Class SharedObject

exir/memory_planning.py:703–726

r""" We define the concept of shared object, which represents a segment in the memory buffer that can be shared by multiple tensors. In order to check if a shared object is available for a tensor, we maintain the last_used_index attribute. The shared object will be available for node

@dataclass
class SharedObject:
    r"""
    We define the concept of shared object, which represents a segment
    in the memory buffer that can be shared by multiple tensors. In order to
    check if a shared object is available for a tensor, we maintain the
    last_used_index attribute. The shared object will be available for nodes
    with index greater than last_used_index.
    """

    # index of the shared object in the list of shared objects, used as a unique id
    idx: int
    # offset in the memory buffer
    offset: int
    # size of this shared object in bytes
    size: int
    # When the object is first created
    first_used_index: int
    # the object will be available for index (last_used_index + 1)
    last_used_index: int
    # list of allocations belong to this shared object
    allocations: List[AllocationSpec] = field(default_factory=list)

    def __repr__(self) -> str:
        return f"SharedObject(idx={self.idx}, offset={self.offset}, size={self.size}, lifetime=[{self.first_used_index, self.last_used_index}])"


@dataclass