# LICENSE HEADER MANAGED BY add-license-header # # Copyright 2018 Kornia Team # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # import importlib.util import platform import sys import warnings from dataclasses import asdict, fields, is_dataclass from functools import wraps from inspect import isclass, isfunction from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Tuple, Type, TypeVar, Union, overload import torch from torch.linalg import inv_ex from kornia.core import Tensor from kornia.utils._compat import torch_version_ge def xla_is_available() -> bool: """Return whether `torch_xla` is available in the system.""" if importlib.util.find_spec("torch_xla") is not None: return True return False def is_mps_tensor_safe(x: Tensor) -> bool: """Return whether tensor is on MPS device.""" return "mps" in str(x.device) def get_cuda_device_if_available(index: int = 0) -> torch.device: """Try to get cuda device, if fail, return cpu. Args: index: cuda device index Returns: torch.device """ if torch.cuda.is_available(): return torch.device(f"cuda:{index}") return torch.device("cpu") def get_mps_device_if_available() -> torch.device: """Try to get mps device, if fail, return cpu. Returns: torch.device """ dev = "cpu" if hasattr(torch.backends, "mps"): if torch.backends.mps.is_available(): dev = "mps" return torch.device(dev) def get_cuda_or_mps_device_if_available() -> torch.device: """Check OS and platform and run get_cuda_device_if_available or get_mps_device_if_available. Returns: torch.device """ if sys.platform == "darwin" and platform.machine() == "arm64": return get_mps_device_if_available() else: return get_cuda_device_if_available() @overload def map_location_to_cpu(storage: Tensor, location: str) -> Tensor: ... @overload def map_location_to_cpu(storage: str) -> str: ... def map_location_to_cpu(storage: Union[str, Tensor], *args: Any, **kwargs: Any) -> Union[str, Tensor]: """Map location of device to CPU, util for loading things from HUB.""" return storage def deprecated( replace_with: Optional[str] = None, version: Optional[str] = None, extra_reason: Optional[str] = None ) -> Any: """Mark methods as deprecated.""" def _deprecated(func: Callable[..., Any]) -> Any: @wraps(func) def wrapper(*args: Any, **kwargs: Any) -> Any: name = "" beginning = f"Since kornia {version} the " if version is not None else "" if isclass(func): name = func.__class__.__name__ if isfunction(func): name = func.__name__ warnings.simplefilter("always", DeprecationWarning) if replace_with is not None: warnings.warn( f"{beginning}`{name}` is deprecated in favor of `{replace_with}`.{extra_reason}", category=DeprecationWarning, stacklevel=2, ) else: warnings.warn( f"{beginning}`{name}` is deprecated and will be removed in the future versions.{extra_reason}", category=DeprecationWarning, stacklevel=2, ) warnings.simplefilter("default", DeprecationWarning) return func(*args, **kwargs) return wrapper return _deprecated def _extract_device_dtype(tensor_list: List[Optional[Any]]) -> Tuple[torch.device, torch.dtype]: """Check if all the input are in the same device (only if when they are Tensor). If so, it would return a tuple of (device, dtype). Default: (cpu, ``get_default_dtype()``). Returns: [torch.device, torch.dtype] """ device, dtype = None, None for tensor in tensor_list: if tensor is not None: if not isinstance(tensor, (Tensor,)): continue _device = tensor.device _dtype = tensor.dtype if device is None and dtype is None: device = _device dtype = _dtype elif device != _device or dtype != _dtype: raise ValueError( "Passed values are not in the same device and dtype." f"Got ({device}, {dtype}) and ({_device}, {_dtype})." ) if device is None: # TODO: update this when having torch.get_default_device() device = torch.device("cpu") if dtype is None: dtype = torch.get_default_dtype() return (device, dtype) def _torch_inverse_cast(input: Tensor) -> Tensor: """Make torch.inverse work with other than fp32/64. The function torch.inverse is only implemented for fp32/64 which makes impossible to be used by fp16 or others. What this function does, is cast input data type to fp32, apply torch.inverse, and cast back to the input dtype. """ if not isinstance(input, Tensor): raise AssertionError(f"Input must be Tensor. Got: {type(input)}.") dtype: torch.dtype = input.dtype if dtype not in (torch.float32, torch.float64): dtype = torch.float32 return torch.linalg.inv(input.to(dtype)).to(input.dtype) def _torch_histc_cast(input: Tensor, bins: int, min: Union[float, bool], max: Union[float, bool]) -> Tensor: """Make torch.histc work with other than fp32/64. The function torch.histc is only implemented for fp32/64 which makes impossible to be used by fp16 or others. What this function does, is cast input data type to fp32, apply torch.inverse, and cast back to the input dtype. """ if not isinstance(input, Tensor): raise AssertionError(f"Input must be Tensor. Got: {type(input)}.") dtype: torch.dtype = input.dtype if dtype not in (torch.float32, torch.float64): dtype = torch.float32 return torch.histc(input.to(dtype), bins, min, max).to(input.dtype) def _torch_svd_cast(input: Tensor) -> Tuple[Tensor, Tensor, Tensor]: """Make torch.svd work with other than fp32/64. The function torch.svd is only implemented for fp32/64 which makes impossible to be used by fp16 or others. What this function does, is cast input data type to fp32, apply torch.svd, and cast back to the input dtype. NOTE: in torch 1.8.1 this function is recommended to use as torch.linalg.svd """ # if not isinstance(input, torch.Tensor): # raise AssertionError(f"Input must be torch.Tensor. Got: {type(input)}.") dtype = input.dtype if dtype not in (torch.float32, torch.float64): dtype = torch.float32 out1, out2, out3H = torch.linalg.svd(input.to(dtype)) if torch_version_ge(1, 11): out3 = out3H.mH else: out3 = out3H.transpose(-1, -2) return (out1.to(input.dtype), out2.to(input.dtype), out3.to(input.dtype)) def _torch_linalg_svdvals(input: Tensor) -> Tensor: """Make torch.linalg.svdvals work with other than fp32/64. The function torch.svd is only implemented for fp32/64 which makes impossible to be used by fp16 or others. What this function does, is cast input data type to fp32, apply torch.svd, and cast back to the input dtype. NOTE: in torch 1.8.1 this function is recommended to use as torch.linalg.svd """ if not isinstance(input, Tensor): raise AssertionError(f"Input must be Tensor. Got: {type(input)}.") dtype: torch.dtype = input.dtype if dtype not in (torch.float32, torch.float64): dtype = torch.float32 if TYPE_CHECKING: # TODO: remove this branch when kornia relies on torch >= 1.10 out: Tensor elif torch_version_ge(1, 10): out = torch.linalg.svdvals(input.to(dtype)) else: # TODO: remove this branch when kornia relies on torch >= 1.10 _, out, _ = torch.linalg.svd(input.to(dtype)) return out.to(input.dtype) # TODO: return only `Tensor` and review all the calls to adjust def _torch_solve_cast(A: Tensor, B: Tensor) -> Tensor: """Make torch.solve work with other than fp32/64. For stable operation, the input matrices should be cast to fp64, and the output will be cast back to the input dtype. However, fp64 is not yet supported on MPS. """ if is_mps_tensor_safe(A): dtype = torch.float32 else: dtype = torch.float64 out = torch.linalg.solve(A.to(dtype), B.to(dtype)) # cast back to the input dtype return out.to(A.dtype) def safe_solve_with_mask(B: Tensor, A: Tensor) -> Tuple[Tensor, Tensor, Tensor]: r"""Solves the system of equations. Avoids crashing because of singular matrix input and outputs the mask of valid solution. """ if not torch_version_ge(1, 10): sol = _torch_solve_cast(A, B) warnings.warn("PyTorch version < 1.10, solve validness mask maybe not correct", RuntimeWarning, stacklevel=1) return sol, sol, torch.ones(len(A), dtype=torch.bool, device=A.device) # Based on https://github.com/pytorch/pytorch/issues/31546#issuecomment-694135622 if not isinstance(B, Tensor): raise AssertionError(f"B must be Tensor. Got: {type(B)}.") dtype: torch.dtype = B.dtype if dtype not in (torch.float32, torch.float64): dtype = torch.float32 if TYPE_CHECKING: # TODO: remove this branch when kornia relies on torch >= 1.13 A_LU: Tensor pivots: Tensor info: Tensor elif torch_version_ge(1, 13): A_LU, pivots, info = torch.linalg.lu_factor_ex(A.to(dtype)) else: # TODO: remove this branch when kornia relies on torch >= 1.13 A_LU, pivots, info = torch.lu(A.to(dtype), True, get_infos=True) valid_mask: Tensor = info == 0 n_dim_B = len(B.shape) n_dim_A = len(A.shape) if n_dim_A - n_dim_B == 1: B = B.unsqueeze(-1) if TYPE_CHECKING: # TODO: remove this branch when kornia relies on torch >= 1.13 X: Tensor elif torch_version_ge(1, 13): X = torch.linalg.lu_solve(A_LU, pivots, B.to(dtype)) else: # TODO: remove this branch when kornia relies on torch >= 1.13 X = torch.lu_solve(B.to(dtype), A_LU, pivots) return X.to(B.dtype), A_LU.to(A.dtype), valid_mask def safe_inverse_with_mask(A: Tensor) -> Tuple[Tensor, Tensor]: r"""Perform inverse. Avoids crashing because of non-invertable matrix input and outputs the mask of valid solution. """ if not isinstance(A, Tensor): raise AssertionError(f"A must be Tensor. Got: {type(A)}.") dtype_original = A.dtype if dtype_original not in (torch.float32, torch.float64): dtype = torch.float32 else: dtype = dtype_original inverse, info = inv_ex(A.to(dtype)) mask = info == 0 return inverse.to(dtype_original), mask def is_autocast_enabled(both: bool = True) -> bool: """Check if torch autocast is enabled. Args: both: if True will consider autocast region for both types of devices Returns: Return a Bool, will always return False for a torch without support, otherwise will be: if both is True `torch.is_autocast_enabled() or torch.is_autocast_enabled('cpu')`. If both is False will return just `torch.is_autocast_enabled()`. """ if TYPE_CHECKING: # TODO: remove this branch when kornia relies on torch >= 1.10.2 return False if not torch_version_ge(1, 10, 2): return False if both: if torch_version_ge(2, 4): return torch.is_autocast_enabled() or torch.is_autocast_enabled("cpu") else: return torch.is_autocast_enabled() or torch.is_autocast_cpu_enabled() return torch.is_autocast_enabled() def dataclass_to_dict(obj: Any) -> Any: """Recursively convert dataclass instances to dictionaries.""" if is_dataclass(obj) and not isinstance(obj, type): return {key: dataclass_to_dict(value) for key, value in asdict(obj).items()} elif isinstance(obj, (list, tuple)): return type(obj)(dataclass_to_dict(item) for item in obj) elif isinstance(obj, dict): return {key: dataclass_to_dict(value) for key, value in obj.items()} else: return obj T = TypeVar("T") def dict_to_dataclass(dict_obj: Dict[str, Any], dataclass_type: Type[T]) -> T: """Recursively convert dictionaries to dataclass instances.""" if not isinstance(dict_obj, dict): raise TypeError("Input conf must be dict") if not is_dataclass(dataclass_type): raise TypeError("dataclass_type must be a dataclass") field_types: dict[str, Any] = {f.name: f.type for f in fields(dataclass_type)} constructor_args = {} for key, value in dict_obj.items(): if key in field_types and is_dataclass(field_types[key]): constructor_args[key] = dict_to_dataclass(value, field_types[key]) else: constructor_args[key] = value # TODO: remove type ignore when https://github.com/python/mypy/issues/14941 be andressed return dataclass_type(**constructor_args)