⚠️ 🚧 TODO / WIP

Introduction

This page contains a short tutorial on MLIR, focusing on what is needed to understand cairo-native’s codebase and usage of MLIR, and a collection of resources.

Useful Notes

LLVM alloca notes

It's better to use opaque pointers for the returned alloca data instead of specifying the full pointer type. Also LLVM is switching to those and deprecating non-opaque pointers.

Creating a constant LLVM array

To create a constant array, you need to use the llvm.mlir.constant operation. The way to represent the array data is by using a dense element attribute, for example:

// The attribute should look like this:
dense<[3, 1, 4, 1, 5]> : tensor<5 x i8>

// Full example
%alloca_size = llvm.mlir.constant(5 : i64) : i64
%array_ptr = llvm.alloca %alloca_size x i8 : (i64) -> !llvm.ptr
%array_data = llvm.mlir.constant(dense<[37, 48, 56, 88, 0]> : tensor<5xi8>) : !llvm.array<5 x i8>
llvm.store %array_data, %array_ptr : !llvm.array<5 x i8>, !llvm.ptr