cs639-parallel-throughput-opt-prog
Parallel and Throughput-Optimized Programming by Prof. Sifakis
Key takeaways
- With sparse matrices, it is important to identify patterns for transformations to result in reordering and parallelism performance at a divided level (i.e. a chunk of a larger full matrix or the factor in cholesky). Think about using ML to apply the right set of transformations to achieve the right performance on sparsity engineering.