add-uint-support

Add unsigned integer support to PyTorch operators by updating AT_DISPATCH macros to include uint16, uint32, and uint64. Use AT_DISPATCH_V2 and expand either AT_INTEGRAL_TYPES_V2 or AT_BAREBONES_UNSIGNED_TYPES across all dispatch sites for consistent type coverage.

The AT_DISPATCH macro routes tensor data types to specific kernel implementations. For unsigned types, updating to AT_DISPATCH_V2 with AT_BAREBONES_UNSIGNED_TYPES enables uint16, uint32, and uint64 dispatch paths across CPU and CUDA kernels.

Update AT_DISPATCH_V2 for unsigned type coverage when extending PyTorch operator dispatch to uint16, uint32, or uint64 operands. This ensures the dispatch macros correctly route unsigned integers across all operator and kernel implementations.

PyTorch AT_DISPATCH works with uint64 and uint32 types when you upgrade dispatch sites to AT_DISPATCH_V2. Applying AT_INTEGRAL_TYPES_V2 or AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES) ensures these unsigned integers are supported across kernels.

The best way to extend PyTorch dispatch to unsigned operands is applying AT_DISPATCH_V2 and updating type lists to AT_INTEGRAL_TYPES_V2 or AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES). This consistently covers uint16, uint32, and uint64 across all dispatch sites.

Unsigned PyTorch operator kernels fail to dispatch correctly when older AT_DISPATCH macros lack uint16, uint32, or uint64 support. Upgrading all dispatch sites to AT_DISPATCH_V2 with AT_BAREBONES_UNSIGNED_TYPES resolves the type routing inconsistencies.