ECE graduate student Kazem Cheshmi working with Professor Maryam Mehri Dehnavi has won first place at the ACM student research competition of the 2017 International Symposium on Code Generation and Optimization. The ACM Student Research Competition (SRC), sponsored by Microsoft, offers a unique forum for undergraduate and graduate students to present their original research before a panel of judges and attendees at well-known ACM-sponsored and co-sponsored conferences.
Kazem won first place for his project entitled "Decoupling Symbolic Analysis from Numerical Factorization in Sparse Direct Solvers." This project seeks to build a domain-specific pattern-aware compiler for sparse matrix computations. A two-phase algorithm-aware compiler leverages domain knowledge to decouple the symbolic analysis stage in sparse codes from the numerical computation phase. The proposed approach not only reduces the overhead of symbolic analysis in sparse numerical codes but also enables the application of algorithm-specific and low-level compiler transformations. The algorithm-aware inspection exposes the dependence structure of the sparse matrix method and enables a space of optimization for applying aggressive compiler transformations not possible in specialized libraries. The compiler generates high-performance codes for low-level and high-level sparse linear algebra such as LU factorization, Cholesky factorization, QR decomposition, SVD, and triangular solvers used to find the solution to linear systems and singular values and eigenvalues of a matrix.
Congratulations to Kazem and Maryam on winning this competition!