Struct BatchedSpGEMM 

pub struct BatchedSpGEMM { /* private fields */ }

Batched sparse-sparse matrix multiplication.

Computes C_i = A_i * B_i for a batch of sparse matrix pairs. The baseline runs each SpGEMM independently on the host using a hash-table accumulation approach.

Implementations

impl BatchedSpGEMM

pub fn new() -> Self

Create a new batched SpGEMM executor.

pub fn execute<T: GpuFloat + Add<Output = T> + Mul<Output = T> + AddAssign>( a_batch: &[CsrMatrix<T>], b_batch: &[CsrMatrix<T>], ) -> SparseResult<Vec<(Vec<i32>, Vec<i32>, Vec<T>, u32, u32)>>

Execute batched SpGEMM on the host.

For each pair (A_i, B_i), computes C_i = A_i * B_i using a row-by-row hash-table accumulation.

Returns host-side CSR triples (row_ptr, col_idx, values) for each result matrix.

Errors

Returns SparseError::InvalidArgument if batch sizes differ or are empty. Returns SparseError::DimensionMismatch if dimensions are incompatible. Returns SparseError::Cuda on download failure.

pub fn execute_host<T: GpuFloat + Add<Output = T> + Mul<Output = T> + AddAssign>( a_row_ptrs: &[Vec<i32>], a_col_indices: &[Vec<i32>], a_values: &[Vec<T>], a_rows: &[usize], a_cols: &[usize], b_row_ptrs: &[Vec<i32>], b_col_indices: &[Vec<i32>], b_values: &[Vec<T>], b_cols: &[usize], ) -> SparseResult<Vec<(Vec<i32>, Vec<i32>, Vec<T>, usize, usize)>>

Execute batched SpGEMM from host-side CSR arrays (no GPU required).

Errors

Returns SparseError::InvalidArgument on length mismatch. Returns SparseError::DimensionMismatch on dimension mismatch.

Trait Implementations

impl Debug for BatchedSpGEMM

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

Formats the value using the given formatter.

impl Default for BatchedSpGEMM

fn default() -> Self

Returns the “default value” for a type.