Recent years have seen a proliferation of specialized AI accelerators – proposed in both academia (e.g., Gemmini, FEATHER, EVA) and industry (e.g., Google TPU, Intel AMX, AWS Trainium) – that depart significantly from traditional CPU/GPU architectures. However, research on compiler and systems support for these accelerators remains sparse, largely due to the lack of mature open-source ML compiler infrastructures capable of targeting them from popular ML frameworks like PyTorch and JAX. Building such support involves considerable manual effort, slowing innovation and creating a gap between hardware and software research communities.

This tutorial introduces the ACT (Accelerator Compiler Toolkit), an ecosystem that automatically generates complete ML compiler backends and essential software tooling from high-level ISA specifications of AI accelerators.

The ACT ecosystem consists of:

• TAIDL (Tensor Accelerator ISA Definition Language): A Python-based DSL for specifying AI accelerator ISAs
• TAIDL-TO (Test Oracle) Generator: Automatically generates fast & scalable functional simulators from TAIDL specifications
• ACT Backend Generator: Automatically generates sound & complete ML compiler backends just from TAIDL specifications
• XLA Integration: Enables end-to-end compilation from popular ML frameworks like JAX, TensorFlow, and PyTorch

The tutorial is designed for researchers, practitioners, and students interested in compiler design, programming languages, and AI/ML hardware. By the end, participants will have hands-on experience with the complete ACT workflow and understand how to rapidly prototype ML compiler support for novel AI accelerator architectures.

The outline can be found on the tutorial webpage.

Note that the tutorial is on the second floor in the Beak Peak room