TileSpMM: A Variable-Size Tiled Algorithm for Sparse Matrix-Matrix Multiplication on Tensor Cores
This work proposes TileSpMM, which breaks the static-granularity bottleneck through a variable-size tiling algorithm that dynamically adapts to local sparsity patterns, and is equipped with an adaptive load-balancing strategy and customized granularity-specific kernels to improve hardware utilization and mitigate compu...