uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment > Struct Template Reference

Packed sparse storage whose disjoint dense blocks may be processed in a scheduler batch. More...

#include <block_tensor_storage.hpp>

Public Types
using	backend_selector_type = typename leaf_storage_policy::backend_selector_type
using	block_execution_policy = SchedulerBatchBlockExecution
using	leaf_storage_policy = LeafStorage
template<typename T , std::size_t KeyCoordinateCount, std::size_t DenseBlockOrder>
using	storage_t = detail::PackedBlockStorageData< T, KeyCoordinateCount, DenseBlockOrder, leaf_storage_policy, BlockAlignment >

Static Public Member Functions
static constexpr auto	backend_selector () noexcept -> backend_selector_type

Static Public Attributes
static constexpr bool	is_distributed = false
static constexpr bool	stores_all_legal_blocks = false

Detailed Description

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>
struct uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >

Packed sparse storage whose disjoint dense blocks may be processed in a scheduler batch.

The packed buffer remains fixed while numerical operations execute. Algorithms must place all writes to one logical output block in the same batch item, so concurrently written block ranges are disjoint. The batch call is synchronous. A BlockAlignment greater than one pads block starts to that byte alignment.

Template Parameters

LeafStorage	Tensor storage policy used by the packed buffer.
BlockAlignment	Required byte alignment of every block start.

Member Typedef Documentation

backend_selector_type

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

using uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::backend_selector_type = typename leaf_storage_policy::backend_selector_type

block_execution_policy

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

using uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::block_execution_policy = SchedulerBatchBlockExecution

leaf_storage_policy

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

using uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::leaf_storage_policy = LeafStorage

storage_t

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

template<typename T , std::size_t KeyCoordinateCount, std::size_t DenseBlockOrder>

using uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::storage_t = detail::PackedBlockStorageData<T, KeyCoordinateCount, DenseBlockOrder, leaf_storage_policy, BlockAlignment>

Member Function Documentation

backend_selector()

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

static constexpr auto uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::backend_selector ( ) -> backend_selector_type

inlinestaticconstexprnoexcept

Member Data Documentation

is_distributed

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

constexpr bool uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::is_distributed = false

staticconstexpr

stores_all_legal_blocks

template<class LeafStorage = HostStorage, std::size_t BlockAlignment = 1>

constexpr bool uni20::ParallelPackedSparseBlockStorage< LeafStorage, BlockAlignment >::stores_all_legal_blocks = false

staticconstexpr

---

The documentation for this struct was generated from the following file:

• block_tensor_storage.hpp