ninetoothed.make

ninetoothed.make(arrangement, application, tensors, caller='torch', kernel_name=None, output_dir=None, num_warps=None, num_stages=None, max_num_configs=None)

Integrate the arrangement and the application of the tensors.

Parameters:

• arrangement – The arrangement of the tensors.

• application – The application of the tensors.

• tensors – The tensors.

• caller – Who will call the compute kernel.

• kernel_name – The name for the generated kernel.

• output_dir – The directory to store the generated files.

• num_warps – The number of warps to use.

• num_stages – The number of pipeline stages.

• max_num_configs – The maximum number of auto-tuning configurations to use.

Returns:

A handle to the compute kernel.