GitLab

* Reduce conversion. Assume that ray_indices should always be long and pack_info always int

* Remove some more casting

* More casting fixes

* Remove unecessary long casting

* Fix changes in precision at python level, ray_pdf_query test is already broken since it was reverted

* Support long when rendering transmittance

* Support long

* Woops

File proposal for #117

This reverts commit 3090d7bd.

- Faster rendering function via nvidia-cub, shipped with cuda >= 11.0 (Require >=11.6 for out use). ~10% speedup
- Expose transmittance computation.

This reverts commit 301b4dfa.

* support alpha for both marching and rendering; relax pytorch requirement

* bump version