[places] Guard empty tiled mdspan shape size

[fallintoplace]

What changed

• return 0 early from tiled_mdspan_shape::size() when the original shape is empty
• add a regression test covering the empty-shape case

Why

tiled_mdspan_shape::size() computes tile_per_part and then uses (tile_per_part - 1) in unsigned arithmetic.

For empty input, original_shape.size() is 0, so ntiles == 0 and tile_per_part == 0. That makes (tile_per_part - 1) underflow and can turn an empty shape into a huge nonzero size.

Impact

Empty tiled mdspan shapes now report 0 elements instead of underflowing.

Validation

• added a focused header unittest for the empty-shape case
• attempted cmake --preset cudax-cpp17, but validation is blocked in this environment because nvcc / CUDAToolkit is not available

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 5be93582-1bb9-45cd-bd0e-f44a1e0af5ab

📥 Commits

Reviewing files that changed from the base of the PR and between 1e2d0c6 and 8e5f3ad.

📒 Files selected for processing (1)

• cudax/include/cuda/experimental/__places/partitions/tiled_partition.cuh

🚧 Files skipped from review as they are similar to previous changes (1)

• cudax/include/cuda/experimental/__places/partitions/tiled_partition.cuh

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Tiling partition now correctly reports a size of zero for empty (zero-length) inputs, preventing unnecessary tiling computations.

• Tests

  • Added a unit test to verify empty-input handling for tiled partition sizing.

important:

Walkthrough

tiled_mdspan_shape::size() adds an early-return guard returning 0 when the underlying 1D shape has zero elements, avoiding tiling math for empty inputs. A unit test validates this behavior, confirming that partitioned empty shapes correctly report size 0.

Changes

Empty tiled partition size handling

Layer / File(s)	Summary
Early return for empty shapes and test cudax/include/cuda/experimental/__places/partitions/tiled_partition.cuh	tiled_mdspan_shape::size() returns 0 immediately when input shape size is zero, bypassing tile calculation. Unit test constructs a mocked zero-element shape and verifies the method returns 0 across tile_size=4, part_id=0, nparts=3 configurations.

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[miscco]

/ok to test 1e2d0c6

[github-actions]

😬 CI Workflow Results

🟥 Finished in 1h 03m: Pass: 98%/55 | Total: 1d 00h | Max: 1h 02m | Hits: 12%/168641

See results here.