Fix ignored errors/durations in generate_preset_pass_manager if dt is set
#14065
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
One or more of the following people are relevant to this code:
|
…er target to reset internal instruction_durations value so that new dt can be taken into account
ElePT
commented
Mar 21, 2025
qiskit/transpiler/preset_passmanagers/generate_preset_pass_manager.py
Outdated
Show resolved
Hide resolved
mtreinish
reviewed
Mar 21, 2025
qiskit/transpiler/preset_passmanagers/generate_preset_pass_manager.py
Outdated
Show resolved
Hide resolved
Comment on lines
2316
to
2320
| tqc = transpile( | ||
| qc, backend=backend, seed_transpiler=4242, callback=callback, dt=backend.dt * 2 | ||
| ) | ||
| self.assertTrue(vf2_post_layout_called) | ||
| self.assertEqual([2, 1, 0], _get_index_layout(tqc, qubits)) |
There was a problem hiding this comment.
I think to test this we either should run transpile twice once with a custom dt once with out and assert the layouts are the same providing indication that we've got the same error rates for both runs.
Or alternatively we could probably test this more directly by building a preset pass manager and poking the internals of one of the passes to assert the target has the expected error rates.
Co-authored-by: Matthew Treinish <mtreinish@kortar.org>
Pull Request Test Coverage Report for Build 13995670918Warning: This coverage report may be inaccurate.This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.
Details
💛 - Coveralls |
8000
mtreinish
reviewed
Mar 21, 2025
qiskit/transpiler/preset_passmanagers/generate_preset_pass_manager.py
Outdated
Show resolved
Hide resolved
mtreinish
approved these changes
Mar 25, 2025
There was a problem hiding this comment.
Thanks for the update. I'm mildly concerned that the test will be fragile longer term because we're indirectly testing the root cause of the bug here. But at the end of the day I don't think it's likely to cause an issue and the test would catch the regression even if it's indirect.
mergify bot
pushed a commit
that referenced
this pull request
Mar 25, 2025
…` is set (#14065) * Fix dt oversight in generate_preset_pass_manager * Update warning to mention explicitly the invalidation of custom durations and error rates. * Do not overwrite original backend values inside transpile, iterate over target to reset internal instruction_durations value so that new dt can be taken into account * One iteration is enough to reset the value of self._instruction_durations * Estimate duration instead of hardcoding it * Apply suggestion from Matt's code review Co-authored-by: Matthew Treinish <mtreinish@kortar.org> * Fix example * Move cache invalidation to target dt setter. * Add reno --------- Co-authored-by: Matthew Treinish <mtreinish@kortar.org> (cherry picked from commit 6ce7788)
github-merge-queue bot
pushed a commit
that referenced
this pull request
Mar 25, 2025
…` is set (#14065) (#14098) * Fix dt oversight in generate_preset_pass_manager * Update warning to mention explicitly the invalidation of custom durations and error rates. * Do not overwrite original backend values inside transpile, iterate over target to reset internal instruction_durations value so that new dt can be taken into account * One iteration is enough to reset the value of self._instruction_durations * Estimate duration instead of hardcoding it * Apply suggestion from Matt's code review Co-authored-by: Matthew Treinish <mtreinish@kortar.org> * Fix example * Move cache invalidation to target dt setter. * Add reno --------- Co-authored-by: Matthew Treinish <mtreinish@kortar.org> (cherry picked from commit 6ce7788) Co-authored-by: Elena Peña Tapia <57907331+ElePT@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
#14056 (comment) reported an unexpected behavior of
generate_preset_pass_managerwhere setting a customdtwould invalidate the backend's gate durations and error rates. This was due to an oversight in #9256 , where a custom target would be built from scratch usingTarget.from_configurationwhenever any loose constraint was set, and any information regarding instruction properties would be lost. This was intentional forcoupling_mapandbasis_gates, as they modify the target gate map, but not necessary fordt, where the target could be kept and simply updated.This PR fixes this use case, adds a test with vf2 + dt, and modifies the user warning to explicitly communicate what will happen with gate durations and errors if
coupling_maporbasis_gatesare set with abackend.Details and comments
The changelog is None as #9256 has not been released yet.