qumat_qdp: intersect GPU compute capability with PyTorch's arch list

[andrewmusselman]

Fixes Issue #1320 .

Adds _select_torch_device(torch, device_id) in qumat_qdp.loader. It:

• returns "cpu" when CUDA isn't available,
• raises ValueError on out-of-range device_id (preserves prior
  contract),
• checks torch.cuda.get_device_capability(device_id) against
  torch.cuda.get_arch_list() and falls back to "cpu" with a clear
  warnings.warn when the device's sm_NN isn't in the list,
• otherwise returns f"cuda:{device_id}".
  Both qumat_qdp.loader.QuantumDataLoader._create_pytorch_iterator and
  qumat_qdp.api.QdpBenchmark._run_throughput_pytorch use the new helper
  (the latter previously duplicated the same incomplete selection logic).

testing/qdp_python/test_torch_ref.py gets a mirror helper
_torch_cuda_usable() used by the two @skipifs that previously only
checked is_available().

After this change, on an incompatible GPU:

• 8 pytorch-backend loader tests + 4 benchmark tests silently fall back
  to CPU and pass (each emits one UserWarning),
• the 5 explicit GPU tests skip with "CUDA not available or GPU compute capability not supported by this PyTorch build".
  Verified on Linux + GTX 1060 (sm_61) with PyTorch wheel targeting
  sm_70+: tests that previously errored with
  cudaErrorNoKernelImageForDevice now pass or skip cleanly.

[ryankert01]

Make sense! Let copilot double check for me.

[Copilot]

Pull request overview

Fixes #1320 by ensuring PyTorch device selection accounts for the GPU's compute capability against the installed wheel's compiled arch list, avoiding opaque cudaErrorNoKernelImageForDevice failures on older GPUs (e.g., Pascal sm_61 against a sm_70+ wheel). Introduces a shared _select_torch_device helper in qumat_qdp.loader, reuses it in the benchmark path, and mirrors the logic in test skip predicates.

Changes:

• Add _select_torch_device(torch, device_id) in loader.py that validates device_id, intersects device capability with torch.cuda.get_arch_list(), and warns + falls back to CPU on mismatch.
• Replace duplicated CUDA selection logic in QdpBenchmark._run_throughput_pytorch with the shared helper.
• Add _torch_cuda_usable() in test_torch_ref.py and use it in the two GPU-only test skipifs.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
qdp/qdp-python/qumat_qdp/loader.py	New _select_torch_device helper; loader uses it for pytorch backend device pick.
qdp/qdp-python/qumat_qdp/api.py	Benchmark throughput path delegates to the shared helper instead of duplicated logic.
testing/qdp_python/test_torch_ref.py	New _torch_cuda_usable() mirroring the helper's capability check, applied to GPU-only test skipifs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.