- class qiskit.dagcircuit.BlockCollector(dag)#
This class implements various strategies of dividing a DAG (direct acyclic graph) into blocks of nodes that satisfy certain criteria. It works both with the
DAGDependencyrepresentations of a DAG, where DagDependency takes into account commutativity between nodes.
Collecting nodes from DAGDependency generally leads to more optimal results, but is slower, as it requires to construct a DAGDependency beforehand. Thus, DAGCircuit should be used with lower transpiler settings, and DAGDependency should be used with higher transpiler settings.
In general, there are multiple ways to collect maximal blocks. The approaches used here are of the form ‘starting from the input nodes of a DAG, greedily collect the largest block of nodes that match certain criteria’. For additional details, see https://github.com/Qiskit/qiskit-terra/issues/5775.
DAGCircuitError – the input object is not a DAG.
- collect_all_matching_blocks(filter_fn, split_blocks=True, min_block_size=2, split_layers=False, collect_from_back=False)#
Collects all blocks that match a given filtering function filter_fn. This iteratively finds the largest block that does not match filter_fn, then the largest block that matches filter_fn, and so on, until no more uncollected nodes remain. Intuitively, finding larger blocks of non-matching nodes helps to find larger blocks of matching nodes later on.
After the blocks are collected, they can be optionally refined. The option
split_blocksallows to split collected blocks into sub-blocks over disjoint qubit subsets. The option
split_layersallows to split collected blocks into layers of non-overlapping instructions. The option
min_block_sizespecifies the minimum number of gates in the block for the block to be collected.
By default, blocks are collected in the direction from the inputs towards the outputs of the circuit. The option
collect_from_backallows to change this direction, that is collect blocks from the outputs towards the inputs of the circuit.
Returns the list of matching blocks only.
Iteratively collects the largest block of input nodes (that is, nodes with
_in_degreeequal to 0) that match a given filtering function. Examples of this include collecting blocks of swap gates, blocks of linear gates (CXs and SWAPs), blocks of Clifford gates, blocks of single-qubit gates, blocks of two-qubit gates, etc. Here ‘iteratively’ means that once a node is collected, the
_in_degreeof each of its immediate successor is decreased by 1, allowing more nodes to become input and to be eligible for collecting into the current block. Returns the block of collected nodes.