[Bug] benchmark files (latency, throughput) allow basis/angle selections without cross-framework encoding parity

[viiccwen]

The main cross-framework benchmark scripts accept:

• --encoding-method amplitude
• --encoding-method angle
• --encoding-method basis

But only the Mahout leg actually uses the selected encoding method.

The PennyLane and Qiskit legs are still hard-coded to amplitude-style state preparation, so a run such as:

uv run --project qdp/qdp-python python qdp/qdp-python/benchmark/benchmark_throughput.py \
  --frameworks mahout,pennylane \
  --encoding-method basis

does not produce a fair basis benchmark.

Evidence

• qdp/qdp-python/benchmark/benchmark_throughput.py
  • Mahout path uses .encoding(encoding_method)
  • PennyLane path uses qml.AmplitudeEmbedding(...)
  • Qiskit path uses qc.initialize(...)
• qdp/qdp-python/benchmark/benchmark_latency.py
  • Mahout path uses .encoding(encoding_method)
  • PennyLane path uses qml.AmplitudeEmbedding(...)
  • Qiskit paths use qc.initialize(...) / Statevector(...)

Why this matters

This makes the benchmark output potentially misleading:

• Mahout may run basis or angle
• competitor frameworks still run amplitude-style preparation
• the script still prints a single comparison table as if all legs measured the same workload

That is a benchmark validity bug, not just a docs problem.

Proposed fix

Short term:

• reject non-amplitude cross-framework combinations
• allow angle / basis only for Mahout-only runs
• update benchmark docs/examples to state that cross-framework parity currently exists only for amplitude

Longer term:

• add true angle / basis parity for competitor frameworks where semantics can be matched reliably

[viiccwen]

@guan404ming, @rich7420 IMO, we should restrict first (only allow fair comparison), and extends more encoding method to Pennylane & Qiskit, WDYT?

[400Ping]

@viiccwen you wanna take a stab at this since there's no other contributors want to do it?

[viiccwen]

@viiccwen you wanna take a stab at this since there's no other contributors want to do it?

Sure, I'll take this one after 14 Apr.