Allow circuits using only SI units to pass through TimeUnitConversion #14027

wshanks · 2025-03-14T17:47:44Z

In f0576fa64, the TimeUnitConversion pass was modified so that it raises an exception if a circuit has only SI unit delays with no dt value set, but the intent seemed to be catch cases where SI units were used alongside instructions with dt units. Here the conditions for raising an exception are modified to allow circuits with all SI unit delays (which usually only happens when working with a backend like AerSimulator) to pass through.

Closes #14016

In f0576fa64, the `TimeUnitConversion` pass was modified so that it raises an exception if a circuit has only SI unit delays with no `dt` value set, but the intent seemed to be catch cases where SI units were used alongside instructions with `dt` units. Here the conditions for raising an exception are modified to allow circuits with all SI unit delays (which usually only happens when working with a backend like `AerSimulator`) to pass through. Closes Qiskit#14016

qiskit-bot · 2025-03-14T17:47:49Z

One or more of the following people are relevant to this code:

@Qiskit/terra-core

mtreinish

I think this change makes sense, do you want to add a test to ensure we don't regress here again?

coveralls · 2025-03-14T18:16:20Z

Pull Request Test Coverage Report for Build 13909958997

Warning: 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.

For more information on this, see Tracking coverage changes with pull request builds.
To avoid this issue with future PRs, see these Recommended CI Configurations.
For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

7 of 7 (100.0%) changed or added relevant lines in 1 file are covered.
No unchanged relevant lines lost coverage.
Overall coverage decreased (-0.03%) to 88.087%

Totals
Change from base Build 13853660516:	-0.03%
Covered Lines:	72603
Relevant Lines:	82422

💛 - Coveralls

wshanks · 2025-03-14T18:20:21Z

I thought about that but I wasn't sure where the test would go. I didn't see much in the way of existing tests. I think a new test/python/transpiler/test_time_unit_conversion.py would be needed?

kevinhartman · 2025-03-14T20:08:51Z

That is most probably a bug, thanks for finding it.

There is already a test that makes sure that SI-only can pass through when dt is not specified:
https://github.com/Qiskit/qiskit/blob/main/test/python/compiler/test_transpiler.py#L1439

But it tests this only for new expression-based delay durations. Perhaps you can add a test beside it for the old-style delay duration specification.

kevinhartman · 2025-03-14T20:38:38Z

qiskit/transpiler/passes/scheduling/time_unit_conversion.py

@@ -126,9 +126,9 @@ def run(self, dag: DAGCircuit):
            # Check what units are used in other instructions: dt or SI or mixed
            units_other = inst_durations.units_used()
            unified_unit = self._unified(units_other)
-            if unified_unit == "SI" and not has_dt:
+            if unified_unit in ("SI", "none") and has_si and not has_dt:


I think we may just want this:

if (has_si or unified_unit == "SI") and not has_dt: time_unit = "s" elif (has_dt or unified_unit == "dt") and not has_si: time_unit = "dt"

If _unified returns "dt" because its input set is empty, then the elif gets executed, which is the behavior we want (defaulting to dt).

And wouldn't need the change to return "none".

Ugh, there are so many edge cases 🙂 With (has_si or unified_unit == "SI") and not has_dt is that it would be true for a circuit with all SI unit delays but with instructions in InstructionDurations that were in dt. Isn't that a problem? (You could also consider the edge case where the instructions with dt units in InstructionDurations are not used in the current circuit).

Yup, you're right. Would something like this be more readable? This assumes the presence of your addition of returning "none" in _unified.

if inst_durations.dt is None: # Check what units are used in other instructions: dt or SI or mixed units_other = inst_durations.units_used() unified_unit = self._unified(units_other) has_si = has_si or unified_unit in {"SI", "mixed"} has_dt = has_dt or unified_unit in {"dt", "mixed"} if has_dt and has_si: raise TranspilerError( "Fail to unify time units. SI units " "and dt unit must not be mixed when dt is not supplied." ) if has_si: time_unit = "s" else: # has_dt may be false (i.e. 'unified_unit' is "none"), but we still default to 'dt' # in this case. time_unit = "dt" else: time_unit = "dt"

That looks good to me. I tried pushing a version like that.

wshanks · 2025-03-14T21:03:14Z

@kevinhartman I wrote some tests before I saw your comment about the existing tests. I pushed them for reference. We move or rework them if you think that is better.

kevinhartman

Looking good, just a few minor comments with the testing.

kevinhartman · 2025-03-17T15:19:35Z

test/python/transpiler/test_time_unit_conversion.py

+    @unpack
+    def test_gate_and_delay(self, delay_unit, inst_durations_unit, target_type):
+        """Test delays are converted, passed through or flagged as appropriate"""
+        if delay_unit == "s":


I think we generally prefer to "unroll" test code so it's simple to debug when something goes wrong. The orange flag for me here is that the test contains logic to special-case itself based on the ddt inputs. If you don't mind, I think it'd be helpful if you could break this into three separate tests instead, even though it'll be more verbose.

I moved the tests to test_transpiler in b749029. To fit in there, I used transpile() instead of the TimeUnitConversion interface. That shifted the scope of what seemed like relevant tests. TimeUnitConversion takes instruction durations and a target and behaves differently depending on the units of the InstructionDurations object and whether or not it or the target have a dt. With transpile, the durations only come from the backend and that just comes from backend.target.durations(). In Qiskit, InstructionProperties specifies duration in seconds so that prevents dt units in the InstructionDurations. So that removes a lot of the variants I was covering with my tests before. The si->dt case already had coverage so I added the most common cases of si->si with no backend or a backend without dt and si->error with mixed units and no dt. I think the other nuances my previous tests were covering about instruction durations versus target precedence are not so important for transpile since the input is only backend or target and not instruction durations directly.

Makes sense. Presumably, the tests you've added wouldn't have passed before your change at least, yeah?

test.python.compiler.test_transpiler.TestTranspile.test_delay_converts_to_seconds fails as expected:

test.python.compiler.test_transpiler.TestTranspile.test_delay_converts_to_seconds --------------------------------------------------------------------------------- Captured traceback: ~~~~~~~~~~~~~~~~~~~ Traceback (most recent call last): File "/qiskit-terra/test/python/compiler/test_transpiler.py", line 1382, in test_delay_converts_to_seconds out = transpile([qc, qc], seed_transpiler=42) File "/qiskit-terra/qiskit/compiler/transpiler.py", line 291, in transpile out_circuits = pm.run(circuits, callback=callback, num_processes=num_processes) ... File "/qiskit-terra/qiskit/transpiler/passes/scheduling/time_unit_conversion.py", line 134, in run raise TranspilerError( ...<2 lines>... ) qiskit.transpiler.exceptions.TranspilerError: 'Fail to unify time units. SI units and dt unit must not be mixed when dt is not supplied.'

test_rejects_mixed_units_delay_without_target_dt passes, but I added that one because I was trying to carry over my ddt tests and didn't see a test like it (there was an expression version) and not because that behavior was affected by my change.

test/python/transpiler/test_time_unit_conversion.py

kevinhartman

Looks great, thanks a lot for the fix!

If it doesn't pass lint, I'll come back to approve again after you update.

kevinhartman · 2025-03-17T19:31:45Z

test/python/compiler/test_transpiler.py

@@ -90,6 +90,7 @@
 from qiskit.transpiler.passmanager_config import PassManagerConfig
 from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager, level_0_pass_manager
 from qiskit.transpiler.target import (
+    InstructionDurations,


Lint is likely to yell at you here 🙂.

Oh, right, this was from when I was trying to cover InstructionDurations in the old test before I realized that it was not really directly relevant for transpile since InstructionProperties is in seconds and transpile does not take instruction_durations as an input.

wshanks · 2025-03-17T21:14:34Z

Oops, there was a bad variable name left in from when I was debugging and some stray changes in tox.ini which I have removed now. Hopefully the tests are clean now.

(I did lint locally but just missed the warnings. I don't have a smooth setup for Qiskit right now which is part of why I was fiddling with tox.ini. I was getting extra linting warnings possibly because of some other pylintrc file I have or something like that).

kevinhartman

Thanks for the updates!

This looks good to me now.

…#14027) * Allow circuits using only SI units to pass through TimeUnitConversion In f0576fa64, the `TimeUnitConversion` pass was modified so that it raises an exception if a circuit has only SI unit delays with no `dt` value set, but the intent seemed to be catch cases where SI units were used alongside instructions with `dt` units. Here the conditions for raising an exception are modified to allow circuits with all SI unit delays (which usually only happens when working with a backend like `AerSimulator`) to pass through. Closes #14016 * Add TimeUnitConverion tests * Update test description * Remove comment * lint test file * Refactor TimeUnitConversion for readability * Move time unit converison tests to test_transpiler * lint * lint (cherry picked from commit 740a9dc)

wshanks requested a review from a team as a code owner March 14, 2025 17:47

wshanks mentioned this pull request Mar 14, 2025

Transpiling circuit with SI unit delay raises exception with Qiskit 2.0 #14016

Closed

mtreinish added the stable backport potential The bug might be minimal and/or import enough to be port to stable label Mar 14, 2025

mtreinish added this to the 2.0.0 milestone Mar 14, 2025

mtreinish reviewed Mar 14, 2025

View reviewed changes

kevinhartman reviewed Mar 14, 2025

View reviewed changes

wshanks added 2 commits March 14, 2025 17:00

Add TimeUnitConverion tests

fcddba3

Update test description

615f56d

wshanks added 3 commits March 14, 2025 17:04

Remove comment

1b8f1c1

lint test file

c2b934d

Refactor TimeUnitConversion for readability

e457f5d

kevinhartman reviewed Mar 17, 2025

View reviewed changes

Move time unit converison tests to test_transpiler

b749029

kevinhartman previously approved these changes Mar 17, 2025

View reviewed changes

kevinhartman self-assigned this Mar 17, 2025

lint

275a9dd

wshanks dismissed kevinhartman’s stale review via 275a9dd March 17, 2025 19:47

lint

8cd5e67

kevinhartman approved these changes Mar 18, 2025

View reviewed changes

kevinhartman added this pull request to the merge queue Mar 18, 2025

Merged via the queue into Qiskit:main with commit 740a9dc Mar 18, 2025
20 checks passed

mergify bot mentioned this pull request Mar 18, 2025

Allow circuits using only SI units to pass through TimeUnitConversion (backport #14027) #14050

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow circuits using only SI units to pass through TimeUnitConversion #14027

Allow circuits using only SI units to pass through TimeUnitConversion #14027

wshanks commented Mar 14, 2025

qiskit-bot commented Mar 14, 2025

mtreinish left a comment

coveralls commented Mar 14, 2025 •

edited

Loading

wshanks commented Mar 14, 2025

kevinhartman commented Mar 14, 2025 •

edited

Loading

kevinhartman Mar 14, 2025

wshanks Mar 14, 2025

kevinhartman Mar 14, 2025

wshanks Mar 14, 2025

wshanks commented Mar 14, 2025

kevinhartman left a comment

kevinhartman Mar 17, 2025

wshanks Mar 17, 2025

kevinhartman Mar 17, 2025

wshanks Mar 17, 2025

kevinhartman left a comment

kevinhartman Mar 17, 2025

wshanks Mar 17, 2025

wshanks commented Mar 17, 2025 •

edited

Loading

kevinhartman left a comment

Allow circuits using only SI units to pass through TimeUnitConversion #14027

Allow circuits using only SI units to pass through TimeUnitConversion #14027

Conversation

wshanks commented Mar 14, 2025

qiskit-bot commented Mar 14, 2025

mtreinish left a comment

Choose a reason for hiding this comment

coveralls commented Mar 14, 2025 • edited Loading

Pull Request Test Coverage Report for Build 13909958997

Warning: This coverage report may be inaccurate.

Details

💛 - Coveralls

wshanks commented Mar 14, 2025

kevinhartman commented Mar 14, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wshanks commented Mar 14, 2025

kevinhartman left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kevinhartman left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wshanks commented Mar 17, 2025 • edited Loading

kevinhartman left a comment

Choose a reason for hiding this comment

coveralls commented Mar 14, 2025 •

edited

Loading

kevinhartman commented Mar 14, 2025 •

edited

Loading

wshanks commented Mar 17, 2025 •

edited

Loading