[XEB] Support parallel execution #3760

mpharrigan · 2021-02-06T05:36:40Z

Support parallel execution of XEB circuits.

Instead of generating a set of circuits for each pair, we generate one library of two-qubit circuits which are mixed-and-matched during execution. This speeds up circuit generation, circuit serialization (for historical record), and speeds up simulation (when not optimizing ... when you're optimizing each pair presumably gets its own set of angles and needs to be simulated separately)
As such, a new function is added to the random_circuit_generation module to create a combinations object that describes how to mix-and-match these two-qubit circuits
The sample function is modified to accept this combination description to on-the-fly zip the circuits, sample from them, and split out the results. This allows maximal code re-use between isolated and parallel XEB! Analysis is mostly shared! Huge caveat below
If you run the characterize_phased_fsim_parameters_with_xeb on the code in this PR, it will find only one set of angles that works for the average of all pairs. I've split up the modifications to that half of the code-base into a subsequent PR.

This can be reviewed independently of #3753

review-notebook-app · 2021-02-06T05:36:43Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

…the time

karlunho · 2021-02-09T00:47:54Z

The tutorials look good, and I suggest that we create entries into the book.yaml so they show up on the documentation site.

cirq/experiments/fidelity_estimation.py

cirq/experiments/fidelity_estimation_test.py

cirq/experiments/random_quantum_circuit_generation.py

cirq/experiments/fidelity_estimation.py

cirq/experiments/random_quantum_circuit_generation.py

docs/characterization/parallel_xeb.ipynb

mpharrigan

address some of the comments

cirq/experiments/fidelity_estimation.py

mpharrigan · 2021-02-09T19:00:53Z

cirq/experiments/fidelity_estimation.py

+    Args:
+        wide_circuit: The zipped circuit on all pairs
+        pairs: The pairs of qubits operated on in the wide circuit.
+        layer_i: The index of the GridInteractionLayer to which the pairs correspond.


This might become clear in a following PR, but let me try to explain and then I'll think about how to make it clearer in code.

The primary idea is that you have a number of different moments that define the spacial layout of simultaneous two-qubit gates. The essential essence of each of those moments/layers is just the list of pairs, here as pairs. Indeed, pairs (and the circuit) is really the only essential part of this dataclass. The other attributes are metadata to maintain the provenance of the particular set of pairs.

Previous XEB implementations relied heavily on these GridInteractionLayer objects to semi-implicitly define the set of pairs by making several assumptions (like you're operating on a nice grid, all the couplings on the grid work, you'll be using the whole device and/or the calling function will downsample only the pairs that actually exist). So in traditional XEB there are always four layers (one for each of the N, S, E, W set of pairs) and they're represented by these objects.

Of course, for the calibration API we re-purpose cirq.Moment as a "layer". What you'll see in a following PR is that these fields will either contain a GridInteractionLayer, a Moment, or even None since it's just metadata that doesn't really matter as long as you fill in pairs.

layer_i and combination_i are just propagated through to be part of the result object. Let me see if I can clarify that

cirq/experiments/fidelity_estimation.py

mpharrigan · 2021-02-09T19:13:03Z

cirq/experiments/fidelity_estimation.py

+    Args:
+        wide_circuit: The zipped circuit on all pairs
+        pairs: The pairs of qubits operated on in the wide circuit.
+        layer_i: The index of the GridInteractionLayer to which the pairs correspond.


I'm changing the docstrings to be more clear about what layer_i and combination_i are. I've added a note to both that they don't modify any behavior, but are just propagated to the output. I'm also re-ordering the fields in this dataclass to put them last, because they are less important. I've added a note to combination that it's important to be able to "unzip" the circuit.

cirq/experiments/fidelity_estimation.py

cirq/experiments/fidelity_estimation_test.py

cirq/experiments/random_quantum_circuit_generation.py

mrwojtek · 2021-02-10T14:24:42Z

Matthew, are you still working on the changes or this is ready for next round of review?

mpharrigan · 2021-02-10T16:29:31Z

Ready for review! I think I've addressed everything

cirq/experiments/fidelity_estimation.py

mrwojtek · 2021-02-11T08:28:13Z

cirq/experiments/fidelity_estimation.py

+            # Slicing creates a copy, although this isn't documented
+            prepared_circuit = zipped_circuit.wide_circuit[:circuit_depth]
+            for pair_i, pair in enumerate(zipped_circuit.pairs):
+                prepared_circuit += ops.measure(*pair, key=str(pair_i))


Have you checked if this aligns all the measurements in one moment?

for the current code, it should just because these are dense circuits by construction; but I've changed it to add the measure operations in one moment. Good spot!

cirq/experiments/fidelity_estimation.py

cirq/experiments/random_quantum_circuit_generation.py

mpharrigan · 2021-02-11T19:05:43Z

back at you @mrwojtek !

CirqBot · 2021-02-12T00:07:26Z

Automerge cancelled: Merging is blocked (I don't understand why).

This PR is purely moving things around (on top of #3760) with one exception noted inline. - `random_quantum_circuit_generation` -- left alone; this is a well-scoped file - `xeb_sampling` - functionality related to sampling random circuits efficiently (for xeb) - `xeb_simulation` - functionality related to simulating 2-qubit circuits at various depths (for xeb) efficiently using multiprocessing - `xeb_fitting` - functionality related to using xeb to characterize gates by fitting parameters.

[XEB] Support parallel execution

9653093

mpharrigan requested a review from mrwojtek February 6, 2021 05:36

mpharrigan requested review from cduck, vtomole and a team as code owners February 6, 2021 05:36

google-cla bot added the cla: yes Makes googlebot stop complaining. label Feb 6, 2021

mpharrigan added 5 commits February 5, 2021 22:01

lint, format, docstrings

0eeae65

tests

9ccc3cc

we really need to move these functions out of contrib i use them all …

77851ac

…the time

format

5b99859

nbformat

caa4ed2

mpharrigan added the area/xeb label Feb 8, 2021

weird contrib check

53247f2

mrwojtek reviewed Feb 9, 2021

View reviewed changes

docs/characterization/parallel_xeb.ipynb Show resolved Hide resolved

docs/characterization/parallel_xeb.ipynb Show resolved Hide resolved

docs/characterization/parallel_xeb.ipynb Show resolved Hide resolved

docs/characterization/parallel_xeb.ipynb Show resolved Hide resolved

mpharrigan commented Feb 9, 2021

View reviewed changes

mpharrigan added 8 commits February 9, 2021 11:41

review comments 1

daee903

notebook comments

5f03e24

Merge remote-tracking branch 'origin/master' into 2021-02-xeb-3-ll

250c299

Merge remote-tracking branch 'origin/master' into 2021-02-xeb-3-ll

9a323df

imports

7fa0092

fix

9d5fb03

notebook headers

9d46290

Merge remote-tracking branch 'origin/master' into 2021-02-xeb-3-ll

6ea1476

mpharrigan mentioned this pull request Feb 10, 2021

3.1 Factor code into three files mpharrigan/Cirq#8

Open

karlunho mentioned this pull request Feb 10, 2021

Tutorial on detecting drift using parallel XEB #3784

Closed

mrwojtek reviewed Feb 11, 2021

View reviewed changes

Review 2

d5f7948

mrwojtek approved these changes Feb 11, 2021

View reviewed changes

mpharrigan added the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Feb 11, 2021

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Feb 11, 2021

Merge branch 'master' into 2021-02-xeb-3-ll

261a031

CirqBot removed automerge Tells CirqBot to sync and merge this PR. (If it's running.) front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. labels Feb 12, 2021

balopat added the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Feb 12, 2021

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Feb 12, 2021

Merge branch 'master' into 2021-02-xeb-3-ll

ba72f0f

balopat approved these changes Feb 12, 2021

View reviewed changes

CirqBot merged commit e8697de into quantumlib:master Feb 12, 2021

CirqBot removed automerge Tells CirqBot to sync and merge this PR. (If it's running.) front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. labels Feb 12, 2021

mpharrigan mentioned this pull request Feb 12, 2021

[XEB] Split into three files #3794

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[XEB] Support parallel execution #3760

[XEB] Support parallel execution #3760

mpharrigan commented Feb 6, 2021

review-notebook-app bot commented Feb 6, 2021

karlunho commented Feb 9, 2021

mpharrigan left a comment

mpharrigan Feb 9, 2021

mpharrigan Feb 9, 2021

mrwojtek commented Feb 10, 2021

mpharrigan commented Feb 10, 2021

mrwojtek Feb 11, 2021

mpharrigan Feb 11, 2021

mpharrigan commented Feb 11, 2021

CirqBot commented Feb 12, 2021

[XEB] Support parallel execution #3760

[XEB] Support parallel execution #3760

Conversation

mpharrigan commented Feb 6, 2021

review-notebook-app bot commented Feb 6, 2021

karlunho commented Feb 9, 2021

mpharrigan left a comment

Choose a reason for hiding this comment

mpharrigan Feb 9, 2021

Choose a reason for hiding this comment

mpharrigan Feb 9, 2021

Choose a reason for hiding this comment

mrwojtek commented Feb 10, 2021

mpharrigan commented Feb 10, 2021

mrwojtek Feb 11, 2021

Choose a reason for hiding this comment

mpharrigan Feb 11, 2021

Choose a reason for hiding this comment

mpharrigan commented Feb 11, 2021

CirqBot commented Feb 12, 2021