Improve speed of random circuit generation #4428
Merged
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
daxfohl
requested review from
a team,
cduck,
mrwojtek and
vtomole
as code owners
mpharrigan
requested changes
Aug 13, 2021
Collaborator
mpharrigan
left a comment
mpharrigan
left a comment
There was a problem hiding this comment.
I don't understand the logic of how this works
cirq-core/cirq/experiments/random_quantum_circuit_generation.py
Outdated
Show resolved
Hide resolved
Collaborator
|
Technically a breaking change because of the change of API for the gate factories and the seed issue you noted. I think it should be fine |
mpharrigan
added
the
BREAKING CHANGE
Collaborator
|
Do you have profiling data? Does this indeed speed things up? |
Collaborator
Author
|
Improves generation of 5 circuits of depth 13000 from around 7.3s to around 5.1s on my machine. The time lost due to the "retry" strategy compared to the deterministic strategy was insignificant. |
mpharrigan
approved these changes
Aug 13, 2021
Collaborator
|
Sweet. I'm a little surprised the retry logic doesn't cause a slowdown; but I guess it's exceedingly unlikely for you to have to retry more than a couple times. Thanks for profiling it. |
mpharrigan
added
the
automerge
CirqBot
added
the
front_of_queue_automerge
CirqBot
removed
automerge
Collaborator
Author
|
FWIW I also tried an alternate approach, just caching the canonicalization of the XZPowGate, since that is called by the equality checker, and the same gates were reused over and over. That reduced the time from 7.3s to 6.1s. So that's about half the overhead in the equality tester for XZPowGates. Don't think we should do anything since we've solved the problem better already. Just a data point. |
Collaborator
Author
|
Wait, this was really dumb. After reading the above again, I realized all we had to do for this is change |
daxfohl
added a commit
to daxfohl/Cirq
that referenced
this pull request
Aug 14, 2021
This reverts commit 75eccd3.
CirqBot
pushed a commit
that referenced
this pull request
Aug 14, 2021
Collaborator
|
I'll remove the breaking change label since this was reverted |
mpharrigan
removed
the
BREAKING CHANGE
CirqBot
pushed a commit
that referenced
this pull request
Aug 16, 2021
Fixes #4337 (again) Changes equality check to identity check to give the perf improvement. Since the gate instances are reused, and since equality checking if XZPowGate is very expensive, we change equality check to identity check and get a big perf boost. This issue was originally fixed in a different way in #4428, but that has been reverted in llieu of this fix which is simpler, faster, and not a breaking change. ref #4428 (comment) @mpharrigan
rht
pushed a commit
to rht/Cirq
that referenced
this pull request
May 1, 2023
Fixes quantumlib#4337 Changes single-qubit layer generation into two steps. First step creates the `{qid: gate_index}` lookup so that we can compare against gate_index equality instead of gate equality. Then the second step is generating the moment from that lookup. ~~Note the change in algo will cause the random circuits generated by a specific seed to change (hence the change in one test case). If this is a blocker, it should be possible to revert to the old retry algo but just compare indexes instead of gates.~~ Reverted to the retry algo since some of the xeb tests were tied to the generated circuit. Also note any change like this assumes that each gate only appears once in the allowed-gates list. I assume that's a safe assumption. @mpharrigan
rht
pushed a commit
to rht/Cirq
that referenced
this pull request
May 1, 2023
rht
pushed a commit
to rht/Cirq
that referenced
this pull request
May 1, 2023
Fixes quantumlib#4337 (again) Changes equality check to identity check to give the perf improvement. Since the gate instances are reused, and since equality checking if XZPowGate is very expensive, we change equality check to identity check and get a big perf boost. This issue was originally fixed in a different way in quantumlib#4428, but that has been reverted in llieu of this fix which is simpler, faster, and not a breaking change. ref quantumlib#4428 (comment) @mpharrigan
harry-phasecraft
pushed a commit
to PhaseCraft/Cirq
that referenced
this pull request
Oct 31, 2024
Fixes quantumlib#4337 Changes single-qubit layer generation into two steps. First step creates the `{qid: gate_index}` lookup so that we can compare against gate_index equality instead of gate equality. Then the second step is generating the moment from that lookup. ~~Note the change in algo will cause the random circuits generated by a specific seed to change (hence the change in one test case). If this is a blocker, it should be possible to revert to the old retry algo but just compare indexes instead of gates.~~ Reverted to the retry algo since some of the xeb tests were tied to the generated circuit. Also note any change like this assumes that each gate only appears once in the allowed-gates list. I assume that's a safe assumption. @mpharrigan
harry-phasecraft
pushed a commit
to PhaseCraft/Cirq
that referenced
this pull request
Oct 31, 2024
harry-phasecraft
pushed a commit
to PhaseCraft/Cirq
that referenced
this pull request
Oct 31, 2024
Fixes quantumlib#4337 (again) Changes equality check to identity check to give the perf improvement. Since the gate instances are reused, and since equality checking if XZPowGate is very expensive, we change equality check to identity check and get a big perf boost. This issue was originally fixed in a different way in quantumlib#4428, but that has been reverted in llieu of this fix which is simpler, faster, and not a breaking change. ref quantumlib#4428 (comment) @mpharrigan
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Fixes #4337
Changes single-qubit layer generation into two steps. First step creates the
{qid: gate_index}lookup so that we can compare against gate_index equality instead of gate equality. Then the second step is generating the moment from that lookup.Note the change in algo will cause the random circuits generated by a specific seed to change (hence the change in one test case). If this is a blocker, it should be possible to revert to the old retry algo but just compare indexes instead of gates.Reverted to the retry algo since some of the xeb tests were tied to the generated circuit.Also note any change like this assumes that each gate only appears once in the allowed-gates list. I assume that's a safe assumption.
@mpharrigan