YouTip

## PyTorch `torch.bitwise_or` Reference The `torch.bitwise_or` function in PyTorch computes the element-wise bitwise OR of two tensors. It is a highly efficient operation commonly used for low-level binary data manipulation, masking, and boolean logic operations. --- ## Introduction `torch.bitwise_or` performs a bitwise OR operation on each pair of corresponding elements in the input tensors. * **For Integer Tensors:** It compares the binary representation of each element. The resulting bit is set to `1` if at least one of the corresponding bits of the operands is `1`. Otherwise, it is set to `0`. * **For Boolean Tensors:** It computes the logical OR operation (equivalent to `+` or `logical_or` behavior). --- ## Syntax and Parameters ### Syntax ```python torch.bitwise_or(input, other, *, out=None) -> Tensor ``` ### Parameters | Parameter | Type | Description | | :--- | :--- | :--- | | **`input`** | `Tensor` or `Scalar` | The first input tensor or a scalar value. | | **`other`** | `Tensor` or `Scalar` | The second input tensor or a scalar value. | | **`out`** | `Tensor` (Optional) | The output tensor where the result will be written. | ### Supported Data Types * **Integer types:** `torch.int8`, `torch.uint8`, `torch.int16`, `torch.int32`, `torch.int64` * **Boolean type:** `torch.bool` --- ## Code Examples ### Example 1: Bitwise OR with Integer Tensors This example demonstrates how the bitwise OR operation works on 8-bit integers. ```python import torch # Define two 8-bit integer tensors # 5 in binary: 00000101 # 3 in binary: 00000011 # 7 in binary: 00000111 x = torch.tensor([5, 3, 7], dtype=torch.int8) # 1 in binary: 00000001 # 3 in binary: 00000011 # 4 in binary: 00000100 y = torch.tensor([1, 3, 4], dtype=torch.int8) # Perform element-wise bitwise OR: # 5 | 1 -> 00000101 | 00000001 = 00000101 (5) # 3 | 3 -> 00000011 | 00000011 = 00000011 (3) # 7 | 4 -> 00000111 | 00000100 = 00000111 (7) result = torch.bitwise_or(x, y) print("Result Tensor:") print(result) # Output: tensor([5, 3, 7], dtype=int8) ``` ### Example 2: Logical OR with Boolean Tensors When applied to boolean tensors, `torch.bitwise_or` behaves as a standard logical OR operation. ```python import torch # Define two boolean tensors a = torch.tensor([True, False, True, False], dtype=torch.bool) b = torch.tensor([True, True, False, False], dtype=torch.bool) # Perform logical OR result = torch.bitwise_or(a, b) print("Boolean OR Result:") print(result) # Output: tensor([ True, True, True, False]) ``` ### Example 3: Tensor and Scalar Bitwise OR You can also perform a bitwise OR between a tensor and a scalar value. PyTorch will automatically broadcast the scalar to match the shape of the tensor. ```python import torch # Define a tensor x = torch.tensor([1, 2, 4], dtype=torch.int32) # Perform bitwise OR with a scalar value of 2 # 1 | 2 -> 01 | 10 = 11 (3) # 2 | 2 -> 10 | 10 = 10 (2) # 4 | 2 -> 100 | 010 = 110 (6) result = torch.bitwise_or(x, 2) print("Scalar Bitwise OR Result:") print(result) # Output: tensor([3, 2, 6], dtype=int32) ``` --- ## Important Considerations 1. **Broadcasting Support:** `torch.bitwise_or` supports PyTorch's broadcasting semantics. If the shapes of `input` and `other` do not match but are compatible, they will be broadcast to a common shape. 2. **In-place Operation:** If you want to perform the operation in-place and modify the original tensor, you can use `input.bitwise_or_(other)` or the operator shorthand `input |= other`. 3. **Floating-Point Types:** Bitwise operations are not defined for floating-point types (e.g., `torch.float32`, `torch.float64`). Attempting to use them will result in a `RuntimeError`.