vllm/lib/python3.10/site-packages/cupy/fft/config.py

from cupy import _util

# expose cache handles to this module
from cupy.fft._cache import get_plan_cache  # NOQA
from cupy.fft._cache import clear_plan_cache  # NOQA
from cupy.fft._cache import get_plan_cache_size  # NOQA
from cupy.fft._cache import set_plan_cache_size  # NOQA
from cupy.fft._cache import get_plan_cache_max_memsize  # NOQA
from cupy.fft._cache import set_plan_cache_max_memsize  # NOQA
from cupy.fft._cache import show_plan_cache_info  # NOQA

# on Linux, expose callback handles to this module
import sys as _sys
if _sys.platform.startswith('linux'):
    from cupy.fft._callback import get_current_callback_manager  # NOQA
    from cupy.fft._callback import set_cufft_callbacks  # NOQA
else:
    def get_current_callback_manager(*args, **kwargs):
        return None

    class set_cufft_callbacks:  # type: ignore
        def __init__(self, *args, **kwargs):
            raise RuntimeError('cuFFT callback is only available on Linux')


enable_nd_planning = True
use_multi_gpus = False
_devices = None


def set_cufft_gpus(gpus):
    '''Set the GPUs to be used in multi-GPU FFT.

    Args:
        gpus (int or list of int): The number of GPUs or a list of GPUs
            to be used. For the former case, the first ``gpus`` GPUs
            will be used.

    .. warning::
        This API is currently experimental and may be changed in the future
        version.

    .. seealso:: `Multiple GPU cuFFT Transforms`_

    .. _Multiple GPU cuFFT Transforms:
        https://docs.nvidia.com/cuda/cufft/index.html#multiple-GPU-cufft-transforms
    '''
    _util.experimental('cupy.fft.config.set_cufft_gpus')
    global _devices

    if isinstance(gpus, int):
        devs = [i for i in range(gpus)]
    elif isinstance(gpus, list):
        devs = gpus
    else:
        raise ValueError("gpus must be an int or a list of int.")
    if len(devs) <= 1:
        raise ValueError("Must use at least 2 GPUs.")

    # make it hashable
    _devices = tuple(devs)