Fully QPY compatible pulse builder for V2 and runtime backend #8949

nkanazawa1989 · 2022-10-19T10:59:17Z

Summary

At present, Qiskit Pulse supports both Schedule and ScheduleBlock representation, but QPY only supports ScheduleBlock representation. This causes an issue in V2 IBM provider and runtime provider, since they assume QPY payload, and the pulse builder still accepts Schedule. For example,

x_gate_sched = backend.target["x"][(0, )].calibration

with builder() as schedule:
    call(x_gate_sched)

backend.run(schedule)

this execution will fail with V2 instance because x_gate_sched is still provided in legacy schedule format (because pulse defaults data is still PulseQobj format in IBM provider, regardless of backend version). Same issue happens in RZX calibration builder passes, which is in high demand for pulse level optimization of application circuits.

This PR allows V2 provider and runtime provider to execute such a program, and circuit optimized with RZX calibration builder passes.

Details and comments

This PR consists of several commits.

805a5b2

This commit removes usage of Schedule and Call from the pulse builder. Previously Schedule was wrapped by Call instruction and added to the schedule block as an instruction. Note that Call instruction is not QPY compatible because it internally keeps Schedule as-is (just a wrapper).
With this commit, schedule is internally converted into schedule block representation to guarantee the QPY compatibility. Because schedule block is not aware of absolute time interval of instruction, a helper compiler directive AreaBarrier is introduced to support instruction scheduling. This instruction must be removed before execution because this is not necessary supported by the hardware.

a02bcb5

This commit update QPY to support AreaBarrier instruction. Now builder-generated program is fully QPY compatible.

f2bc607

This commit updates the RZX calibration builder passes to use the updated pulse builder to construct stretched pulse schedules. Thus QPY compatibility is guaranteed.

(note)
AreaBarrier is renamed to TimeBlocked according to the review comment.

Alternative

Alternatively, we could add Schedule support to QPY to fix this problem. However, I hesitate this approach because QPY supports indicates we cannot drop this representation from the code stack permanently, due to the QPY backward compatibility. Schedule tends to be legacy, because we cannot support frame properly on top of it (because of its poor time slot representation).

Fix #8933

…tly converted into ScheduleBlock with AreaBarrier instructions.

qiskit-bot · 2022-10-19T10:59:20Z

Thank you for opening a new pull request.

Before your PR can be merged it will first need to pass continuous integration tests and be reviewed. Sometimes the review process can be slow, so please be patient.

While you're waiting, please feel free to review other open PRs. While only a subset of people are authorized to approve pull requests for merging, everyone is encouraged to review open pull requests. Doing reviews helps reduce the burden on the core team and helps make the project's code better for everyone.

One or more of the the following people are requested to review this:

@Qiskit/terra-core
@mtreinish
@nkanazawa1989

…ial context.

eggerdj

Thanks a lot. Overall, this looks fine. I have some questions that need clarification. See the comments.

qiskit/pulse/builder.py

eggerdj · 2022-10-19T12:35:54Z

qiskit/pulse/instructions/directives.py

+            block = ScheduleBlock()
+            block.append(AreaBarrier(120, DriveChannel(0)))
+            block.append(Play(Constant(10, 0.1), DriveChannel(0)))


This feels like we are mixing representations. How does this work with parametric duration? What about

block = ScheduleBlock() block.append(AreaBarrier(120, DriveChannel(0))) block.append(Play(Constant(Parameter("duration"), 0.1), DriveChannel(0)))

As I type hinted, AreaBarrier doesn't accept parameter. This is internally used to convert schedule into block, in which schedule doesn't take parameterized duration.

qiskit/transpiler/passes/calibration/rzx_builder.py

releasenotes/notes/upgrade-pulse-builder-and-rzx-builder-033ac8ad8ad2a192.yaml

Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com>

coveralls · 2022-10-19T17:43:37Z

Pull Request Test Coverage Report for Build 3528859271

151 of 163 (92.64%) changed or added relevant lines in 6 files are covered.
1 unchanged line in 1 file lost coverage.
Overall coverage increased (+0.02%) to 84.22%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/pulse/instructions/directives.py	15	16	93.75%
qiskit/pulse/transforms/canonicalization.py	19	20	95.0%
qiskit/transpiler/passes/calibration/rzx_builder.py	58	60	96.67%
qiskit/pulse/builder.py	53	61	86.89%

Files with Coverage Reduction	New Missed Lines	%
qiskit/pulse/builder.py	1	90.84%

Totals
Change from base Build 3527305551:	0.02%
Covered Lines:	62400
Relevant Lines:	74092

💛 - Coveralls

eggerdj

A few more comments and a suggestion for the directive name.

eggerdj · 2022-10-21T14:56:42Z

qiskit/pulse/instructions/directives.py

+    """Pulse ``AreaBarrier`` directive.
+
+    This instruction is intended to be used internally within the pulse builder,
+    to naively convert :class:`.Schedule` into :class:`.ScheduleBlock`.


If this is used for a naive conversion then what would a non-naive conversion look like?

In the current mechanism I "give up" to infer the original alignment context. For example,

with align_sequential(): with align_left(): play(cr45p, ControlChannel(1)) play(cr45p_d0, DriveChannel(0)) play(x, DriveChannel(1)) with align_left(): play(cr45m, ControlChannel(1)) play(cr45m_d0, DriveChannel(0)) play(x, DriveChannel(1))

this would be the context of the ECR schedule, but current mechanism will generate

with align_left(): play(cr45p, ControlChannel(1)) play(cr45p_d0, DriveChannel(0)) time_block(xxx, DriveChannel(1)) play(x, DriveChannel(1)) ...

Note that we cannot modify any pulse duration (otherwise you must also modify time block duration), but this should be okey because we are consuming "already scheduled" sequence. To recover original alignment context, we may need to generate DAG to find parallel context. Such logic may become heavy and still has a rick of some edge case. So I just avoid doing over engineering and choose to say "this is naive version of alignment context recovery".

Added bit more detail in 7ab2710

qiskit/pulse/instructions/directives.py

Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com>

eggerdj

LGTM.

…d_schedule_block_like

eggerdj

LGTM

mtreinish

Overall this LGTM I just had a couple of small questions inline

mtreinish · 2022-11-11T12:09:56Z

qiskit/pulse/builder.py

@@ -731,17 +737,6 @@ def append_instruction(self, instruction: instructions.Instruction):
        """
        self._context_stack[-1].append(instruction)

-    @_compile_lazy_circuit_before
-    def append_block(self, context_block: ScheduleBlock):


Are there any backwards compatibility implications with this removal. The documentation referred to this method previously so was it user facing or is this all internal?

This should be okey because builder itself is a protected class.
https://github.com/Qiskit/qiskit-terra/blob/8bfb058b9d94353936d7e2aea61015095487b873/qiskit/pulse/builder.py#L520-L521

qiskit/pulse/instructions/__init__.py

qiskit/qpy/type_keys.py

releasenotes/notes/upgrade-pulse-builder-and-rzx-builder-033ac8ad8ad2a192.yaml

- render class docs of the directive instructions - update qpy format documentation - add separate reno for time blockade

mtreinish

Overall I'm happy with the state of the code now, thanks for making the updates. I just have a few small inline comments. The only one I think is really required is adding a unit test for RZXCalibration* with QPY, other than that I'm happy to approve this.

releasenotes/notes/upgrade-pulse-builder-and-rzx-builder-033ac8ad8ad2a192.yaml

test/python/transpiler/test_calibrationbuilder.py

mtreinish · 2022-11-16T13:42:08Z

test/python/qpy/test_block_load_from_qpy.py

@@ -187,6 +194,20 @@ def test_nested_blocks(self):
                    builder.delay(200, DriveChannel(1))
        self.assert_roundtrip_equal(test_sched)

+    def test_called_schedule(self):


One thing I think would be good is to maybe have a qpy test with the rzx calibration passes. Since a big part of this PR is fixing the QPY support with the pass output having a test to verify the pass output can be round tripped through qpy is simple way to ensure we don't regress on this moving forward.

Thanks Matthew. I added new tests for QPY. I also refactored the RZX builder test in 8562d98 since the output schedule has been never really tested. This is why the alignment error has been existed before this PR (this PR unintentionally fixed it).

…es more robust to the rounding error.

mtreinish

Thanks for the updates it lgtm now.

…#8949) * Eliminate use of Schedule in the builder context. Schedule is implicitly converted into ScheduleBlock with AreaBarrier instructions. * Support AreaBarrier instruction in QPY * Update RZX calibration builder to generate ScheduleBlock * Add release note * Support python3.7 * Fix hadamard schedule in the RZX builder. This should not use sequential context. * Review comments Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com> * Unified append_block and append_subroutine * Update instruction name AreaBarrier -> TimeBlockade * Documentation updates Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com> * Remove old name * lint fix * add release note for bugfix * review comments - render class docs of the directive instructions - update qpy format documentation - add separate reno for time blockade * more detailed upgrade notes * add new unittests. rescale_cr_inst method is updated so that it becomes more robust to the rounding error. Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com>

nkanazawa1989 added 4 commits October 19, 2022 19:10

Eliminate use of Schedule in the builder context. Schedule is implici…

805a5b2

…tly converted into ScheduleBlock with AreaBarrier instructions.

Support AreaBarrier instruction in QPY

a02bcb5

Update RZX calibration builder to generate ScheduleBlock

f2bc607

Add release note

17d2a46

nkanazawa1989 requested review from a team, eggerdj and wshanks as code owners October 19, 2022 10:59

nkanazawa1989 mentioned this pull request Oct 19, 2022

CalibrationBuilders should not use Schedule #8933

Closed

nkanazawa1989 added 2 commits October 19, 2022 20:33

Support python3.7

4d0ddf5

Fix hadamard schedule in the RZX builder. This should not use sequent…

6a90761

…ial context.

nkanazawa1989 force-pushed the upgrade/add_schedule_block_like branch from 0f02d6a to 6a90761 Compare October 19, 2022 11:33

eggerdj requested changes Oct 19, 2022

View reviewed changes

nkanazawa1989 and others added 2 commits October 20, 2022 02:14

Review comments

0ca8efb

Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com>

Unified append_block and append_subroutine

de2f7d5

Merge branch 'main' into upgrade/add_schedule_block_like

a69aac0

eggerdj reviewed Oct 21, 2022

View reviewed changes

nkanazawa1989 and others added 4 commits October 24, 2022 11:38

Update instruction name AreaBarrier -> TimeBlockade

1f1db8d

Documentation updates

7ab2710

Co-authored-by: Daniel J. Egger <38065505+eggerdj@users.noreply.github.com>

Remove old name

15887c8

lint fix

37f4cbd

eggerdj previously approved these changes Oct 25, 2022

View reviewed changes

kevinsung mentioned this pull request Oct 27, 2022

RZXCalibrationBuilder aligns pulses incorrectly #9013

Closed

jakelishman linked an issue Oct 27, 2022 that may be closed by this pull request

RZXCalibrationBuilder aligns pulses incorrectly #9013

Closed

nkanazawa1989 added 2 commits October 28, 2022 16:23

Merge branch 'main' of github.com:Qiskit/qiskit-terra into upgrade/ad…

5434f79

…d_schedule_block_like

add release note for bugfix

1504aaf

nkanazawa1989 dismissed eggerdj’s stale review via 1504aaf October 28, 2022 07:27

eggerdj previously approved these changes Oct 30, 2022

View reviewed changes

Merge branch 'main' into upgrade/add_schedule_block_like

bee3fca

mtreinish added Changelog: New Feature Include in the "Added" section of the changelog Changelog: API Change Include in the "Changed" section of the changelog Changelog: Bugfix Include in the "Fixed" section of the changelog labels Oct 31, 2022

mtreinish added this to the 0.23.0 milestone Oct 31, 2022

mtreinish self-assigned this Oct 31, 2022

mtreinish reviewed Nov 11, 2022

View reviewed changes

review comments

1756e35

- render class docs of the directive instructions - update qpy format documentation - add separate reno for time blockade

nkanazawa1989 dismissed eggerdj’s stale review via 1756e35 November 14, 2022 05:21

mtreinish reviewed Nov 16, 2022

View reviewed changes

nkanazawa1989 added 2 commits November 22, 2022 12:51

more detailed upgrade notes

b17595e

add new unittests. rescale_cr_inst method is updated so that it becom…

8562d98

…es more robust to the rounding error.

mtreinish approved these changes Nov 22, 2022

View reviewed changes

mtreinish added the automerge label Nov 22, 2022

Merge branch 'main' into upgrade/add_schedule_block_like

41f6d1b

mergify bot merged commit e5ea411 into Qiskit:main Nov 23, 2022

nkanazawa1989 deleted the upgrade/add_schedule_block_like branch November 25, 2022 02:56

This was referenced Feb 3, 2023

Update qpy terra to 0.23.1 version Qiskit/qiskit-ibm-provider#496

Merged

Update qpy version to terra 0.23.1 Qiskit/qiskit-ibm-runtime#577

Merged

nkanazawa1989 mentioned this pull request May 11, 2023

Deprecate pulse Call instruction #10103

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fully QPY compatible pulse builder for V2 and runtime backend #8949

Fully QPY compatible pulse builder for V2 and runtime backend #8949

nkanazawa1989 commented Oct 19, 2022 •

edited

Loading

qiskit-bot commented Oct 19, 2022

eggerdj left a comment

eggerdj Oct 19, 2022

nkanazawa1989 Oct 19, 2022

coveralls commented Oct 19, 2022 •

edited

Loading

eggerdj left a comment

eggerdj Oct 21, 2022

nkanazawa1989 Oct 24, 2022 •

edited

Loading

nkanazawa1989 Oct 24, 2022

eggerdj left a comment

eggerdj left a comment

mtreinish left a comment

mtreinish Nov 11, 2022

nkanazawa1989 Nov 14, 2022

mtreinish left a comment

mtreinish Nov 16, 2022

nkanazawa1989 Nov 22, 2022

mtreinish left a comment

Fully QPY compatible pulse builder for V2 and runtime backend #8949

Fully QPY compatible pulse builder for V2 and runtime backend #8949

Conversation

nkanazawa1989 commented Oct 19, 2022 • edited Loading

Summary

Details and comments

Alternative

qiskit-bot commented Oct 19, 2022

eggerdj left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

coveralls commented Oct 19, 2022 • edited Loading

Pull Request Test Coverage Report for Build 3528859271

💛 - Coveralls

eggerdj left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nkanazawa1989 Oct 24, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

eggerdj left a comment

Choose a reason for hiding this comment

eggerdj left a comment

Choose a reason for hiding this comment

mtreinish left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mtreinish left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mtreinish left a comment

Choose a reason for hiding this comment

nkanazawa1989 commented Oct 19, 2022 •

edited

Loading

coveralls commented Oct 19, 2022 •

edited

Loading

nkanazawa1989 Oct 24, 2022 •

edited

Loading