Print mps

print_mps

MPSTarget dataclass

Bases: Target

Source code in xdsl/backend/mps/print_mps.py

@dataclass(frozen=True)
class MPSTarget(Target):
    name = "mps"

    def emit(self, ctx: Context, module: ModuleOp, output: IO[str]) -> None:
        print_to_mps(module, output)

name = 'mps' class-attribute instance-attribute

__init__() -> None

emit(ctx: Context, module: ModuleOp, output: IO[str]) -> None

Source code in xdsl/backend/mps/print_mps.py

def emit(self, ctx: Context, module: ModuleOp, output: IO[str]) -> None:
    print_to_mps(module, output)

print_to_mps(prog: ModuleOp, output: IO[str]) -> None

Lower an MLIR module to the proprietary Air Intermediate Representation (AIR).

This function emits the AIR bitcode required to target Apple GPUs directly, bypassing the Metal Shading Language source generation step.

Parameters:

Name	Type	Description	Default
prog	ModuleOp	The MLIR ModuleOp to be lowered.	required
output	IO[str]	The IO stream (e.g., stdout or a file) to write the generated bitcode to.	required

Returns:

Type	Description
None	None. The output is written directly to the provided stream.

Source code in xdsl/backend/mps/print_mps.py

def print_to_mps(prog: ModuleOp, output: IO[str]) -> None:
    """
    Lower an MLIR module to the proprietary Air Intermediate Representation (AIR).

    This function emits the AIR bitcode required to target Apple GPUs directly,
    bypassing the Metal Shading Language source generation step.

    Args:
        prog: The MLIR ModuleOp to be lowered.
        output: The IO stream (e.g., stdout or a file) to write the generated bitcode to.

    Returns:
        None. The output is written directly to the provided stream.
    """
    raise NotImplementedError("MPS backend not yet implemented")