CustomSparseChunkedMatrix.hpp

Go to the documentation of this file.

#ifndef TATAMI_CHUNKED_CUSTOM_SPARSE_CHUNKED_MATRIX_HPP
#define TATAMI_CHUNKED_CUSTOM_SPARSE_CHUNKED_MATRIX_HPP
 
#include "custom_internals.hpp"
#include "SparseSlabFactory.hpp"
#include "SlabCacheStats.hpp"
#include "LruSlabCache.hpp"
#include "OracularSlabCache.hpp"
#include "OracularSubsettedSlabCache.hpp"
#include "utils.hpp"
 
#include <vector>
#include <cstddef>
 
#include "tatami/tatami.hpp"
#include "sanisizer/sanisizer.hpp"
 
namespace tatami_chunked {
 
struct CustomSparseChunkedMatrixOptions {
    std::size_t maximum_cache_size = sanisizer::cap<std::size_t>(100000000);
 
    bool require_minimum_cache = true;
 
    bool cache_subset = false;
};
 
template<typename ChunkValue_, typename Index_>
class CustomSparseChunkedMatrixWorkspace {
public:
    CustomSparseChunkedMatrixWorkspace() = default;
    CustomSparseChunkedMatrixWorkspace(const CustomSparseChunkedMatrixWorkspace&) = default;
    CustomSparseChunkedMatrixWorkspace(CustomSparseChunkedMatrixWorkspace&&) = default;
    CustomSparseChunkedMatrixWorkspace& operator=(const CustomSparseChunkedMatrixWorkspace&) = default;
    CustomSparseChunkedMatrixWorkspace& operator=(CustomSparseChunkedMatrixWorkspace&&) = default;
    virtual ~CustomSparseChunkedMatrixWorkspace() = default;
    virtual void extract(
        Index_ chunk_row_id,
        Index_ chunk_column_id,
        bool row,
        Index_ target_start,
        Index_ target_length,
        Index_ non_target_start,
        Index_ non_target_length,
        const std::vector<ChunkValue_*>& output_values,
        const std::vector<Index_*>& output_indices,
        Index_* output_number,
        Index_ shift
    ) = 0;
 
    virtual void extract(
        Index_ chunk_row_id,
        Index_ chunk_column_id,
        bool row,
        Index_ target_start,
        Index_ target_length,
        const std::vector<Index_>& non_target_indices,
        const std::vector<ChunkValue_*>& output_values,
        const std::vector<Index_*>& output_indices,
        Index_* output_number,
        Index_ shift
    ) = 0;
 
    virtual void extract(
        Index_ chunk_row_id,
        Index_ chunk_column_id,
        bool row,
        const std::vector<Index_>& target_indices,
        Index_ non_target_start,
        Index_ non_target_length,
        const std::vector<ChunkValue_*>& output_values,
        const std::vector<Index_*>& output_indices,
        Index_* output_number,
        Index_ shift
    ) = 0;
 
    virtual void extract(
        Index_ chunk_row_id,
        Index_ chunk_column_id,
        bool row,
        const std::vector<Index_>& target_indices,
        const std::vector<Index_>& non_target_indices,
        const std::vector<ChunkValue_*>& output_values,
        const std::vector<Index_*>& output_indices,
        Index_* output_number,
        Index_ shift
    ) = 0;
};
 
template<typename ChunkValue_, typename Index_>
class CustomSparseChunkedMatrixManager {
public:
    CustomSparseChunkedMatrixManager() = default;
    CustomSparseChunkedMatrixManager(const CustomSparseChunkedMatrixManager&) = default;
    CustomSparseChunkedMatrixManager(CustomSparseChunkedMatrixManager&&) = default;
    CustomSparseChunkedMatrixManager& operator=(const CustomSparseChunkedMatrixManager&) = default;
    CustomSparseChunkedMatrixManager& operator=(CustomSparseChunkedMatrixManager&&) = default;
    virtual ~CustomSparseChunkedMatrixManager() = default;
    virtual std::unique_ptr<CustomSparseChunkedMatrixWorkspace<ChunkValue_, Index_> > new_workspace() const = 0;
 
    std::unique_ptr<CustomSparseChunkedMatrixWorkspace<ChunkValue_, Index_> > new_workspace_exact() const {
        return new_workspace();
    }
 
    virtual bool prefer_rows() const = 0;
 
    virtual const ChunkDimensionStats<Index_>& row_stats() const = 0;
 
    virtual const ChunkDimensionStats<Index_>& column_stats() const = 0;
};
 
namespace CustomChunkedMatrix_internal {
 
/*********************
 **** Base classes ***
 *********************/
 
template<bool oracle_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class SoloSparseCore {
    WorkspacePtr_ my_chunk_workspace;
    const ChunkCoordinator<true, ChunkValue_, Index_>& my_coordinator;
 
    tatami::MaybeOracle<oracle_, Index_> my_oracle;
    typename std::conditional<oracle_, tatami::PredictionIndex, bool>::type my_counter = 0;
 
    SparseSlabFactory<ChunkValue_, Index_, Index_> my_factory;
    typedef typename I<decltype(my_factory)>::Slab Slab;
 
    // These two instances are not fully allocated Slabs; rather, tmp_solo just
    // holds the content for a single chunk, while final_solo holds the content
    // across chunks but only for the requested dimension element. Both cases
    // are likely to be much smaller than a full Slab, so we're already more
    // memory-efficient than 'require_minimum_cache = true'.
    SparseSingleWorkspace<ChunkValue_, Index_> my_tmp_solo;
    Slab my_final_solo;
 
public:
    SoloSparseCore(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        [[maybe_unused]] const SlabCacheStats<Index_>& slab_stats, // for consistency with the other base classes.
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle,
        Index_ non_target_length,
        bool needs_value,
        bool needs_index
    ) :
        my_chunk_workspace(std::move(chunk_workspace)),
        my_coordinator(coordinator),
        my_oracle(std::move(oracle)),
        my_factory(1, non_target_length, 1, needs_value, needs_index),
        my_tmp_solo(
            my_coordinator.get_target_chunkdim(row),
            my_coordinator.get_non_target_chunkdim(row), 
            needs_value, 
            needs_index
        ),
        my_final_solo(my_factory.create())
    {}
 
    template<typename ... Args_>
    std::pair<const Slab*, Index_> fetch_raw(Index_ i, bool row, Args_&& ... args) {
        if constexpr(oracle_) {
            i = my_oracle->get(my_counter++);
        }
        return my_coordinator.fetch_single(row, i, std::forward<Args_>(args)..., *my_chunk_workspace, my_tmp_solo, my_final_solo);
    }
};
 
template<typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class MyopicSparseCore {
    WorkspacePtr_ my_chunk_workspace;
    const ChunkCoordinator<true, ChunkValue_, Index_>& my_coordinator;
 
    SparseSlabFactory<ChunkValue_, Index_, Index_> my_factory;
    typedef typename I<decltype(my_factory)>::Slab Slab;
 
    LruSlabCache<Index_, Slab> my_cache;
 
public:
    MyopicSparseCore(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator,
        const SlabCacheStats<Index_>& slab_stats, 
        bool row,
        [[maybe_unused]] tatami::MaybeOracle<false, Index_> oracle, // for consistency with the other base classes
        Index_ non_target_length,
        bool needs_value,
        bool needs_index
    ) : 
        my_chunk_workspace(std::move(chunk_workspace)),
        my_coordinator(coordinator),
        my_factory(coordinator.get_target_chunkdim(row), non_target_length, slab_stats, needs_value, needs_index),
        my_cache(slab_stats.max_slabs_in_cache) 
    {}
 
    template<typename ... Args_>
    std::pair<const Slab*, Index_> fetch_raw(Index_ i, bool row, Args_&& ... args) {
        return my_coordinator.fetch_myopic(row, i, std::forward<Args_>(args)..., *my_chunk_workspace, my_cache, my_factory);
    }
};
 
template<bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class OracularSparseCore {
protected:
    WorkspacePtr_ my_chunk_workspace;
    const ChunkCoordinator<true, ChunkValue_, Index_>& my_coordinator;
 
    SparseSlabFactory<ChunkValue_, Index_, Index_> my_factory;
    typedef typename I<decltype(my_factory)>::Slab Slab;
 
    typename std::conditional<use_subset_, OracularSubsettedSlabCache<Index_, Index_, Slab>, OracularSlabCache<Index_, Index_, Slab> >::type my_cache;
 
public:
    OracularSparseCore(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator,
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<true, Index_> oracle,
        Index_ non_target_length,
        bool needs_value,
        bool needs_index
    ) : 
        my_chunk_workspace(std::move(chunk_workspace)),
        my_coordinator(coordinator), 
        my_factory(coordinator.get_target_chunkdim(row), non_target_length, slab_stats, needs_value, needs_index),
        my_cache(std::move(oracle), slab_stats.max_slabs_in_cache) 
    {}
 
    template<typename ... Args_>
    std::pair<const Slab*, Index_> fetch_raw([[maybe_unused]] Index_ i, bool row, Args_&& ... args) {
        if constexpr(use_subset_) {
            return my_coordinator.fetch_oracular_subsetted(row, std::forward<Args_>(args)..., *my_chunk_workspace, my_cache, my_factory);
        } else {
            return my_coordinator.fetch_oracular(row, std::forward<Args_>(args)..., *my_chunk_workspace, my_cache, my_factory);
        }
    }
};
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
using SparseCore = typename std::conditional<solo_, 
      SoloSparseCore<oracle_, Value_, Index_, ChunkValue_, WorkspacePtr_>,
      typename std::conditional<oracle_,
          OracularSparseCore<use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_>,
          MyopicSparseCore<Value_, Index_, ChunkValue_, WorkspacePtr_>
      >::type
>::type;
 
/***********************
 **** Sparse classes ***
 ***********************/
 
template<class Slab_, typename Index_, typename Value_>
tatami::SparseRange<Value_, Index_> process_sparse_slab(const std::pair<const Slab_*, Index_>& fetched, Value_* value_buffer, Index_* index_buffer, bool needs_value, bool needs_index) {
    auto num = fetched.first->number[fetched.second];
 
    if (needs_value) {
        auto vptr = fetched.first->values[fetched.second];
        std::copy_n(vptr, num, value_buffer);
    } else {
        value_buffer = NULL;
    }
 
    if (needs_index) {
        auto iptr = fetched.first->indices[fetched.second];
        std::copy_n(iptr, num, index_buffer);
    } else {
        index_buffer = NULL;
    }
 
    return tatami::SparseRange<Value_, Index_>(num, value_buffer, index_buffer);
}
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class SparseFull : public tatami::SparseExtractor<oracle_, Value_, Index_> {
public:
    SparseFull(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle, 
        const tatami::Options& opt
    ) :
        my_row(row),
        my_non_target_dim(coordinator.get_non_target_dim(row)),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index),
        my_core(
            std::move(chunk_workspace),
            coordinator, 
            slab_stats,
            row,
            std::move(oracle), 
            my_non_target_dim,
            opt.sparse_extract_value,
            opt.sparse_extract_index
        )
    {}
 
    tatami::SparseRange<Value_, Index_> fetch(Index_ i, Value_* value_buffer, Index_* index_buffer) {
        auto fetched = my_core.fetch_raw(i, my_row, 0, my_non_target_dim);
        return process_sparse_slab(fetched, value_buffer, index_buffer, my_needs_value, my_needs_index); 
    }
 
private:
    bool my_row;
    Index_ my_non_target_dim;
    bool my_needs_value, my_needs_index;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class SparseBlock : public tatami::SparseExtractor<oracle_, Value_, Index_> {
public:
    SparseBlock(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle, 
        Index_ block_start, 
        Index_ block_length, 
        const tatami::Options& opt
    ) :
        my_row(row),
        my_block_start(block_start),
        my_block_length(block_length),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index),
        my_core(
            std::move(chunk_workspace),
            coordinator,
            slab_stats,
            row,
            std::move(oracle),
            block_length,
            my_needs_value,
            my_needs_index
        )
    {}
 
    tatami::SparseRange<Value_, Index_> fetch(Index_ i, Value_* value_buffer, Index_* index_buffer) {
        auto fetched = my_core.fetch_raw(i, my_row, my_block_start, my_block_length);
        return process_sparse_slab(fetched, value_buffer, index_buffer, my_needs_value, my_needs_index);
    }
 
private:
    bool my_row;
    Index_ my_block_start, my_block_length;
    bool my_needs_value, my_needs_index;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class SparseIndex : public tatami::SparseExtractor<oracle_, Value_, Index_> {
public:
    SparseIndex(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle, 
        tatami::VectorPtr<Index_> indices_ptr, 
        const tatami::Options& opt
    ) :
        my_row(row),
        my_indices_ptr(std::move(indices_ptr)),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index),
        my_core(
            std::move(chunk_workspace),
            coordinator, 
            slab_stats,
            row,
            std::move(oracle), 
            my_indices_ptr->size(),
            my_needs_value,
            my_needs_index
        )
    {}
 
    tatami::SparseRange<Value_, Index_> fetch(Index_ i, Value_* value_buffer, Index_* index_buffer) {
        auto fetched = my_core.fetch_raw(i, my_row, *my_indices_ptr, my_tmp_indices);
        return process_sparse_slab(fetched, value_buffer, index_buffer, my_needs_value, my_needs_index);
    }
 
private:
    bool my_row;
    tatami::VectorPtr<Index_> my_indices_ptr;
    std::vector<Index_> my_tmp_indices;
    bool my_needs_value, my_needs_index;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
/**************************
 **** Densified classes ***
 **************************/
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class DensifiedFull : public tatami::DenseExtractor<oracle_, Value_, Index_> {
public:
    DensifiedFull(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle, 
        const tatami::Options&
    ) :
        my_row(row),
        my_non_target_dim(coordinator.get_non_target_dim(row)),
        my_core(
            std::move(chunk_workspace),
            coordinator,
            slab_stats,
            row,
            std::move(oracle), 
            my_non_target_dim,
            true, 
            true
        )
    {}
 
    const Value_* fetch(Index_ i, Value_* buffer) {
        auto contents = my_core.fetch_raw(i, my_row, 0, my_non_target_dim);
 
        Index_ num = contents.first->number[contents.second];
        auto vptr = contents.first->values[contents.second];
        auto iptr = contents.first->indices[contents.second];
 
        std::fill_n(buffer, my_non_target_dim, 0);
        for (Index_ x = 0; x < num; ++x, ++iptr, ++vptr) {
            buffer[*iptr] = *vptr;
        }
        return buffer;
    }
 
private:
    bool my_row;
    Index_ my_non_target_dim;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class DensifiedBlock : public tatami::DenseExtractor<oracle_, Value_, Index_> {
public:
    DensifiedBlock(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle,
        Index_ block_start, 
        Index_ block_length,
        const tatami::Options&
    ) :
        my_row(row),
        my_block_start(block_start),
        my_block_length(block_length),
        my_core(
            std::move(chunk_workspace),
            coordinator,
            slab_stats,
            row,
            std::move(oracle), 
            block_length,
            true,
            true
        )
    {}
 
    const Value_* fetch(Index_ i, Value_* buffer) {
        auto contents = my_core.fetch_raw(i, my_row, my_block_start, my_block_length);
 
        auto vptr = contents.first->values[contents.second];
        auto iptr = contents.first->indices[contents.second];
        auto num = contents.first->number[contents.second];
 
        std::fill_n(buffer, my_block_length, 0);
        for (Index_ x = 0; x < num; ++x, ++iptr, ++vptr) {
            buffer[*iptr - my_block_start] = *vptr;
        }
        return buffer;
    }
 
private:
    bool my_row;
    Index_ my_block_start, my_block_length;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
template<bool solo_, bool oracle_, bool use_subset_, typename Value_, typename Index_, typename ChunkValue_, class WorkspacePtr_>
class DensifiedIndex : public tatami::DenseExtractor<oracle_, Value_, Index_> {
public:
    DensifiedIndex(
        WorkspacePtr_ chunk_workspace,
        const ChunkCoordinator<true, ChunkValue_, Index_>& coordinator, 
        const SlabCacheStats<Index_>& slab_stats,
        bool row,
        tatami::MaybeOracle<oracle_, Index_> oracle, 
        tatami::VectorPtr<Index_> indices_ptr,
        const tatami::Options&
    ) :
        my_row(row),
        my_indices_ptr(std::move(indices_ptr)),
        my_core(
            std::move(chunk_workspace),
            coordinator, 
            slab_stats,
            row,
            std::move(oracle), 
            my_indices_ptr->size(),
            true,
            true
        )
    {
        const auto& indices = *my_indices_ptr;
        if (!indices.empty()) {
            my_remap_offset = indices.front();
            Index_ alloc = indices.back() - my_remap_offset + 1; // alloc must be <= dim extent, which should fit in an Index_.
            tatami::resize_container_to_Index_size(my_remap, alloc);
            Index_ counter = 0;
            for (auto i : indices) {
                my_remap[i - my_remap_offset] = counter;
                ++counter;
            }
        }
    }
 
    const Value_* fetch(Index_ i, Value_* buffer) {
        auto contents = my_core.fetch_raw(i, my_row, *my_indices_ptr, my_tmp_indices);
 
        auto vptr = contents.first->values[contents.second];
        auto iptr = contents.first->indices[contents.second];
        auto num = contents.first->number[contents.second];
 
        auto nidx = my_indices_ptr->size();
        std::fill_n(buffer, nidx, 0);
        for (Index_ x = 0; x <num; ++x, ++iptr, ++vptr) {
            buffer[my_remap[*iptr - my_remap_offset]] = *vptr;
        }
        return buffer;
    }
 
private:
    bool my_row;
    tatami::VectorPtr<Index_> my_indices_ptr;
    Index_ my_remap_offset = 0;
    std::vector<Index_> my_remap;
    std::vector<Index_> my_tmp_indices;
    SparseCore<solo_, oracle_, use_subset_, Value_, Index_, ChunkValue_, WorkspacePtr_> my_core;
};
 
}
template<typename Value_, typename Index_, typename ChunkValue_, class Manager_ = CustomSparseChunkedMatrixManager<ChunkValue_, Index_> >
class CustomSparseChunkedMatrix : public tatami::Matrix<Value_, Index_> {
public:
    CustomSparseChunkedMatrix(std::shared_ptr<Manager_> manager, const CustomSparseChunkedMatrixOptions& opt) : 
        my_manager(std::move(manager)),
        my_coordinator(my_manager->row_stats(), my_manager->column_stats()),
        my_cache_size_in_bytes(opt.maximum_cache_size),
        my_require_minimum_cache(opt.require_minimum_cache),
        my_cache_subset(opt.cache_subset)
    {}
 
private:
    std::shared_ptr<Manager_> my_manager;
    CustomChunkedMatrix_internal::ChunkCoordinator<true, ChunkValue_, Index_> my_coordinator;
    std::size_t my_cache_size_in_bytes;
    bool my_require_minimum_cache;
    bool my_cache_subset;
 
public:
    Index_ nrow() const { 
        return my_coordinator.get_nrow();
    }
 
    Index_ ncol() const { 
        return my_coordinator.get_ncol();
    }
 
    bool prefer_rows() const { 
        return my_manager->prefer_rows();
    }
 
    bool uses_oracle(bool) const { 
        return true; 
    }
 
    double prefer_rows_proportion() const { 
        return static_cast<double>(my_manager->prefer_rows());
    }
 
    bool is_sparse() const {
        return true;
    }
 
    double is_sparse_proportion() const {
        return 1;
    }
 
    using tatami::Matrix<Value_, Index_>::dense;
 
    using tatami::Matrix<Value_, Index_>::sparse;
 
    /********************
     *** Myopic dense ***
     ********************/
private:
    template<
        template<bool, typename, typename> class Interface_, 
        bool oracle_, 
        template<bool, bool, bool, typename, typename, typename, class> class Extractor_,
        typename ... Args_
    >
    std::unique_ptr<Interface_<oracle_, Value_, Index_> > raw_internal(bool row, Index_ non_target_length, const tatami::Options& opt, Args_&& ... args) const {
        std::size_t element_size = (opt.sparse_extract_value ? sizeof(ChunkValue_) : 0) + (opt.sparse_extract_index ? sizeof(Index_) : 0);
        auto stats = [&]{
            if (row) {
                // Remember, the num_chunks_per_column is the number of slabs needed to divide up all the *rows* of the matrix.
                return SlabCacheStats<Index_>(
                    my_coordinator.get_chunk_nrow(),
                    non_target_length,
                    my_coordinator.get_num_chunks_per_column(), // already Index_, no need to do a protected cast.
                    my_cache_size_in_bytes,
                    element_size,
                    my_require_minimum_cache
                );
            } else {
                // Same as above, but this time, the num_chunks_per_row is the number of slabs needed to divide up all the *columns* of the matrix.
                return SlabCacheStats<Index_>(
                    my_coordinator.get_chunk_ncol(),
                    non_target_length,
                    my_coordinator.get_num_chunks_per_row(),
                    my_cache_size_in_bytes,
                    element_size,
                    my_require_minimum_cache
                );
            }
        }();
 
        auto wrk = my_manager->new_workspace_exact();
        if (stats.max_slabs_in_cache == 0) {
            return std::make_unique<Extractor_<true, oracle_, false, Value_, Index_, ChunkValue_, I<decltype(wrk)> > >(std::move(wrk), my_coordinator, stats, row, std::forward<Args_>(args)...);
        } else if constexpr(oracle_) {
            if (my_cache_subset) {
                return std::make_unique<Extractor_<false, true, true, Value_, Index_, ChunkValue_, I<decltype(wrk)> > >(std::move(wrk), my_coordinator, stats, row, std::forward<Args_>(args)...);
            } else {
                return std::make_unique<Extractor_<false, true, false, Value_, Index_, ChunkValue_, I<decltype(wrk)> > >(std::move(wrk), my_coordinator, stats, row, std::forward<Args_>(args)...);
            }
        } else {
            return std::make_unique<Extractor_<false, false, false, Value_, Index_, ChunkValue_, I<decltype(wrk)> > >(std::move(wrk), my_coordinator, stats, row, std::forward<Args_>(args)...);
        }
    }
 
public:
    std::unique_ptr<tatami::MyopicDenseExtractor<Value_, Index_> > dense(bool row, const tatami::Options& opt) const {
        return raw_internal<tatami::DenseExtractor, false, CustomChunkedMatrix_internal::DensifiedFull>(row, my_coordinator.get_non_target_dim(row), opt, false, opt);
    }
 
    std::unique_ptr<tatami::MyopicDenseExtractor<Value_, Index_> > dense(bool row, Index_ block_start, Index_ block_length, const tatami::Options& opt) const {
        return raw_internal<tatami::DenseExtractor, false, CustomChunkedMatrix_internal::DensifiedBlock>(row, block_length, opt, false, block_start, block_length, opt);
    }
 
    std::unique_ptr<tatami::MyopicDenseExtractor<Value_, Index_> > dense(bool row, tatami::VectorPtr<Index_> indices_ptr, const tatami::Options& opt) const {
        auto num_indices = indices_ptr->size();
        return raw_internal<tatami::DenseExtractor, false, CustomChunkedMatrix_internal::DensifiedIndex>(row, num_indices, opt, false, std::move(indices_ptr), opt);
    }
 
    /***********************
     *** Oracular dense ***
     ***********************/
public:
    std::unique_ptr<tatami::OracularDenseExtractor<Value_, Index_> > dense(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        const tatami::Options& opt) 
    const {
        return raw_internal<tatami::DenseExtractor, true, CustomChunkedMatrix_internal::DensifiedFull>(row, my_coordinator.get_non_target_dim(row), opt, std::move(oracle), opt);
    }
 
    std::unique_ptr<tatami::OracularDenseExtractor<Value_, Index_> > dense(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        Index_ block_start, 
        Index_ block_length, 
        const tatami::Options& opt) 
    const {
        return raw_internal<tatami::DenseExtractor, true, CustomChunkedMatrix_internal::DensifiedBlock>(row, block_length, opt, std::move(oracle), block_start, block_length, opt);
    }
 
    std::unique_ptr<tatami::OracularDenseExtractor<Value_, Index_> > dense(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        tatami::VectorPtr<Index_> indices_ptr, 
        const tatami::Options& opt) 
    const {
        auto num_indices = indices_ptr->size();
        return raw_internal<tatami::DenseExtractor, true, CustomChunkedMatrix_internal::DensifiedIndex>(row, num_indices, opt, std::move(oracle), std::move(indices_ptr), opt);
    }
 
    /*********************
     *** Myopic sparse ***
     *********************/
public:
    std::unique_ptr<tatami::MyopicSparseExtractor<Value_, Index_> > sparse(bool row, const tatami::Options& opt) const {
        return raw_internal<tatami::SparseExtractor, false, CustomChunkedMatrix_internal::SparseFull>(row, my_coordinator.get_non_target_dim(row), opt, false, opt);
    }
 
    std::unique_ptr<tatami::MyopicSparseExtractor<Value_, Index_> > sparse(bool row, Index_ block_start, Index_ block_length, const tatami::Options& opt) const {
        return raw_internal<tatami::SparseExtractor, false, CustomChunkedMatrix_internal::SparseBlock>(row, block_length, opt, false, block_start, block_length, opt);
    }
 
    std::unique_ptr<tatami::MyopicSparseExtractor<Value_, Index_> > sparse(bool row, tatami::VectorPtr<Index_> indices_ptr, const tatami::Options& opt) const {
        auto num_indices = indices_ptr->size();
        return raw_internal<tatami::SparseExtractor, false, CustomChunkedMatrix_internal::SparseIndex>(row, num_indices, opt, false, std::move(indices_ptr), opt);
    }
 
    /***********************
     *** Oracular sparse ***
     ***********************/
public:
    std::unique_ptr<tatami::OracularSparseExtractor<Value_, Index_> > sparse(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        const tatami::Options& opt) 
    const {
        return raw_internal<tatami::SparseExtractor, true, CustomChunkedMatrix_internal::SparseFull>(row, my_coordinator.get_non_target_dim(row), opt, std::move(oracle), opt);
    }
 
    std::unique_ptr<tatami::OracularSparseExtractor<Value_, Index_> > sparse(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        Index_ block_start, 
        Index_ block_length, 
        const tatami::Options& opt) 
    const {
        return raw_internal<tatami::SparseExtractor, true, CustomChunkedMatrix_internal::SparseBlock>(row, block_length, opt, std::move(oracle), block_start, block_length, opt);
    }
 
    std::unique_ptr<tatami::OracularSparseExtractor<Value_, Index_> > sparse(
        bool row, 
        std::shared_ptr<const tatami::Oracle<Index_> > oracle, 
        tatami::VectorPtr<Index_> indices_ptr, 
        const tatami::Options& opt) 
    const {
        auto num_indices = indices_ptr->size();
        return raw_internal<tatami::SparseExtractor, true, CustomChunkedMatrix_internal::SparseIndex>(row, num_indices, opt, std::move(oracle), std::move(indices_ptr), opt);
    }
};
 
}
 
#endif