[QDP] Delegate GPU pointer encoding to QuantumEncoder trait

[viiccwen]

Purpose of PR

Refactors qdp/qdp-core/src/lib.rs so that GPU-pointer encoding (encode_from_gpu_ptr and encode_batch_from_gpu_ptr) is delegated to the QuantumEncoder trait instead of being implemented in large match blocks inside the engine. This aligns the design with the existing host-side encode() and encode_batch() flow and reduces duplication.

Changes

1. QuantumEncoder trait (gpu/encodings/mod.rs)

• Added two optional trait methods (Linux-only, with default implementations that return NotImplemented):
  • encode_from_gpu_ptr(device, input_d, input_len, num_qubits, stream) -> Result<GpuStateVector>
  • encode_batch_from_gpu_ptr(device, input_batch_d, num_samples, sample_size, num_qubits, stream) -> Result<GpuStateVector>
• Encoders that do not implement these (e.g. IQP) get the default and return a clear NotImplemented error.

2. Encoder implementations

• AmplitudeEncoder (gpu/encodings/amplitude.rs): Implemented encode_from_gpu_ptr and encode_batch_from_gpu_ptr (logic moved from lib.rs: norm kernel, amplitude kernel, sync, validation).
• AngleEncoder (gpu/encodings/angle.rs): Same; angle validation and launch_angle_encode / launch_angle_encode_batch with stream.
• BasisEncoder (gpu/encodings/basis.rs): Same; launch_basis_encode_batch for both single (num_samples=1) and batch.

3. QdpEngine (lib.rs)

• encode_from_gpu_ptr_with_stream: Replaced the large match method.as_str() { "amplitude" => ..., "angle" => ..., "basis" => ..., _ => ... } with:
  • get_encoder(encoding_method)?
  • encoder.encode_from_gpu_ptr(&self.device, input_d, input_len, num_qubits, stream)?
  • state_vector.to_precision(&self.device, self.precision)?
  • Ok(state_vector.to_dlpack())
• encode_batch_from_gpu_ptr_with_stream: Same pattern using encode_batch_from_gpu_ptr; kept the num_samples == 0 check in the engine before calling the encoder.
• Removed the inline TODOs that asked for this refactor.

Related Issues or PRs

Closes #1005

Changes Made

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Breaking Changes

• Yes
• No

Checklist

• Added or updated unit tests for all changes
• Added or updated documentation for all changes
• Successfully built and ran all unit tests or manual tests locally
• PR title follows "MAHOUT-XXX: Brief Description" format (if related to an issue)
• Code follows ASF guidelines

[guan404ming]

Overall looks nice!

[guan404ming]

I think this computes L2 norms (expensive) just to check finiteness (cheap side effect). Maybe we could introduce a dedicated check_finite_batch kernel would be more direct. I think we could do it in a follow up and please correct me if I'm wrong. Thanks!

cc @rich7420 @ryankert01

[viiccwen]

Agree : )

[viiccwen]

I'll fix conflict later :)

[viiccwen]

cc @guan404ming
sorry for late.