# MLIR

Multi-Level IR Compiler Framework

# Multi-Level Intermediate Representation Overview

The MLIR project is a novel approach to building reusable and extensible compiler infrastructure. MLIR aims to address software fragmentation, improve compilation for heterogeneous hardware, significantly reduce the cost of building domain specific compilers, and aid in connecting existing compilers together.

## More resources

We host a weekly public meeting about MLIR and the ecosystem. If you’d like to discuss a particular topic or have questions, please add it to the agenda doc . Details on how to join the meeting are in the agenda doc, you can get a Google calendar invite by joining this googlegroup . The meetings are recorded and published in the talks section.

## What is MLIR for?

MLIR is intended to be a hybrid IR which can support multiple different requirements in a unified infrastructure. For example, this includes:

• The ability to represent dataflow graphs (such as in TensorFlow), including dynamic shapes, the user-extensible op ecosystem, TensorFlow variables, etc.
• Optimizations and transformations typically done on such graphs (e.g. in Grappler).
• Ability to host high-performance-computing-style loop optimizations across kernels (fusion, loop interchange, tiling, etc.), and to transform memory layouts of data.
• Code generation “lowering” transformations such as DMA insertion, explicit cache management, memory tiling, and vectorization for 1D and 2D register architectures.
• Ability to represent target-specific operations, e.g. accelerator-specific high-level operations.
• Quantization and other graph transformations done on a Deep-Learning graph.
• Polyhedral primitives .
• Hardware Synthesis Tools / HLS .

MLIR is a common IR that also supports hardware specific operations. Thus, any investment into the infrastructure surrounding MLIR (e.g. the compiler passes that work on it) should yield good returns; many targets can use that infrastructure and will benefit from it.

MLIR is a powerful representation, but it also has non-goals. We do not try to support low level machine code generation algorithms (like register allocation and instruction scheduling). They are a better fit for lower level optimizers (such as LLVM). Also, we do not intend MLIR to be a source language that end-users would themselves write kernels in (analogous to CUDA C++). On the other hand, MLIR provides the backbone for representing any such DSL and integrating it in the ecosystem.

## Compiler infrastructure

We benefited from experience gained from building other IRs (LLVM IR, XLA HLO, and Swift SIL) when building MLIR. The MLIR framework encourages existing best practices, e.g. writing and maintaining an IR spec, building an IR verifier, providing the ability to dump and parse MLIR files to text, writing extensive unit tests with the FileCheck tool, and building the infrastructure as a set of modular libraries that can be combined in new ways.

Other lessons have been incorporated and integrated into the design in subtle ways. For example, LLVM has non-obvious design mistakes that prevent a multithreaded compiler from working on multiple functions in an LLVM module at the same time. MLIR solves these problems by having limited SSA scope to reduce the use-def chains and by replacing cross-function references with explicit symbol reference .

## Citing MLIR

Please see the FAQ entry on how to cite MLIR in publications.