TY - GEN
T1 - Sympiler
T2 - 2017 International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2017
AU - Cheshmi, Kazem
AU - Kamil, Shoaib
AU - Strout, Michelle Mills
AU - Dehnavi, Maryam Mehri
N1 - Funding Information:
This work is supported by the U.S. National Science Foundation (NSF) Award Numbers CCF-1657175 and CCF-1564074. This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by the National Science Foundation grant number ACI-1548562.
Publisher Copyright:
© 2017 ACM.
PY - 2017
Y1 - 2017
N2 - Sympiler is a domain-specific code generator that optimizes sparse matrix computations by decoupling the symbolic analysis phase from the numerical manipulation stage in sparse codes. The computation patterns in sparse numerical methods are guided by the input sparsity structure and the sparse algorithm itself. In many real-world simulations, the sparsity pattern changes little or not at all. Sympiler takes advantage of these properties to symbolically analyze sparse codes at compile time and to apply inspector-guided transformations that enable applying low-level transformations to sparse codes. As a result, the Sympiler-generated code outperforms highly-optimized matrix factorization codes from commonly-used specialized libraries, obtaining average speedups over Eigen and CHOLMOD of 3.8 × and 1.5 × respectively.
AB - Sympiler is a domain-specific code generator that optimizes sparse matrix computations by decoupling the symbolic analysis phase from the numerical manipulation stage in sparse codes. The computation patterns in sparse numerical methods are guided by the input sparsity structure and the sparse algorithm itself. In many real-world simulations, the sparsity pattern changes little or not at all. Sympiler takes advantage of these properties to symbolically analyze sparse codes at compile time and to apply inspector-guided transformations that enable applying low-level transformations to sparse codes. As a result, the Sympiler-generated code outperforms highly-optimized matrix factorization codes from commonly-used specialized libraries, obtaining average speedups over Eigen and CHOLMOD of 3.8 × and 1.5 × respectively.
KW - domain-specific compilation
KW - loop transformations
KW - Matrix computations
KW - sparse methods
UR - http://www.scopus.com/inward/record.url?scp=85142252952&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85142252952&partnerID=8YFLogxK
U2 - 10.1145/3126908.3126936
DO - 10.1145/3126908.3126936
M3 - Conference contribution
AN - SCOPUS:85040185861
T3 - International Conference for High Performance Computing, Networking, Storage and Analysis, SC
BT - SC 2017 - International Conference for High Performance Computing, Networking, Storage and Analysis
PB - IEEE Computer Society
Y2 - 12 November 2017 through 17 November 2017
ER -