from enum import Enum
from typing_extensions import Protocol, TypedDict
from typing import Union
from sparseconverter import (
CUDA, CUPY, CUPY_SCIPY_COO, CUPY_SCIPY_CSC, CUPY_SCIPY_CSR, NUMPY,
SCIPY_COO, SCIPY_CSC, SCIPY_CSR, SPARSE_COO, SPARSE_DOK, SPARSE_GCXS,
CPU_BACKENDS, CUDA_BACKENDS, CUPY_BACKENDS, SPARSE_BACKENDS, DENSE_BACKENDS,
ND_BACKENDS, D2_BACKENDS,
)
import numpy as np
# markers for special values:
class TileDepthEnum(Enum):
TILE_DEPTH_DEFAULT = object()
class TileSizeEnum(Enum):
TILE_SIZE_BEST_FIT = object()
class TilingPreferences(TypedDict):
depth: Union[int, TileDepthEnum]
total_size: Union[float, int]
[docs]
class UDFMethod(Enum):
TILE = 'tile'
FRAME = 'frame'
PARTITION = 'partition'
class UDFProtocol(Protocol):
'''
Parts of the UDF interface required for MIT code in LiberTEM
'''
USE_NATIVE_DTYPE = bool #: Neutral element for type conversion
TILE_SIZE_BEST_FIT = TileSizeEnum.TILE_SIZE_BEST_FIT #: Suggest using recommended tile size
TILE_SIZE_MAX = np.inf #: Suggest using maximum tile size
TILE_DEPTH_DEFAULT = TileDepthEnum.TILE_DEPTH_DEFAULT #: Suggest using recommended tile depth
TILE_DEPTH_MAX = np.inf #: Suggest using maximum tile depth
BACKEND_NUMPY = NUMPY #: NumPy array
BACKEND_CUPY = CUPY #: CuPy array
BACKEND_CUDA = CUDA #: NumPy array, but run on CUDA device class
BACKEND_SPARSE_COO = SPARSE_COO #: sparse.COO array
BACKEND_SPARSE_GCXS = SPARSE_GCXS #: sparse.GCXS array
BACKEND_SPARSE_DOK = SPARSE_DOK #: sparse.DOK array -- not recommended since slow!
BACKEND_SCIPY_COO = SCIPY_COO #: scipy.sparse.coo_matrix
BACKEND_SCIPY_CSR = SCIPY_CSR #: scipy.sparse.csr_matrix
BACKEND_SCIPY_CSC = SCIPY_CSC #: scipy.sparse.csc_matrix
BACKEND_CUPY_SCIPY_COO = CUPY_SCIPY_COO #: cupyx.scipy.sparse.coo_matrix
BACKEND_CUPY_SCIPY_CSR = CUPY_SCIPY_CSR #: cupyx.scipy.sparse.csr_matrix
BACKEND_CUPY_SCIPY_CSC = CUPY_SCIPY_CSC #: cupyx.scipy.sparse.csc_matrix
# Excludes sparse.DOK, numpy.matrix and CUDA, prefers scipy.sparse and GPU
# Deprioritizes sparse.pydata.org due to their high call overhead
#: Tuple with all backends in suggested priority
BACKEND_ALL = (
CUPY_SCIPY_CSR, CUPY_SCIPY_CSC, CUPY_SCIPY_COO,
SCIPY_CSR, SCIPY_CSC, SCIPY_COO,
CUPY, NUMPY,
SPARSE_COO, SPARSE_GCXS,
)
CPU_BACKENDS = CPU_BACKENDS #: Set of backends that run on device class CPU
CUDA_BACKENDS = CUDA_BACKENDS #: Set of backends that run on device class CUDA
CUPY_BACKENDS = CUPY_BACKENDS #: Set of backends that use CuPy, subset of class CUDA
SPARSE_BACKENDS = SPARSE_BACKENDS #: Set of backends that are sparse arrays
DENSE_BACKENDS = DENSE_BACKENDS #: Set of backends that are dense arrays
ND_BACKENDS = ND_BACKENDS #: Set of backends that support n-dimensional arrays
D2_BACKENDS = D2_BACKENDS #: Set of backends that only support two-dimensional arrays
UDF_METHOD = UDFMethod #: Enum of process_ methods accepted by the UDF interface
def get_method() -> UDFMethod:
raise NotImplementedError()
def get_tiling_preferences() -> TilingPreferences:
raise NotImplementedError()