MLIR Dialect Usage Estimates

This page estimates the usage of MLIR dialects as lowering targets for the following repositories:

• mlir
• iree
• tensorflow
• xla
• stablehlo
• torch-mlir
• triton
• mlir-aie
• onnx-mlir
• Polygeist
• circt
• Accera

Usage is estimated by counting the number of matches after // CHECK: for each dialect operation in the repository tests. This is based on the assumption that each compiler project lowers to various dialects, that a more important dialect is mentioned more often in test files, and that the downstream repositories use FileCheck. Specifically, the following ripgrep rg command is used:

rg regex -g "*.mlir" repo_dir

where regex is the following regular expression:

[ ]*//[ ]*[^:]+: [ ]*[^%]*%[\[\]\.\+:a-zA-Z0-9_]+ = "?[^.]*\.

which, for example, matches the following lines:

// CHECK: %0 = arith.constant 1 : index

  //      FULL-UNROLL:    %cst = arith.constant 1 : index

 // CHECK: %[[VAR:.+]] = arith.constant 1 : index

The source code for this page is available at https://github.com/rikhuijzer/mlir-dialects.