mlir-opt CLI Wrapper¶
mlir-opt is the standard command line tool for applying passes to MLIR and lowering from one dialect
to another.
mlir_graphblas.MlirOptCli is a wrapper around the mlir-opt command line executable.
cli = MlirOptCli(executable=None, options=None)
The executable defaults to mlir-opt, but can be specified to include the full path or a differently
named executable if needed.
options, if provided, must be a list of strings to pass to mlir-opt with every call. These options
are in addition to the passes that will be applied.
Applying Passes¶
The first way to apply passes is by calling
result = cli.apply_passes(input_mlir, list_of_passes)
This will return a string containing the final result of applying the list of passes to the input.
If any errors occur, MlirOptError will be raised. This error contains a .debug_result attribute,
which is explained below.
The second way to apply passes is by calling
result = cli.debug_passes(input_mlir, list_of_passes)
This always returns a DebugResult object.
DebugResult¶
A DebugResult object contains a list of .passes applied (or attempted to apply) and a list of
.stages which resulted, including the original. As a result, there is always one more stage than
pass.
These stages and passes can be inspected manually, but the easiest way to interact with them is through the MLIR Explorer. To open the explorer, call
result.explore()
A new browser tab will appear showing the explorer.