MLIR - AshokBhat/notes GitHub Wiki

About

• Multi-Level Intermediate Representation
• An intermediate representation (IR) system
• Between a language (like C) or library (like TensorFlow) and the compiler backend (like LLVM)
• Allows code reuse between compiler stack of different languages
• Can be used for any language in general.

MLIR Architecture

Source: iq.opengenus.org

TensorFlow connection

• Developed primarily to improve the support of TensorFlow on different backends
• Used as part of TensorFlow Runtime

LLVM connection

• Originated by Chris Lattner while he was at Google
• Part of LLVM family of projects

See Also

• TensorFlow Runtime
• TVM
• LLVM