Saman Amarasinghe gives his talk, “Compiler 2.0: Building the Next Generation of Compilers with Machine Learning,” on October 28, 2026.

YouTube Webinar

EECS Colloquium

Wednesday, October 28, 2026

Gateway 1220
4:00 – 5:00 pm

Saman Amarasinghe

Thomas and Gerd Perkins Professor of EECS, MIT

Bio

Saman Amarasinghe is the Thomas and Gerd Perkins Professor in the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL), where he leads the Commit Compiler Research Group. Under his leadership, the Commit group has developed a wide range of innovative programming languages and compilers, including StreamIt, StreamJIT, PetaBricks, Halide, TACO, Finch, SySTeC, GraphIt, Simit, MILK, Cimple, BioStream, NetBlocks, BREeze, CoLa, Shim, AskIt, and Seq. Additionally, the group has created compiler and runtime frameworks such as DynamoRIO, Helium, Tiramisu, Codon, BuildIt, and D2X, as well as vectorization techniques and tools such as Superword Level Parallelism (SLP), goSLP, and VeGen. Saman’s team also developed Ithemal, a machine-learning-based performance predictor; Program Shepherding, which protects programs from external attacks; the extensible OpenTuner autotuning framework; and Kendo, a deterministic execution system. He co-led the Raw architecture project. Outside academia, Saman has founded or co-founded several companies, including Determina, Lanka Internet Services Ltd., Venti Technologies, DataCebo, Exaloop, and Build4Sim. He earned his BS in Electrical Engineering and Computer Science from Cornell University in 1988, and his MSEE and Ph.D. from Stanford University in 1990 and 1997, respectively. He is a Fellow of the ACM and a recipient of the ACM-IEEE CS Ken Kennedy Award.

Abstract

The genius of FORTRAN, introduced in 1957, was not just that it was the first widely successful high-level language, but that it made hardware disappear. The same FORTRAN program could run on an IBM System/360, a DEC PDP-8, a CDC 6600, or even a Cray-1, with enough performance that most programmers never had to touch assembly or write architecture-specific code.

Over time, as both hardware and software grew more complex, compilers quietly lost much of their ability to maintain this abstraction. Multicores, complex vector instructions, and specialized accelerators have pushed more of the performance burden back onto the programmer. Today, we have, in many ways, come full circle. For example, getting peak performance from NVIDIA’s V100, A100, H100, and B100 generations often requires different, architecture-specific CUDA kernels and, on the latest parts, even hand-written PTX assembly to fully exploit tensor cores and the Tensor Memory Accelerator. To the best of our knowledge, the Apple/Arm SME accelerator does not even have a high-level compiler, and most high-performance code must be written using intrinsics or assembly.

The recent AI revolution is bringing capabilities that are poised to upend the traditional compiler. Machine learning can give existing compilers capabilities they have never had, from making complex optimization decisions to generating substantial portions of compiler passes. At the extreme, LLMs may replace much of the traditional compiler as we know it.

This raises a fundamental question: can the next generation of compilers finally fulfill the promise that began with FORTRAN? Can they transform a single, high-level description of an algorithm into light-speed implementations across CPUs, GPUs, and specialized accelerators, while once again making the underlying hardware disappear?

In this talk, I will hallucinate about this exciting possibility, drawing lessons from the history of compiler research and exploring possible paths toward building Compiler 2.0.