4.6.8. pde.tools.numba module¶
Helper functions for just-in-time compilation with numba
- class Counter(value: int = 0)¶
helper class for implementing JIT_COUNT
We cannot use a simple integer for this, since integers are immutable, so if one imports JIT_COUNT from this module it would always stay at the fixed value it had when it was first imported. The workaround would be to import the symbol every time the counter is read, but this is error-prone. Instead, we implement a thin wrapper class around an int, which only supports reading and incrementing the value. Since this object is now mutable it can be used easily. A disadvantage is that the object needs to be converted to int before it can be used in most expressions.
helper function that turns 0d-arrays into scalars
This helps to avoid the bug discussed in https://github.com/numba/numba/issues/6000
- flat_idx(arr, i)¶
helper function allowing indexing of scalars as if they arrays
returns a numba numerical type in which all arrays can be represented
*args – All items to be tested
Returns: numba.complex128 if any entry is complex, otherwise numba.double
- jit(function: TFunc, signature=None, parallel: bool = False, **kwargs) TFunc ¶
apply nb.jit with predefined arguments
function – The function which is jitted
signature – Signature of the function to compile
parallel (bool) – Allow parallel compilation of the function
**kwargs – Additional arguments to nb.jit
Function that will be compiled using numba
- jit_allocate_out(func: Callable, parallel: bool = False, out_shape: Optional[Tuple[int, ...]] = None, num_args: int = 1, **kwargs) Callable ¶
Decorator that compiles a function with allocating an output array.
This decorator compiles a function that takes the arguments arr and out. The point of this decorator is to make the out array optional by supplying an empty array of the same shape as arr if necessary. This is implemented efficiently by using
func – The function to be compiled
parallel (bool) – Determines whether the function is jitted with parallel=True.
out_shape (tuple) – Determines the shape of the out array. If omitted, the same shape as the input array is used.
num_args (int, optional) – Determines the number of input arguments of the function.
**kwargs – Additional arguments used in
The decorated function
- make_array_constructor(arr: ndarray) Callable[, ndarray] ¶
returns an array within a jitted function using basic information
ndarray) – The array that should be accessible within jit
A reference to the array needs to be retained outside the numba code to prevent garbage collection from removing the array
- numba_dict(data: Optional[Dict[str, Any]] = None) Optional[Dict] ¶
converts a python dictionary to a numba typed dictionary