Struct CooMatrix 

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

A sparse matrix in Coordinate (COO) format, stored on GPU.

The matrix has shape (rows, cols) with nnz non-zero elements. Each non-zero is represented by its row index, column index, and value.

Implementations

impl<T: GpuFloat> CooMatrix<T>

pub fn from_host( rows: u32, cols: u32, row_idx: &[i32], col_idx: &[i32], values: &[T], ) -> SparseResult<Self>

Creates a COO matrix from host-side arrays, uploading to GPU.

The triplets are assumed to be unsorted unless the caller knows otherwise (use with_sorted to override).

Arguments

• rows – Number of rows.
• cols – Number of columns.
• row_idx – Row indices of length nnz.
• col_idx – Column indices of length nnz.
• values – Non-zero values of length nnz.

Errors

Returns SparseError::InvalidFormat if array lengths are inconsistent.

pub fn from_device( rows: u32, cols: u32, nnz: u32, row_idx: DeviceBuffer<i32>, col_idx: DeviceBuffer<i32>, values: DeviceBuffer<T>, ) -> SparseResult<Self>

Creates a COO matrix from pre-allocated device buffers.

This is the unchecked escape hatch: only buffer lengths are validated. The contents of row_idx/col_idx are not range checked against rows/cols. Passing indices outside [0, rows)/[0, cols) will cause out-of-bounds device memory access in downstream operations (e.g. to_csr, to_csc, or SpMV/SpMM kernels). Callers are responsible for ensuring indices are in range before calling this constructor; prefer from_host when the data originates on the host, as it validates the full range.

Errors

Returns SparseError::InvalidFormat if buffer lengths are inconsistent.

pub fn with_sorted(self, sorted: bool) -> Self

Mark the COO matrix as sorted by (row, col).

This is a hint for conversion routines; the caller is responsible for ensuring correctness.

pub fn to_host(&self) -> SparseResult<(Vec<i32>, Vec<i32>, Vec<T>)>

Downloads the COO arrays from GPU to host memory.

Errors

Returns SparseError::Cuda on transfer failure.

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

Converts this COO matrix to CSR format.

Downloads to host, builds row pointers via histogram, then uploads.

Errors

Returns SparseError::Cuda on transfer failure. Returns SparseError::InvalidFormat if any downloaded row or column index is out of range (this can happen for matrices constructed via the unchecked from_device constructor).

pub fn to_csc(&self) -> SparseResult<CscMatrix<T>>

Converts this COO matrix to CSC format.

Downloads to host, builds column pointers, then uploads.

Errors

Returns SparseError::Cuda on transfer failure. Returns SparseError::InvalidFormat if any downloaded row or column index is out of range (this can happen for matrices constructed via the unchecked from_device constructor).

pub fn is_sorted(&self) -> bool

Returns whether the triplets are sorted by (row, col).

pub fn rows(&self) -> u32

Returns the number of rows.

pub fn cols(&self) -> u32

Returns the number of columns.

pub fn nnz(&self) -> u32

Returns the number of non-zero elements.

pub fn row_idx(&self) -> &DeviceBuffer<i32>

Returns a reference to the row index device buffer.

pub fn col_idx(&self) -> &DeviceBuffer<i32>

Returns a reference to the column index device buffer.

pub fn values(&self) -> &DeviceBuffer<T>

Returns a reference to the values device buffer.