Annual Joy Modi Lecture: SPIRAL AI for High Performance Code

Wednesday, May 21, 2025

11:00 AM-12:00 PM

Talk Summary:
This talk presents an overview of the SPIRAL system, a 25-year collaborative effort by Carnegie Mellon, Drexel, and UIUC, now available as BSD open-source. SPIRAL is a rule-based AI system that encodes expert knowledge about algorithms, architectures, and program transformations using a formalism called operator language (OL). The system synthesizes highly optimized computational kernels by solving a constrained optimization problem through semantics-preserving rewriting. This enables formal correctness proofs of the generated code. The talk also introduces a novel semantic lifting approach for deriving code meaning and briefly covers SPIRAL-based libraries—FFTX, PROTOX, GBTLX, and NTTX—available as C++ eDSLs and plugins for LLVM, Python, and Julia.

Bio:
Franz Franchetti is a Professor of Electrical and Computer Engineering at Carnegie Mellon University. He holds an M.Sc. and Ph.D. in Computational Mathematics from the Vienna University of Technology. His research centers on automatic performance tuning, program generation for parallel platforms, and algorithm/hardware co-synthesis, with a focus on developing SPIRAL for generating high-performance software libraries.