MLIR - Intermediate representation language

home edit

MLIR, aka Multi-Level Intermediate Representation, is an intermediate representation language created in 2019 by Chris Lattner.

#230on PLDB

6Years Old

84Repos

Download source code:

git clone https://github.com/tensorflow/mlir

Homepage · Source Code

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.

Tags: intermediate representation language
MLIR is developed on GitHub and has 1,728 stars
There are at least 84 MLIR repos on GitHub
Early development of MLIR happened in Google
LLVM IR and Swift SIL and XLA influenced the design of MLIR
The Google BigQuery Public Dataset GitHub snapshot shows 27 users using MLIR in 30 repos on GitHub
First announcement of MLIR
Read more about MLIR on the web: 1.
1 PLDB concepts link to MLIR: Mojo

Example from the web:

// Syntactically similar to LLVM:
func @testFunction(%arg0: i32) {
  %x = call @thingToCall(%arg0) : (i32) -> i32
  br ^bb1
^bb1:
  %y = addi %x, %x : i32
  return %y : i32
}

Example from Compiler Explorer:

// Example code of an affine reduction.
// MLIR example code may not always work out of the box because the textual MLIR format is not stable.
// The example tries to be compatible with the latest MLIR version, which may not work on previous versions.

func @affine_parallel_with_reductions_i64(%arg0: memref<3x3xi64>, %arg1: memref<3x3xi64>) -> (i64, i64) {
  %0:2 = affine.parallel (%kx, %ky) = (0, 0) to (2, 2) reduce ("addi", "muli") -> (i64, i64) {
            %1 = affine.load %arg0[%kx, %ky] : memref<3x3xi64>
            %2 = affine.load %arg1[%kx, %ky] : memref<3x3xi64>
            %3 = arith.muli %1, %2 : i64
            %4 = arith.addi %1, %2 : i64
            affine.yield %3, %4 : i64, i64
          }
  return %0#0, %0#1 : i64, i64
}

Language features

Feature	Supported	Example	Token
Comments	✓	// A comment
Line Comments	✓	// A comment	//
Semantic Indentation	X