26470

DISTAL: The Distributed Tensor Algebra Compiler

Rohan Yadav, Alex Aiken, Fredrik Kjolstad
Stanford University, Stanford, CA, USA
arXiv:2203.08069 [cs.PL], (17 Mar 2022)

@article{yadav2022distal,

   title={DISTAL: The Distributed Tensor Algebra Compiler},

   author={Yadav, Rohan and Aiken, Alex and Kjolstad, Fredrik},

   journal={arXiv preprint arXiv:2203.08069},

   year={2022}

}

Download Download (PDF)   View View   Source Source   

397

views

We introduce DISTAL, a compiler for dense tensor algebra that targets modern distributed and heterogeneous systems. DISTAL lets users independently describe how tensors and computation map onto target machines through separate format and scheduling languages. The combination of choices for data and computation distribution creates a large design space that includes many algorithms from both the past (e.g., Cannon’s algorithm) and the present (e.g., COSMA). DISTAL compiles a tensor algebra domain specific language to a distributed task-based runtime system and supports nodes with multi-core CPUs and multiple GPUs. Code generated by DISTAL is competitive with optimized codes for matrix multiply on 256 nodes of the Lassen supercomputer and outperforms existing systems by between 1.8x to 3.7x (with a 45.7x outlier) on higher order tensor operations.
No votes yet.
Please wait...

* * *

* * *

* * *

HGPU group © 2010-2022 hgpu.org

All rights belong to the respective authors

Contact us: