FastCuda is a handwritten CUDA operator library featuring progressive GEMM and Reduce kernels, cuBLAS benchmarking, and C/C++/Python interfaces for learning, profiling, and performance optimization.
Fast SpMM implementation on GPUs for GNN (IPDPS'23)
Codes for DTC-SpMM (ASPLOS'24)
Awesome list of sparsity in deep learning: pruning, sparse attention, N:M structured sparsity, KV-cache & LLM sparsity, MoE, sparse matrix accelerators (SpMM/SpGEMM), GNN hardware, sparse GPU kernels & PIM. 590 papers organized by mechanism.
📚 Coursework for "Introduction to High Performance Computing" (30240192)
Searching for peptide candidates using sparse matrix + matrix/vector multiplication.
Proof-of-concept implementation of a search engine that uses sparse matrix multiplication to identify the best peptide candidates for a given mass spectrum.
Input-aware CUDA auto-scheduler for CSR SpMM, SDDMM, and sparse attention in graph neural networks.
Regime-aware sparse matrix multiplication for GNN workloads on GPUs (Future Generation Computer Systems, Elsevier). Six structural regimes, an interpretable seven-rule router, and five specialized kernel families over native CSR and Tensor Cores.
Reproducible Instruction Roofline analysis of cuSPARSE and Ginkgo SpMM on RTX 4090 using Nsight Compute metrics.
To associate your repository with the spmm topic, visit your repo's landing page and select "manage topics."