Add rust representation of the UnitaryGate class
#13272
Labels
Milestone
Comments
mtreinish
added
performance
Rust
mtreinish
added a commit
to mtreinish/qiskit-core
that referenced
this issue
Oct 4, 2024
In the accelerate crate we currently have two places that are building `UnitaryGate`, the quantum_volume() function that builds a quantum volume model circuit and the Split2QUnitaries transpiler pass. Currently this can only be done by calling Python (until Qiskit#13272 is implemented) and there was a potential optimization we could make to specify the number of qubits as an argument to the Python constructor. This skipss the need for a few python operations to compute the number of qubits from the size of the matrix. These operations are not exceedingly slow, but as these Python space constructors are often the bottleneck so it should help runtime performance of these two functions.
github-merge-queue bot
pushed a commit
that referenced
this issue
Oct 7, 2024
In the accelerate crate we currently have two places that are building `UnitaryGate`, the quantum_volume() function that builds a quantum volume model circuit and the Split2QUnitaries transpiler pass. Currently this can only be done by calling Python (until #13272 is implemented) and there was a potential optimization we could make to specify the number of qubits as an argument to the Python constructor. This skipss the need for a few python operations to compute the number of qubits from the size of the matrix. These operations are not exceedingly slow, but as these Python space constructors are often the bottleneck so it should help runtime performance of these two functions.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The
UnitaryGateclass is defined in Python and that means that when we need to work with it from Rust (either for transpiler passes or circuit construction) we treat it as aPyInstructionand that requires that we use Python for interacting with it. For the most part for reading this isn't a huge overhead because aUnitaryGateis a gate defined around a unitary matrix as a numpy array and we typically just need to inspect that matrix which can be done by reference without any copying or conversion. However, on writing (typically creation), which we do a fair amount of in the transpiler and the circuit library, the python overhead is significant. In rust it will be easy to build aUnitaryGatestruct that implements theOperationtrait and owns anArray2<Complex64>. We might need to think a bit about the python interface and split of responsibilities between the languages because we can generate a numpy array from a owned array but the traits in rust-numpy for doing that either consume the array or copy it.This weakly depends on #12966 as we'll need to figure out the best way to expand PackedOperation to support this and the pattern we establish in #12966 will inform this.
The text was updated successfully, but these errors were encountered: