Module ops 

Sparse matrix operations.

This module provides GPU-accelerated sparse matrix operations:

• spmv – Sparse matrix-vector multiplication (y = alphaAx + beta*y)
• spmm – Sparse matrix-dense matrix multiplication (C = alphaAB + beta*C)
• spgemm – Sparse matrix-sparse matrix multiplication (C = A*B)
• spgemm_symbolic – Gustavson symbolic SpGEMM: the sparsity pattern of C = A*B (host CSR, values not computed)
• sptrsv – Sparse triangular solve (Lx = b or Ux = b)
• sddmm – Sampled Dense-Dense Matrix Multiply
• krylov – Krylov subspace methods (Lanczos & Arnoldi iteration)
• matrix_powers – Sparse matrix powers (A^k) and polynomial evaluation

Re-exports

pub use auto_spmv::RecommendedFormat;

pub use auto_spmv::SpMatFormat;

pub use auto_spmv::analyze_sparsity;

pub use auto_spmv::auto_spmv;

pub use auto_spmv::recommend_format;

pub use auto_spmv::select_format;

pub use batched::BatchScheduler;

pub use batched::BatchedSpGEMM;

pub use batched::BatchedSpMV;

pub use batched::BatchedSpMVPlan;

pub use batched::BatchedTriSolve;

pub use batched::Strategy;

pub use batched::UniformBatchedSpMV;

pub use batched::batched_spmv_cpu;

pub use batched::generate_batched_spmv_ptx;

pub use batched::mixed_precision_spmv_cpu;

pub use krylov::ArnoldiConfig;

pub use krylov::ArnoldiPlan;

pub use krylov::ArnoldiResult;

pub use krylov::EigenTarget;

pub use krylov::LanczosConfig;

pub use krylov::LanczosPlan;

pub use krylov::LanczosResult;

pub use matrix_powers::MatrixPowerConfig;

pub use matrix_powers::MatrixPowerResult;

pub use matrix_powers::estimate_power_nnz;

pub use matrix_powers::sparse_identity;

pub use matrix_powers::sparse_matrix_polynomial;

pub use matrix_powers::sparse_matrix_power;

pub use mixed_precision_spmv::ComputePrecision;

pub use mixed_precision_spmv::MixedPrecisionConfig;

pub use mixed_precision_spmv::MixedPrecisionPlan;

pub use mixed_precision_spmv::MixedPrecisionStats;

pub use mixed_precision_spmv::MixedSpMVAlgo;

pub use mixed_precision_spmv::StoragePrecision;

pub use mixed_precision_spmv::estimate_precision_loss;

pub use mixed_precision_spmv::generate_mixed_scalar_spmv_ptx;

pub use mixed_precision_spmv::generate_mixed_vector_spmv_ptx;

pub use mixed_precision_spmv::generate_packed_vector_spmv_ptx;

pub use mixed_precision_spmv::plan_mixed_precision_spmv;

pub use mixed_precision_spmv::validate_mixed_precision_config;

pub use sddmm::sddmm;

pub use spgemm::spgemm_numeric;

pub use spgemm::spgemm_symbolic;

pub use spgemm_estimate::EstimationMethod;

pub use spgemm_estimate::SpGEMMEstimate;

pub use spgemm_estimate::auto_estimate_spgemm;

pub use spgemm_estimate::count_nnz_exact;

pub use spgemm_estimate::estimate_nnz_sampling;

pub use spgemm_estimate::estimate_nnz_upper_bound;

pub use spgemm_estimate::estimate_spgemm_memory;

pub use spgemm_merge::spgemm_merge;

pub use spgemm_symbolic::SymbolicPattern;

pub use spgemm_symbolic::spgemm_symbolic_pattern;

pub use spmm::spmm;

pub use spmv::SpMVAlgo;

pub use spmv::spmv;

pub use spmv_bsr::spmv_bsr;

pub use spmv_csr5::csr5_spmv;

pub use spmv_ell::spmv_ell;

pub use sptrsv::sptrsv;

pub use tensor::EdgeFeatures;

pub use tensor::GnnSparseConfig;

pub use tensor::MessagePassingOp;

pub use tensor::add_self_loops;

pub use tensor::compute_degree_matrix;

pub use tensor::gather;

pub use tensor::scatter_reduce;

pub use tensor::sparse_attention_message;

pub use tensor::sparse_message_passing;

pub use tensor::sparse_row_softmax;

pub use tensor::symmetric_normalize;

Modules

auto_spmv

Heuristic-based format selection and auto-dispatching SpMV.

batched

Batched sparse operations.

krylov

Krylov subspace methods for sparse eigenvalue computation.

matrix_powers

Sparse matrix powers and polynomial evaluation.

mixed_precision_spmv

Mixed-precision sparse matrix-vector multiplication (SpMV).

sddmm

Sampled Dense-Dense Matrix Multiply (SDDMM).

spgemm

Sparse matrix-sparse matrix multiplication (SpGEMM).

spgemm_estimate

SpGEMM memory estimation for C = A * B.

spgemm_merge

Merge-based SpGEMM: C = A * B using merge-path load balancing.

spgemm_symbolic

Symbolic SpGEMM (Gustavson) – the sparsity pattern of C = A * B.

spmm

Sparse matrix-dense matrix multiplication (SpMM).

spmv

Sparse matrix-vector multiplication (SpMV).

spmv_bsr

BSR SpMV kernel.

spmv_csr5

CSR5 SpMV kernel.

spmv_ell

ELL-optimized SpMV kernel.

sptrsv

Sparse triangular solve (SpTRSV).

tensor

Sparse tensor operations for GNN (Graph Neural Network) workloads.