mlsgpu
mlsgpu copied to clipboard
bmerry
Reconstruction of triangulated surfaces from range-scanned point clouds
There is currently no testing for the case where the total output exceeds 232 vertices (split across multiple files), or for when a single output file exceeds 4GiB.
Currently internal welding is baked into Marching::shipOut, but it could be implemented as a MeshFilter. This would make it easier to test in isolation.
This should at a minimum remove any temporary files that are in progress.
Currently the startup code checks that the device supports OpenCL 1.1. It should be sufficient if the platform supports OpenCL 1.1, even if the device is only OpenCL 1.0 compliant...
Currently every rank outputs the usage message, and they end up all interleaved.
The binary_io tests on Linux create a >9GB file to test that the interfaces are 64-bit clean, but this is currently scaled back on Windows due to poor performance (as...