Towards a high-performance AI compiler with upstream MLIR

TL;DR

This paper introduces a compilation flow using open-source passes to optimize high-level linear algebra IR for high-performance AI applications, achieving near hand-optimized performance.

Contribution

It presents a novel compilation pipeline with cache-aware tensor distribution, shape propagation, and micro-kernel lowering for efficient AI model execution.

Findings

Achieves over 90% performance of hand-optimized code

Supports TensorFlow and PyTorch IR inputs

Includes cache-aware tensor distribution and micro-kernel lowering

Abstract

This work proposes a compilation flow using open-source compiler passes to build a framework to achieve ninja performance from a generic linear algebra high-level abstraction. We demonstrate this flow with a proof-of-concept MLIR project that uses input IR in Linalg-on-Tensor from TensorFlow and PyTorch, performs cache-level optimizations and lowering to micro-kernels for efficient vectorization, achieving over 90% of the performance of ninja-written equivalent programs. The contributions of this work include: (1) Packing primitives on the tensor dialect and passes for cache-aware distribution of tensors (single and multi-core) and type-aware instructions (VNNI, BFDOT, BFMMLA), including propagation of shapes across the entire function; (2) A linear algebra pipeline, including tile, fuse and bufferization strategies to get model-level IR into hardware friendly tile calls; (3) A mechanism for micro-kernel lowering to an open source library that supports various CPUs.