SparseDitto: Customizing GPU Kernels for Different Sparsity Patterns with LLM-Based Agentic System
This work presents SparseDitto, an LLM-based system that constructs a GPU kernel for each matrix, operator, and target GPU, and supports SpMV, SpMM, and SpGEMM within a unified design framework.
Shiyang Li, Yan-Zhi Wang, Ming-Yi Hong
· 0 citations