af hamming matcher produces incorrect results for 64-bit unsigned integer types on CUDA
af hamming matcher produces incorrect results for 64-bit unsigned integer types on CUDA: a task in LegoFlow-SWE (Harbor dataset). When using the af hamming matcher function with query and training arrays of type uintl (64‑bit unsigned integers) on the CUDA backend, the returned nearest‑neighbor…
The task
When using the `af_hamming_matcher` function with query and training arrays of type `uintl` (64‑bit unsigned integers) on the CUDA backend, the returned nearest‑neighbor indices and Hamming distances are incorrect or the operation fails silently. The function works correctly for smaller unsigned types (`uchar`,…
Part of Lego-X/LegoFlow-SWE.