[ENH] Optimize TRX output

[36000]

No description provided.

[Copilot]

Pull request overview

This PR aims to optimize TRX generation in the CUDA tractography pipeline by adjusting batching defaults and TRX preallocation/output dtypes.

Changes:

• Reduce the default chunk_size (seeds processed per chunk per GPU) from 100000 to 25000 and document it in the class docstring.
• Increase TRX preallocation heuristic sl_per_seed_guess from 3 to 4.
• Cast TRX streamline storage arrays to float32 (points) and uint64 (offsets).

Comments suppressed due to low confidence (1)

cuslines/cuda_python/cu_tractography.py:52

• Changing the default chunk_size from 100000 to 25000 is a behavioral change for callers that rely on defaults (it can increase kernel launch/host overhead and alter memory/perf characteristics). If this is intended, consider documenting it as a breaking/behavioral change (e.g., in release notes/README) or keeping the previous default and adjusting it only in the CLI/config path that needs it.

        chunk_size: int = 25000,

---

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

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

cuslines/cuda_python/cu_tractography.py:258

• The TODO comment about outputting in VOX space directly was removed, but the code on line 290 still calls tractogram.to_world(), which suggests the conversion is still happening. If the TODO was meant to be addressed by this PR, the conversion should be removed or modified. If the TODO is not being addressed, it should remain in the code.

    def generate_trx(self, seeds, ref_img):

---

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