What is it about?
Tensors with many zeros are called sparse tensors. Sparse tensors provide opportunities for reducing computational time as well as storage requirements. Exploiting such opportunities by hand is cumbersome and error-prone, however. Therefore, we propose treating sparsity as a property of tensor, not a tedious implementation detail and letting a compiler automatically generate sparse code. This paper discusses integrating this idea into MLIR.
Featured Image
Photo by Pawel Czerwinski on Unsplash
Why is it important?
Sparse compilers can greatly simplify sparse code development and increase productivity.
Perspectives
We welcome the open-source and research communities to contribute to this ongoing effort in MLIR.
Aart Bik
Alphabet Inc
Read the Original
This page is a summary of: Compiler Support for Sparse Tensor Computations in MLIR, ACM Transactions on Architecture and Code Optimization, December 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3544559.
You can read the full text:
Resources
Contributors
The following have contributed to this page
