Add ASV benchmarks for BasicSimulator Clifford optimization #15448
Conversation
|
One or more of the following people are relevant to this code:
|
Pull Request Test Coverage Report for Build 20265672539Details
💛 - Coveralls |
alexanderivrii
left a comment
alexanderivrii
left a comment
There was a problem hiding this comment.
Thanks for adding these well thought-out set of benchmarks @debasmita2102. A few minor comments, but otherwise LGTM.
| def setup(self, n_qubits): | ||
| """Setup random Clifford circuit for given n_qubits.""" | ||
|
|
||
| cliff = random_clifford(n_qubits) |
There was a problem hiding this comment.
This call to random_clifford is not deterministic, consider providing an explicit seed, for example:
| cliff = random_clifford(n_qubits) | |
| cliff = random_clifford(n_qubits, seed=0) |
I now see that there are other ASV tests that use random_clifford with non-deterministic results, I think we also fix those (but in a separate PR).
| def time_statevector(self, n_qubits): | ||
| """Time statevector simulation of random Clifford circuit.""" | ||
| _ = n_qubits # ASV parameter |
There was a problem hiding this comment.
Would something like this work as well?
| def time_statevector(self, n_qubits): | |
| """Time statevector simulation of random Clifford circuit.""" | |
| _ = n_qubits # ASV parameter | |
| def time_statevector(self, _n_qubits): | |
| """Time statevector simulation of random Clifford circuit.""" |
| def time_clifford(self, n_qubits): | ||
| """Time Clifford-optimized simulation of random Clifford circuit.""" | ||
| _ = n_qubits # ASV parameter |
| """Time Clifford-optimized simulation of random Clifford circuit.""" | ||
| _ = n_qubits # ASV parameter | ||
| backend = BasicSimulator() | ||
| backend.run( |
There was a problem hiding this comment.
Would it make sense to add _ = here for readability, expressing that the call returns a result (which is however ignored)?
| backend.run( | |
| _ = backend.run( |
If you think it does, then this comment applies to other similar calls in this file.
| from qiskit.providers.basic_provider import BasicSimulator | ||
| from qiskit.quantum_info import random_clifford | ||
|
|
||
| # pylint: disable=attribute-defined-outside-init |
There was a problem hiding this comment.
I don't know if it matters (and most probably it does not), but in all other tests the # pylint: disable line appears before the module dostring/imports. Feel free to change this or leave it as is.
| def time_statevector(self, n_qubits): | ||
| """Time statevector simulation of GHZ circuit.""" | ||
| _ = n_qubits # ASV parameter | ||
| backend = BasicSimulator() |
There was a problem hiding this comment.
This is fast, right (otherwise, you could also move this construction to setup as well).
|
Note that we are able merge this PR before #15159: calling |
ShellyGarion
added
the
type: qa
ShellyGarion
left a comment
ShellyGarion
left a comment
There was a problem hiding this comment.
@debasmita2102 - Thanks for adding ASV tests!
Are there similar tests for the BasicSimulator for non-Clifford circuits? If not, maybe it's worth adding them in a seperate PR.
My main comment is that with a Clifford simulator we can extend the number of qubits in the tests for more than 24. But this can be done only after we merge #15159.
| class BasicSimulatorGHZBenchmark: | ||
| """Benchmark BasicSimulator on GHZ Clifford circuits.""" | ||
|
|
||
| params = ([4, 8, 12, 16],) |
There was a problem hiding this comment.
I wonder if after we merge #15159, we could extend the number of qubits here
| class BasicSimulatorRandomCliffordBenchmark: | ||
| """Benchmark BasicSimulator on random Clifford circuits.""" | ||
|
|
||
| params = ([4, 8, 12, 16],) |
There was a problem hiding this comment.
I wonder if after we merge #15159, we could extend the number of qubits here
ShellyGarion
added
the
mod: quantum info
5 participants
Summary
Add ASV benchmarks to track performance of the
BasicSimulatorClifford (StabilizerState) path versus the existing statevector path.The benchmarks are split out from PR Add Clifford circuit optimization to BasicSimulator #15159 so that code changes and performance tracking can be reviewed independently.
Details and comments
BasicSimulatorGHZBenchmarkto compare Clifford vs statevector performance on GHZ-like Clifford circuits (two output bitstrings) forn_qubits = [4, 8, 12, 16].BasicSimulatorRandomCliffordBenchmarkto compare Clifford vs statevector performance on random Clifford circuits generated viarandom_clifford(n_qubits)for the same qubit sizes.