Overview

template<typename T>
struct batchlas::StedcParams< T >

Parameters for stedc().

Every tuning knob at 0 (or Auto) takes the value from the checked-in tuning tables.

See also: STEDC: the divide-and-conquer merge, section "stedc: current tuning values"

Definition at line 1847 of file extensions.hh.

#include <batchlas/blas/extensions.hh>

Variables

◆ recursion_threshold

template<typename T>

int64_t batchlas::StedcParams< T >::recursion_threshold = 0

Leaf size; <= 0 uses BatchLAS tuning, otherwise this exact threshold.

Definition at line 1848 of file extensions.hh.

◆ algorithm

template<typename T>

StedcAlgorithm batchlas::StedcParams< T >::algorithm = StedcAlgorithm::Auto

Merge-tree driver.

Definition at line 1849 of file extensions.hh.

◆ secular_solver

template<typename T>

StedcSecularSolver batchlas::StedcParams< T >::secular_solver = StedcSecularSolver::Rocm

Secular root finder.

Definition at line 1850 of file extensions.hh.

◆ leaf_steqr_params

template<typename T>

SteqrParams<T> batchlas::StedcParams< T >::leaf_steqr_params = SteqrParams<T>()

Parameters of the leaf steqr() solves.

Definition at line 1851 of file extensions.hh.

◆ merge_variant

template<typename T>

StedcMergeVariant batchlas::StedcParams< T >::merge_variant = StedcMergeVariant::Auto

Merge-step dispatch.

Definition at line 1853 of file extensions.hh.

◆ merge_threads

template<typename T>

int batchlas::StedcParams< T >::merge_threads = 128

work-group size for fused kernel

Definition at line 1854 of file extensions.hh.

◆ max_sec_iter

template<typename T>

int batchlas::StedcParams< T >::max_sec_iter = 50

iteration cap for secular root solver

Definition at line 1855 of file extensions.hh.

◆ enable_rescale

template<typename T>

bool batchlas::StedcParams< T >::enable_rescale = true

keep ROCm-style v rescale (disable for perf experiments)

Definition at line 1856 of file extensions.hh.

◆ secular_threads_per_root

template<typename T>

int batchlas::StedcParams< T >::secular_threads_per_root = 0

<= 0 uses BatchLAS tuning; otherwise this exact partition width

Definition at line 1857 of file extensions.hh.

◆ secular_cta_wg_size_multiplier

template<typename T>

int batchlas::StedcParams< T >::secular_cta_wg_size_multiplier = 0

<= 0 uses BatchLAS tuning; otherwise this exact multiplier

Definition at line 1858 of file extensions.hh.