Skip to content

beaver-lodge/manx

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🐱 Manx 🐈

MLIR-Accelerated-Nx. MLIR compiler/backend for the Nx.

What does this library do?

You can think of Manx as IREE implemented in Elixir and unlike IREE with new dedicated runtime Manx uses BEAM as the runtime. Nx's expressions are very close to XLA's MHLO so Manx would borrow a lot of conversion/lowering implementations from XLA and IREE.

Why do we need it?

  • Instead of repurposing compilers built for Python, Manx is about building a Nx compiler in Elixir and tailored for Elixir.
  • With Manx, "Tensor compiler" is no longer a giant black box for Erlang world anymore. A non-python programming language should have its full-stack data/ML solution so that it could be truly maintainable.
  • Tighter integration with BEAM. We can build passes and optimizations for Elixir and BEAM and even generate LLVM instructions to send messages or allocate memory with Erlang's allocator.
  • There is a great gap between the understanding "distributed system" in ML and non-ML applications (MPI vs. fault-tolerance). With Manx we could narrow the gap by implementing a ML compiler with a programming language with strong fault-tolerance capability.

Compared to EXLA

  • EXLA is the Nx backend for XLA.

  • In the short run, Manx's performance won't be on-per with XLA/EXLA's.

  • EXLA's lowering:

Nx |> EXLA |> XLA |> MLIR |> LLVM |> hardware
  • Manx's lowering
Nx |> Manx |> MLIR |> LLVM |> hardware

Installation

If available in Hex, the package can be installed by adding manx to your list of dependencies in mix.exs:

def deps do
  [
    {:manx, "~> 0.1.0"}
  ]
end

Documentation can be generated with ExDoc and published on HexDocs. Once published, the docs can be found at https://hexdocs.pm/manx.

About

MLIR backend for Nx

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages