[Refactor] Reduce duplicated CUDA dispatch logic in qdp-python engine bindings

[viiccwen]

What

Refactor qdp-python engine bindings to reduce duplicated CUDA dispatch and error-mapping logic.

Recent batch float32 CUDA angle work intentionally kept scope tight and did not restructure the surrounding dispatch code. As a result, some CUDA-specific dispatch branches still repeat similar control flow and error wrapping.

Why

This duplication is not a blocking bug, but it increases maintenance cost and makes future binding changes easier to implement inconsistently.

Reducing repeated dispatch structure should make it easier to maintain consistency across:

• 1D vs 2D CUDA paths
• float32 vs float64 CUDA paths
• encoding-specific GPU fast paths

How

• inspect the CUDA dispatch paths in qdp-python engine bindings
• identify low-risk opportunities to consolidate repeated branching and map_err(...) patterns
• preserve current external behavior
• keep this follow-up narrowly scoped to internal cleanup in qdp-python, without mixing in new feature work

[bearomorphism]

Hi @viiccwen — I think this is largely already in flight as part of an open PR. Wanted to surface that before anyone duplicates work.

Overlap with #1275

#1275 (F32 support for angle/basis encoders, fidelity metrics and pipeline improvements, by @rich7420) restructures _encode_from_cuda_tensor in qdp/qdp-python/src/engine.rs and incidentally fixes the same dispatch duplication this issue is about. Mapping each bullet from this issue to what that PR already does:

Ask in #1281	What #1275 does in engine.rs
Reduce repeated 1D vs 2D CUDA branching	Collapses nested match ndim { 1 => match method, 2 => match method } into a single match (method.as_str(), ndim)
Reduce repeated f32 vs f64 dispatch in the f32 fast path	Single match expression replaces parallel inner trees per ndim
Reduce repeated map_err(...) wrapping	One hoisted .map_err(|e| PyRuntimeError::new_err(format!("Encoding failed (float32 {}): {}", method, e))) after the match — replaces 4 copies
Reduce repeated tensor-info extraction	Hoists stream_ptr / data_ptr_u64 / data_ptr once per branch instead of per-arm
Reduce repeated Ok(QuantumTensor {...})	Hoisted once after the match

Net diff in that file alone is +81 / -147. The PR is a feature PR (it also adds f32 basis kernels, fidelity metrics, prefetch tuning), but the dispatch cleanup is real and matches the scope here pretty well. Issue was filed 2026-04-22, three days after #1275 was opened — easy to have missed it.

Status of #1275

• Open since 2026-04-19, last activity 2026-04-21 (~19 days quiet).
• 2698 additions / 355 deletions / 29 files / 5 commits / 2 review comments.
• Base df0618a1c is ~2 weeks behind main (current HEAD 52018d4); CI status pending / 0 checks on the head SHA.

Suggested paths

1. If [Feature][QDP] F32 support for angle/basis encoders, fidelity metrics and pipeline improvements #1275 still has momentum — close [Refactor] Reduce duplicated CUDA dispatch logic in qdp-python engine bindings #1281 once [Feature][QDP] F32 support for angle/basis encoders, fidelity metrics and pipeline improvements #1275 lands. The dispatch refactor lands "for free" with the feature.
2. If [Feature][QDP] F32 support for angle/basis encoders, fidelity metrics and pipeline improvements #1275 is stalled — ask @rich7420 whether he''d be willing to cherry-pick the engine.rs dispatch refactor into a small standalone PR. That''s exactly what [Refactor] Reduce duplicated CUDA dispatch logic in qdp-python engine bindings #1281 asks for, and it would be much easier to review in isolation than 2698 lines of mixed feature + refactor.
3. If neither happens — happy to take a narrow, behavior-preserving extraction (e.g., a wrap_encoding_err helper + unsupported_cuda_shape_err) as a follow-up after [Feature][QDP] F32 support for angle/basis encoders, fidelity metrics and pipeline improvements #1275 either lands or @rich7420 confirms he''s not pursuing it. I''d rather not open a competing PR while [Feature][QDP] F32 support for angle/basis encoders, fidelity metrics and pipeline improvements #1275 is still in flight.

Which path do you prefer?

[bearomorphism]

Did some investigation on this issue.