What
Add regression tests for DLPack and zero-copy interoperability on CUDA batch angle encoding paths in qdp-python.
Recent work added batch float32 CUDA angle encoding support in the Python bindings. This follow-up should strengthen regression coverage around the Python <-> PyTorch handoff for that path.
Why
One of the main reasons to use the CUDA tensor path is zero-copy integration through DLPack.
The current feature tests cover correctness and validation behavior, but they do not fully protect interoperability expectations such as:
- compatibility with
torch.from_dlpack(...)
- output device placement
- output dtype following engine precision
- output shape for batched inputs
A dedicated regression layer here would make the zero-copy workflow safer to evolve.
How
- add qdp-python tests that consume CUDA batch angle outputs through
torch.from_dlpack(...)
- assert expected device, dtype, and shape
- include at least one deterministic batch case for state-value sanity checks
- keep the scope limited to DLPack / interop regression coverage
What
Add regression tests for DLPack and zero-copy interoperability on CUDA batch angle encoding paths in qdp-python.
Recent work added batch
float32CUDA angle encoding support in the Python bindings. This follow-up should strengthen regression coverage around the Python <-> PyTorch handoff for that path.Why
One of the main reasons to use the CUDA tensor path is zero-copy integration through DLPack.
The current feature tests cover correctness and validation behavior, but they do not fully protect interoperability expectations such as:
torch.from_dlpack(...)A dedicated regression layer here would make the zero-copy workflow safer to evolve.
How
torch.from_dlpack(...)