full()

template<class T>
histogram_result< T > shamalgs::numeric::device_histogram_full ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )

Compute the histogram and bin properties (center, width) for a set of values and bin edges.

This function returns the histogram counts, the center of each bin, and the width of each bin.

Template Parameters
TThe data type of the values and bin edges.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to compute the histogram on.
lenThe length of the values array.
Returns
histogram_result<T> Structure containing counts, bin centers, and bin widths.

Definition at line 165 of file numeric.hpp.

Here is the call graph for this function:

◆ device_histogram_full_mpi()

template<class T>
histogram_result< T > shamalgs::numeric::device_histogram_full_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )

Compute the histogram and bin properties (center, width) for a set of values and bin edges.

This function returns the histogram counts, the center of each bin, and the width of each bin.

Template Parameters
TThe data type of the values and bin edges.
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1).
nbinsThe number of bins.
valuesThe values to compute the histogram on.
lenThe length of the values array.
Returns
histogram_result<T> Structure containing counts, bin centers, and bin widths.

Definition at line 744 of file numeric.hpp.

Here is the call graph for this function:

◆ device_histogram_mpi()

template<class Tret, class T>
sham::DeviceBuffer< Tret > shamalgs::numeric::device_histogram_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )

Compute the histogram of values between bin_edges across all MPI ranks.

This function computes the local histogram on each MPI rank and then performs an MPI reduction to sum the counts across all ranks, producing the global histogram. Each rank processes its own subset of data, and the final result contains the combined counts from all processes.

Template Parameters
TretThe data type for the return counts (e.g., u32, u64).
TThe data type of the values and bin edges (e.g., float, double).
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins (must be > 0, nbins = bin_edges.size() - 1).
valuesThe local values for this rank to compute the histogram on.
lenThe length of the local values array.
Returns
sham::DeviceBuffer<Tret> The global counts in each bin (length == nbins).

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
dev_sched, bin_edges, nbins, local_values, local_values.get_size());
sham::DeviceBuffer< Tret > device_histogram_mpi(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
Compute the histogram of values between bin_edges across all MPI ranks.
Definition numeric.hpp:466

If rank 0 has values = {0.5, 1.5} and rank 1 has values = {2.5, 3.5} with bin_edges = {0.0, 1.0, 2.0, 3.0, 4.0}, the result = {1, 1, 1, 1}

Definition at line 466 of file numeric.hpp.

Here is the call graph for this function:

◆ device_histogram_u32()

template<class T>
sham::DeviceBuffer< u32 > shamalgs::numeric::device_histogram_u32 ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )
inline

Definition at line 140 of file numeric.hpp.

◆ device_histogram_u32_mpi()

template<class T>
sham::DeviceBuffer< u32 > shamalgs::numeric::device_histogram_u32_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )
inline

Compute the u32 histogram of values between bin_edges across all MPI ranks.

Convenience wrapper for device_histogram_mpi with u32 return type. This function computes the local histogram on each MPI rank and then performs an MPI reduction to sum the counts across all ranks, producing the global histogram with 32-bit unsigned integer counts.

Template Parameters
TThe data type of the values and bin edges (e.g., float, double).
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins (must be > 0, nbins = bin_edges.size() - 1).
valuesThe local values for this rank to compute the histogram on.
lenThe length of the local values array.
Returns
sham::DeviceBuffer<u32> The global counts in each bin (length == nbins).

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
dev_sched, bin_edges, nbins, local_values, local_values.get_size());
sham::DeviceBuffer< u32 > device_histogram_u32_mpi(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
Compute the u32 histogram of values between bin_edges across all MPI ranks.
Definition numeric.hpp:552

Definition at line 552 of file numeric.hpp.

Here is the call graph for this function:

◆ device_histogram_u64()

template<class T>
sham::DeviceBuffer< u64 > shamalgs::numeric::device_histogram_u64 ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )
inline

Definition at line 130 of file numeric.hpp.

◆ device_histogram_u64_mpi()

template<class T>
sham::DeviceBuffer< u64 > shamalgs::numeric::device_histogram_u64_mpi ( const sham::DeviceScheduler_ptr & sched,
const sham::DeviceBuffer< T > & bin_edges,
u64 nbins,
const sham::DeviceBuffer< T > & values,
u32 len )
inline

Compute the u64 histogram of values between bin_edges across all MPI ranks.

Convenience wrapper for device_histogram_mpi with u64 return type. This function computes the local histogram on each MPI rank and then performs an MPI reduction to sum the counts across all ranks, producing the global histogram with 64-bit unsigned integer counts.

Template Parameters
TThe data type of the values and bin edges (e.g., float, double).
Parameters
schedThe device scheduler to run on.
bin_edgesThe edges of the bins (length == nbins + 1). Must be sorted in ascending order and identical across all ranks.
nbinsThe number of bins (must be > 0, nbins = bin_edges.size() - 1).
valuesThe local values for this rank to compute the histogram on.
lenThe length of the local values array.
Returns
sham::DeviceBuffer<u64> The global counts in each bin (length == nbins).

Example:

auto dev_sched = shamsys::get_compute_scheduler_ptr();
u64 nbins = bin_edges.get_size() - 1;
sham::DeviceBuffer<double> local_values = ...; // Different on each rank
dev_sched, bin_edges, nbins, local_values, local_values.get_size());
sham::DeviceBuffer< u64 > device_histogram_u64_mpi(const sham::DeviceScheduler_ptr &sched, const sham::DeviceBuffer< T > &bin_edges, u64 nbins, const sham::DeviceBuffer< T > &values, u32 len)
Compute the u64 histogram of values between bin_edges across all MPI ranks.
Definition numeric.hpp:512

Definition at line 512 of file numeric.hpp.

Here is the call graph for this function:

◆ scan_exclusive() [1/2]

template<class T>
sham::DeviceBuffer< T > shamalgs::numeric::scan_exclusive ( sham::DeviceScheduler_ptr sched,
sham::DeviceBuffer< T > & buf1,
u32 len )

Compute the exclusive sum of a buffer on the device.

Parameters
schedThe scheduler to use for the computation
buf1The buffer to sum
lenThe length of the sum
Returns
A new buffer which is the output of the sum

Definition at line 48 of file numeric.cpp.

◆ scan_exclusive() [2/2]

template<class T>
sycl::buffer< T > shamalgs::numeric::scan_exclusive ( sycl::queue & q,
sycl::buffer< T > & buf1,
u32 len )

Computes the exclusive sum of elements in a SYCL buffer.

Template Parameters
TThe data type of elements in the buffer.
Parameters
qThe SYCL queue to use for computation.
buf1The input buffer whose exclusive sum is to be computed.
lenThe number of elements in the buffer.
Returns
A new SYCL buffer containing the exclusive sum of the input buffer.

Definition at line 35 of file numeric.cpp.

◆ stream_compact() [1/2]

sham::DeviceBuffer< u32 > shamalgs::numeric::stream_compact ( const sham::DeviceScheduler_ptr & sched,
sham::DeviceBuffer< u32 > & buf_flags,
u32 len )

Stream compaction algorithm.

Parameters
schedthe device scheduler to run on
buf_flagsbuffer of only 0 and ones
lenthe length of the buffer considered
Returns
sham::DeviceBuffer<u32> table of the index to extract

Definition at line 70 of file numeric.cpp.

◆ stream_compact() [2/2]

std::tuple< std::optional< sycl::buffer< u32 > >, u32 > shamalgs::numeric::stream_compact ( sycl::queue & q,
sycl::buffer< u32 > & buf_flags,
u32 len )

Stream compaction algorithm.

Parameters
qthe queue to run on
buf_flagsbuffer of only 0 and ones
lenthe length of the buffer considered
Returns
std::tuple<sycl::buffer<u32>, u32> table of the index to extract and its size

Definition at line 65 of file numeric.cpp.