#include <algorithm>
#include <array>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <cstring>
#include <type_traits>
#include <vector>
#include "clustering/always_assert.h"
#include "clustering/math/detail/avx2_helpers.h"
#include "clustering/math/detail/gemm_outer.h"
#include "clustering/math/detail/matrix_desc.h"
#include "clustering/math/pairwise.h"
#include "clustering/math/rng.h"
#include "clustering/math/thread.h"
#include "clustering/ndarray.h"
#include <immintrin.h>
#include "clustering/math/detail/kmpp_score_avx2.h"

Include dependency graph for greedy_kmpp_seeder.h:

This graph shows which files directly or indirectly include this file:

Classes
class	clustering::kmeans::GreedyKmppSeeder< T >
	Greedy k-means++ seeder. More...

Namespaces
namespace	clustering
namespace	clustering::kmeans
namespace	clustering::kmeans::detail

Functions
std::size_t	clustering::kmeans::detail::greedyKmppLocalTrials (std::size_t k) noexcept
	Compute the local-trials count used by greedy k-means++.
constexpr std::size_t	clustering::kmeans::detail::greedyKmppTransposedWidth (std::size_t L) noexcept
	Round `L` up to the nearest multiple of 8 used by the transposed scoring layout.
template<std::size_t B>
void	clustering::kmeans::detail::sqEuclideanRowToBatchAvx2Fixed (const float x, const float candData, std::size_t d, float *out) noexcept
	Compile-time batched scoring kernel: stream `x` once across `B` parallel AVX2 accumulators to compute `B` squared distances against `candData` rows.
template<std::size_t B>
void	clustering::kmeans::detail::sqEuclideanRowToBatchAvx2Fixed (const double x, const double candData, std::size_t d, double *out) noexcept
template<class T>
void	clustering::kmeans::detail::sqEuclideanRowToBatchAvx2 (const T x, const T candData, std::size_t L, std::size_t d, T *out) noexcept
	Compute `L` squared Euclidean distances against an (L, d) row-batched candidate layout in a single streaming pass over the `x` row.
void	clustering::kmeans::detail::sqEuclideanRowAgainst8Transposed (const float x, const float candData, std::size_t d, float *out) noexcept
	Compute `L` squared distances against an (d, 8) transposed candidate layout with one streaming pass over the `x` row.
void	clustering::kmeans::detail::sqEuclideanRowAgainst16Transposed (const float x, const float candData, std::size_t d, float *out) noexcept
	Compute two 8-way squared distance slabs against an (d, 16) transposed candidate layout in one streaming pass over the `x` row.
void	clustering::kmeans::detail::sqEuclideanRowAgainst8TransposedStrided (const float x, const float candData, std::size_t d, std::size_t rowStride, float *out) noexcept
	Compute one 8-way squared distance slab against an (d, W) transposed candidate layout with an explicit row stride `W`.
template<class T>
void	clustering::kmeans::detail::sqEuclideanRowToBatch (const T x, const T candData, std::size_t L, std::size_t d, T *out) noexcept
	Squared Euclidean distance from one `x` row to a batch of `L` candidate rows.

Classes

Namespaces

Functions