Convert `BIPMaping` transpiler pass into an external plugin #8662

mtreinish · 2022-09-02T13:02:41Z

What should we add?

The BIPMapping transpiler pass has been in a weird state since it's introduction. It is a standalone transpiler pass that is used for both the layout and routing stage but we never integrated it into transpile because it has an external dependency on cplex which is a proprietary software package that most people don't have access too.. Now that we've added an external plugin interface for transpiler stages (which was added in #8305) we have a better option for how to advertise and use the BipMapping pass. We can create an external package for the pass that contains the code of the transpiler pass and offers a plugin that cleanly integrates into the transpile() function. Basically the workflow for using the bip mapper would change to be pip install qiskit-bip-mapper (or whatever the package name we pick is) and then doing transpile(qc, backend, layout_method="bip_mapping"). This make it easier for users to run and also separates the dependency on cplex into a separate package where we can more cleanly document the extra requirements around installing and/or purchasing a commercial license depending on the size of the circuits.

Steps:

Create a new package for the BipMapping pass
- Add layout plugin for the pass
Deprecate BIPMapping pass in qiskit-terra after release of p

The text was updated successfully, but these errors were encountered:

This commit updates all the layout and routing passes which leverage hardware constraints to have a new optional keyword argument, target, to specify a Target object for modelling the constraints of the target backend. This option will superscede any other specified arguments for backend constraints. For most of these passes the target argument is just internally leveraged to build a CouplingMap (and in 2 cases a BackendProperties) as most of the algorithms don't need more than the global connectivity graph and there is minimal overhead to convert from a target to a CouplingMap. The 2 passes which take backend properties should be refactored to work natively with the target because the conversion adds extra overhead which isn't needed, but in those cases the passes aren't enabled by default (or are slated to be moved out of tree as a plugin, see Qiskit#8662) so the extra overhead isn't of particular concern. This is part of the first step towards making all backend constraints for the transpiler used the Target internally (see Qiskit#9256). Once all passes that take hardware constraints parameters as an input are updated to have a target we can start internally using the Target only for all preset pass manager construction and start the long process of deprecating the legacy interface in these passes. Related to Qiskit#9256

* Make all Layout and Routing passes target aware This commit updates all the layout and routing passes which leverage hardware constraints to have a new optional keyword argument, target, to specify a Target object for modelling the constraints of the target backend. This option will superscede any other specified arguments for backend constraints. For most of these passes the target argument is just internally leveraged to build a CouplingMap (and in 2 cases a BackendProperties) as most of the algorithms don't need more than the global connectivity graph and there is minimal overhead to convert from a target to a CouplingMap. The 2 passes which take backend properties should be refactored to work natively with the target because the conversion adds extra overhead which isn't needed, but in those cases the passes aren't enabled by default (or are slated to be moved out of tree as a plugin, see #8662) so the extra overhead isn't of particular concern. This is part of the first step towards making all backend constraints for the transpiler used the Target internally (see #9256). Once all passes that take hardware constraints parameters as an input are updated to have a target we can start internally using the Target only for all preset pass manager construction and start the long process of deprecating the legacy interface in these passes. Related to #9256 * Update pass manager drawer reference dot files The pass manager drawer tests were comparing the dot output for statically built pass managers to test that the output visualization is generated correctly. However, the new arguments for the taget are changing the visualization (to show the new option) and caused these tests to fail. This commit updates the reference images to include the new argument. * Fix lint * Optimize CheckMap This commit optimizes the logic of the CheckMap pass. Previously we were unecessarily creating a distance matrix from the coupling graph, and with a target unecessarily creating a CouplingMap object. Instead of doing this instead the pass now creates a list of bidirectional edges for adjacent nodes in the coupling graph and then a set lookup is done for each check instead of checking for a distance of 1 in the distance matrix. * Fix test failures * Update qiskit/transpiler/passes/layout/full_ancilla_allocation.py * Add test coverage for passes not run in preset pass managers * Fix lint * Fix reference dot for new preset pass managers * Rework arguments to take CouplingMap or Target This commit updates the pass constructors to use a single positional argument for both the coupling map or a target. This replaces the previous behavior where a new keyword argument was added to pass in the target. The exception to this is passes where the target was used differently than the coupling map for extra functionality. * Expand test coverage * Small typo fixes from review * Fix lint and test failure * Update qiskit/transpiler/preset_passmanagers/level0.py Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com> * Update logic to favor target in all optimization levels In an earlier commit we updated the logic in optimization level 0 to favor target instead of a coupling map. But, that ignored the other optimization levels. This commit corrects this oversight and adds the same logically change to all the preset pass managers. * Update release note for api changes in earlier iterations * Update docstrings * Use a single positional argument for embed pm function --------- Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com>

This commit deprecates the BIPMapping transpiler pass. Since its introduction the pass has been in a weird state since it's introduction. It is a standalone transpiler pass that we never integrated it into transpile() because it has an external dependency on cplex which is a proprietary software package that most people don't have access too. With the introduction of the transpiler stage plugin interface the pass has been turned into an external package: https://github.com/qiskit-community/qiskit-bip-mapper By using the plugin interface the pass can now be cleanly integrates into the transpile() function and also makes the requirement to have cplex installed much more explicit. For users with cplex it's much easier to run the BIPMapping pass as part of a transpilation workflow with `transpile(..., routing_method="bip")`. Closes Qiskit#8662

* Make all Layout and Routing passes target aware This commit updates all the layout and routing passes which leverage hardware constraints to have a new optional keyword argument, target, to specify a Target object for modelling the constraints of the target backend. This option will superscede any other specified arguments for backend constraints. For most of these passes the target argument is just internally leveraged to build a CouplingMap (and in 2 cases a BackendProperties) as most of the algorithms don't need more than the global connectivity graph and there is minimal overhead to convert from a target to a CouplingMap. The 2 passes which take backend properties should be refactored to work natively with the target because the conversion adds extra overhead which isn't needed, but in those cases the passes aren't enabled by default (or are slated to be moved out of tree as a plugin, see Qiskit#8662) so the extra overhead isn't of particular concern. This is part of the first step towards making all backend constraints for the transpiler used the Target internally (see Qiskit#9256). Once all passes that take hardware constraints parameters as an input are updated to have a target we can start internally using the Target only for all preset pass manager construction and start the long process of deprecating the legacy interface in these passes. Related to Qiskit#9256 * Update pass manager drawer reference dot files The pass manager drawer tests were comparing the dot output for statically built pass managers to test that the output visualization is generated correctly. However, the new arguments for the taget are changing the visualization (to show the new option) and caused these tests to fail. This commit updates the reference images to include the new argument. * Fix lint * Optimize CheckMap This commit optimizes the logic of the CheckMap pass. Previously we were unecessarily creating a distance matrix from the coupling graph, and with a target unecessarily creating a CouplingMap object. Instead of doing this instead the pass now creates a list of bidirectional edges for adjacent nodes in the coupling graph and then a set lookup is done for each check instead of checking for a distance of 1 in the distance matrix. * Fix test failures * Update qiskit/transpiler/passes/layout/full_ancilla_allocation.py * Add test coverage for passes not run in preset pass managers * Fix lint * Fix reference dot for new preset pass managers * Rework arguments to take CouplingMap or Target This commit updates the pass constructors to use a single positional argument for both the coupling map or a target. This replaces the previous behavior where a new keyword argument was added to pass in the target. The exception to this is passes where the target was used differently than the coupling map for extra functionality. * Expand test coverage * Small typo fixes from review * Fix lint and test failure * Update qiskit/transpiler/preset_passmanagers/level0.py Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com> * Update logic to favor target in all optimization levels In an earlier commit we updated the logic in optimization level 0 to favor target instead of a coupling map. But, that ignored the other optimization levels. This commit corrects this oversight and adds the same logically change to all the preset pass managers. * Update release note for api changes in earlier iterations * Update docstrings * Use a single positional argument for embed pm function --------- Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com>

…iskit#9924) * Deprecate the BIPMapping transpiler pass in favor of external plugin This commit deprecates the BIPMapping transpiler pass. Since its introduction the pass has been in a weird state since it's introduction. It is a standalone transpiler pass that we never integrated it into transpile() because it has an external dependency on cplex which is a proprietary software package that most people don't have access too. With the introduction of the transpiler stage plugin interface the pass has been turned into an external package: https://github.com/qiskit-community/qiskit-bip-mapper By using the plugin interface the pass can now be cleanly integrates into the transpile() function and also makes the requirement to have cplex installed much more explicit. For users with cplex it's much easier to run the BIPMapping pass as part of a transpilation workflow with `transpile(..., routing_method="bip")`. Closes Qiskit#8662 * Catch deprecation warnings in tests * Fix lint

* Make all Layout and Routing passes target aware This commit updates all the layout and routing passes which leverage hardware constraints to have a new optional keyword argument, target, to specify a Target object for modelling the constraints of the target backend. This option will superscede any other specified arguments for backend constraints. For most of these passes the target argument is just internally leveraged to build a CouplingMap (and in 2 cases a BackendProperties) as most of the algorithms don't need more than the global connectivity graph and there is minimal overhead to convert from a target to a CouplingMap. The 2 passes which take backend properties should be refactored to work natively with the target because the conversion adds extra overhead which isn't needed, but in those cases the passes aren't enabled by default (or are slated to be moved out of tree as a plugin, see Qiskit#8662) so the extra overhead isn't of particular concern. This is part of the first step towards making all backend constraints for the transpiler used the Target internally (see Qiskit#9256). Once all passes that take hardware constraints parameters as an input are updated to have a target we can start internally using the Target only for all preset pass manager construction and start the long process of deprecating the legacy interface in these passes. Related to Qiskit#9256 * Update pass manager drawer reference dot files The pass manager drawer tests were comparing the dot output for statically built pass managers to test that the output visualization is generated correctly. However, the new arguments for the taget are changing the visualization (to show the new option) and caused these tests to fail. This commit updates the reference images to include the new argument. * Fix lint * Optimize CheckMap This commit optimizes the logic of the CheckMap pass. Previously we were unecessarily creating a distance matrix from the coupling graph, and with a target unecessarily creating a CouplingMap object. Instead of doing this instead the pass now creates a list of bidirectional edges for adjacent nodes in the coupling graph and then a set lookup is done for each check instead of checking for a distance of 1 in the distance matrix. * Fix test failures * Update qiskit/transpiler/passes/layout/full_ancilla_allocation.py * Add test coverage for passes not run in preset pass managers * Fix lint * Fix reference dot for new preset pass managers * Rework arguments to take CouplingMap or Target This commit updates the pass constructors to use a single positional argument for both the coupling map or a target. This replaces the previous behavior where a new keyword argument was added to pass in the target. The exception to this is passes where the target was used differently than the coupling map for extra functionality. * Expand test coverage * Small typo fixes from review * Fix lint and test failure * Update qiskit/transpiler/preset_passmanagers/level0.py Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com> * Update logic to favor target in all optimization levels In an earlier commit we updated the logic in optimization level 0 to favor target instead of a coupling map. But, that ignored the other optimization levels. This commit corrects this oversight and adds the same logically change to all the preset pass managers. * Update release note for api changes in earlier iterations * Update docstrings * Use a single positional argument for embed pm function --------- Co-authored-by: Ali Javadi-Abhari <ajavadia@users.noreply.github.com>

…iskit#9924) * Deprecate the BIPMapping transpiler pass in favor of external plugin This commit deprecates the BIPMapping transpiler pass. Since its introduction the pass has been in a weird state since it's introduction. It is a standalone transpiler pass that we never integrated it into transpile() because it has an external dependency on cplex which is a proprietary software package that most people don't have access too. With the introduction of the transpiler stage plugin interface the pass has been turned into an external package: https://github.com/qiskit-community/qiskit-bip-mapper By using the plugin interface the pass can now be cleanly integrates into the transpile() function and also makes the requirement to have cplex installed much more explicit. For users with cplex it's much easier to run the BIPMapping pass as part of a transpilation workflow with `transpile(..., routing_method="bip")`. Closes Qiskit#8662 * Catch deprecation warnings in tests * Fix lint

mtreinish added type: feature request New feature or request short project labels Sep 2, 2022

mtreinish mentioned this issue Dec 7, 2022

Make all Layout and Routing passes target aware #9263

Merged

mtreinish mentioned this issue Apr 6, 2023

Deprecate the BIPMapping transpiler pass in favor of external plugin #9924

Merged

jakelishman closed this as completed in #9924 Apr 12, 2023

jakelishman closed this as completed in 602d340 Apr 12, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Convert `BIPMaping` transpiler pass into an external plugin #8662

Convert `BIPMaping` transpiler pass into an external plugin #8662

mtreinish commented Sep 2, 2022 •

edited

Loading

Convert BIPMaping transpiler pass into an external plugin #8662

Convert BIPMaping transpiler pass into an external plugin #8662

Comments

mtreinish commented Sep 2, 2022 • edited Loading

What should we add?

Convert `BIPMaping` transpiler pass into an external plugin #8662

Convert `BIPMaping` transpiler pass into an external plugin #8662

mtreinish commented Sep 2, 2022 •

edited

Loading