Method path_matching

ppq/IR/search.py:506–562

path_matching is used for path searching on the graph, and complete paths will be returned. Note that it's possible for results to overflow when there are numerous matchings. path_matching 用于图检索，匹配完整路径 一个最简单的例子： sp_expr = lamdba x: x.type == 'Conv

(
        self,
        sp_expr: Callable,
        rp_expr: Callable,
        ep_expr: Callable,
        direction: str
        )

        return ret_collection

    def path_matching(
        self,
        sp_expr: Callable,
        rp_expr: Callable,
        ep_expr: Callable,
        direction: str
        ) -> List[Path]:
        """path_matching is used for path searching on the graph, and complete
        paths will be returned. Note that it's possible for results to overflow
        when there are numerous matchings.

        path_matching 用于图检索，匹配完整路径
            一个最简单的例子：
                sp_expr = lamdba x: x.type == 'Conv'
                rp_expr = lamdba x, y: y.type == 'Relu'
                ep_expr = lamdba x: x.type == 'Conv'
                limitation = None
                direction = 'down'
            该指令检索出从任意 Conv 出发，到任意 Conv 的所有可行路径
            其中路径上含有任意多个 Relu 节点，并且只能包含 Relu

            你需要注意到 rp_expr 是一个二元表达式，其中 x 代表起点，y 代表终点
        Args:
            sp_expr (Union[Pattern, Callable]):
                start point expression, 用于匹配检索起点的表达式
            rp_expr (Union[Pattern, Callable]):
                relay point expression, 用于匹配检索中继点的表达式
            ep_expr (Union[Pattern, Callable]):
                end point expression, 用于匹配检索终点的表达式
            limitation (Union[Limitation, Callable]):
                用于过滤路径的表达式
            direction (str, optional):
                图检索方向，search direction: up, down. Defaults to 'down'.
            greedy (bool, optional):
                是否执行贪心匹配，设置为True则尝试匹配直到最后一个end point.
                whether to search greedily
        """
        if direction not in {'up', 'down'}:
            raise KeyError("traversal_direction must be one of {'up', 'down'}")

        _ret_collection = []
        for operation in self.graph.operations.values():

            # filter out operation which can not be a start point.
            if not sp_expr(operation): continue

            # match patterns, add patterns towards result collection.
            matchings = self._path_matching(
                start_point=operation, rp_expr=rp_expr,
                ep_expr=ep_expr, direction=direction)
            _ret_collection.extend(matchings)

        # clear cache
        self._cache.clear()

        # use limitation to filter out invalid path.
        return _ret_collection