Beaver.Composer (beaver v0.4.2)

This module provide functions to compose and run passes.

Summary

Types

composer()
operation()
run_option()
run_result()
t()

Functions

append(composer_or_op, pass)
init(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])
nested(composer_or_op, op_name, passes)
new(composer)
run(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])

Run the passes on the operation.

run!(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])
to_pipeline(composer)

Types

composer()

@type composer() :: t()

operation()

@type operation() :: Beaver.MLIR.Module.t() | Beaver.MLIR.Operation.t()

run_option()

@type run_option() ::
  {:debug, boolean()}
  | {:print, boolean()}
  | {:timing, boolean()}
  | {:verifier, boolean()}

run_result()

@type run_result() :: {:ok, any()} | {:error, String.t()}

t()

@type t() :: %Beaver.Composer{ctx: term(), op: operation(), passes: [any()]}

Functions

append(composer_or_op, pass)

init(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])

nested(composer_or_op, op_name, passes)

new(composer)

run(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])

@spec run(composer(), [run_option()]) :: run_result()

Run the passes on the operation.

Must be a multi-threaded context if an Elixir pass is in the pipeline

MLIR context's thread pool is used to run the CAPI. If an Elixir pass is in the pipeline, the context must be multi-threaded otherwise there can be a deadlock. Also note that it can be more expensive than a C/C++ implementation due to the overhead of the thread pool.

run!(composer, opts \\ [debug: false, print: false, timing: false, verifier: true])

@spec run!(composer(), [run_option()]) :: operation()

to_pipeline(composer)