What is it about?
We optimize Sparse Matrix Vector multiplication (SpMV) using a mixed precision strategy (MpSpMV) the decision to lower to single precision is data driven, based on individual nonzero values of the sparse matrix. On all real-valued matrices from the Sparse Matrix Collection, we obtain a maximum speedup of 2.61× and average speedup of 1.06× over double precision, while maintaining higher accuracy compared to single precision.
Photo by Jan Kaluza on Unsplash
Why is it important?
(1) It reduces execution time of the computation (2) it reduces the size of the input data and therefore reduces data movement (3) it provides an opportunity for increased parallelism on the GPU
Read the Original
This page is a summary of: Data-driven Mixed Precision Sparse Matrix Vector Multiplication for GPUs, ACM Transactions on Architecture and Code Optimization, January 2020, ACM (Association for Computing Machinery), DOI: 10.1145/3371275.
You can read the full text:
The following have contributed to this page