Struct HybMatrix 

pub struct HybMatrix<T: GpuFloat> { /* private fields */ }

A sparse matrix in HYB (Hybrid ELL+COO) format, stored on host.

The ELL portion uses column-major layout: element (row, k) is at index k * rows + row. Unused ELL slots have column index HYB_ELL_SENTINEL and a zero value.

The COO portion stores overflow entries as unsorted (row, col, value) triplets.

Implementations

impl<T: GpuFloat> HybMatrix<T>

pub fn new( rows: usize, cols: usize, ell_width: usize, ell_col_indices: Vec<i32>, ell_values: Vec<T>, coo_row_indices: Vec<i32>, coo_col_indices: Vec<i32>, coo_values: Vec<T>, ) -> SparseResult<Self>

Creates a new HYB matrix from raw host-side arrays.

Arguments

• rows – Number of rows.
• cols – Number of columns.
• ell_width – Maximum entries per row in the ELL portion.
• ell_col_indices – ELL column indices, length rows * ell_width, column-major. Unused entries must be HYB_ELL_SENTINEL.
• ell_values – ELL values, length rows * ell_width, column-major.
• coo_row_indices – COO overflow row indices.
• coo_col_indices – COO overflow column indices.
• coo_values – COO overflow values.

Errors

Returns SparseError::InvalidFormat if array lengths are inconsistent.

pub fn from_csr( csr: &CsrMatrix<T>, partition: HybPartition, ) -> SparseResult<Self>

Constructs a HYB matrix from a CSR matrix using the given partition strategy.

Downloads the CSR data from GPU, computes per-row nnz, determines the ELL width according to partition, then fills ELL up to that width and overflows the remainder into COO.

Errors

Returns SparseError::Cuda on GPU transfer failure.

pub fn from_coo( coo: &CooMatrix<T>, partition: HybPartition, ) -> SparseResult<Self>

Constructs a HYB matrix from a COO matrix (via CSR intermediate) using the given partition strategy.

Converts COO to CSR first, then builds HYB from the CSR data.

Errors

Returns SparseError::Cuda on GPU transfer failure.

pub fn to_csr(&self) -> SparseResult<CsrMatrix<T>>

Converts this HYB matrix back to CSR format, uploading to GPU.

Merges the ELL and COO portions into a single CSR representation.

Errors

Returns SparseError::Cuda on GPU allocation failure. Returns SparseError::ZeroNnz if the matrix has no non-zeros.

pub fn total_nnz(&self) -> usize

Returns the total number of non-zeros (ELL + COO).

pub fn nnz(&self) -> usize

Returns the number of non-zeros (alias for total_nnz).

pub fn is_empty(&self) -> bool

Returns true if the matrix contains no non-zero entries.

pub fn ell_nnz(&self) -> usize

Returns the number of non-zeros stored in the ELL portion.

pub fn rows(&self) -> usize

Returns the number of rows.

pub fn cols(&self) -> usize

Returns the number of columns.

pub fn ell_width(&self) -> usize

Returns the ELL width (max entries per row in ELL portion).

pub fn ell_col_indices(&self) -> &[i32]

Returns a reference to the ELL column indices (column-major).

pub fn ell_values(&self) -> &[T]

Returns a reference to the ELL values (column-major).

pub fn coo_nnz(&self) -> usize

Returns the number of overflow entries in the COO portion.

pub fn coo_row_indices(&self) -> &[i32]

Returns a reference to the COO overflow row indices.

pub fn coo_col_indices(&self) -> &[i32]

Returns a reference to the COO overflow column indices.

pub fn coo_values(&self) -> &[T]

Returns a reference to the COO overflow values.

pub fn statistics(&self) -> HybStatistics

Computes statistics describing the quality of the HYB partition.

The returned HybStatistics includes the fraction of nnz in each portion, the ELL padding ratio, and memory consumption compared to an equivalent CSR matrix.

Trait Implementations

impl<T: Clone + GpuFloat> Clone for HybMatrix<T>

fn clone(&self) -> HybMatrix<T>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: Debug + GpuFloat> Debug for HybMatrix<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.