tinygrad runtime.ops_cpu

Note

You likely want the upstream tinygrad, not tinygrab. Tinygrab contains AI generated docstrings for a tinygrad snapshot. Upstream: https://tinygrad.org

class tinygrad.runtime.ops_cpu.NumpyAllocator[source]

Bases: Allocator

Allocator class for numpy arrays.

_alloc[source]

Allocates memory for a given size and dtype.

Type:

method

as_buffer[source]

Converts an np.ndarray to a memoryview object.

Type:

method

copyin[source]

Copies data from a memoryview object to an np.ndarray.

Type:

method

copyout[source]

Copies data from an np.ndarray to a memoryview object.

Type:

method

as_buffer(src: ndarray) → memoryview[source]

Converts an np.ndarray to a memoryview object.

Parameters:

src (np.ndarray) – The numpy array to be converted.

Returns:

A view into the original numpy array’s data.

Return type:

memoryview

copyin(dest: ndarray, src: memoryview)[source]

Copies data from a memoryview object to an np.ndarray.

Parameters:

  • dest (np.ndarray) – The destination numpy array.

  • src (memoryview) – The source memoryview object.

copyout(dest: memoryview, src: ndarray)[source]

Copies data from an np.ndarray to a memoryview object.

Parameters:

  • dest (memoryview) – The destination memoryview object.

  • src (np.ndarray) – The source numpy array.

tinygrad.runtime.ops_cpu.einsum_mulacc(einsum, get_strides, expand)[source]

This function returns a higher-order function named mulacc. The returned function performs multiplication and accumulation using the numpy.einsum function. It takes two arrays as input along with their shapes and strides.

tinygrad.runtime.ops_cpu.einsum

A function that performs numpy einsum operation.

Type:

function

tinygrad.runtime.ops_cpu.get_strides

Function to calculate the strides of an array.

Type:

function

tinygrad.runtime.ops_cpu.expand

Function to perform broadcasting/expansion of arrays.

Type:

function

tinygrad.runtime.ops_cpu.match_types(x, y)[source]

Cast two numpy arrays to a common datatype determined by output_type() function and return the casted arrays.

Parameters:

  • x (numpy.ndarray) – The first numpy array for casting.

  • y (numpy.ndarray) – The second numpy array for casting.

Returns:

A tuple of the casted x and y arrays.

Return type:

Tuple[numpy.ndarray, numpy.ndarray]

tinygrad.runtime.ops_cpu.output_type(x, y)[source]

Determine the datatype with higher priority between two numpy arrays.

Parameters:

  • x (numpy.ndarray) – The first numpy array to compare.

  • y (numpy.ndarray) – The second numpy array to compare.

Returns:

The datatype with higher priority between x and y.

Return type:

numpy.dtype

tinygrad.runtime.ops_cpu.shape_to_axis(old_shape: Tuple[int, ...], new_shape: Tuple[int, ...]) → Tuple[int, ...][source]

Compare two shapes and return a tuple containing the indices of axes that differ between the two shapes.

Parameters:

  • old_shape (Tuple[int, ...]) – The original shape to be compared.

  • new_shape (Tuple[int, ...]) – The new shape to compare against the original shape.

Returns:

A tuple of indices where the axes differ between the two shapes.

Return type:

Tuple[int, …]

Raises:

AssertionError – If the dimensions of old_shape and new_shape are not equal.