cuda.core 1.0.0 Release Notes

Highlights

• TBD

New features

• Added managed-memory range operations to cuda.core.utils: Location, advise(), prefetch(), discard(), and discard_prefetch(). Each operation accepts either a single managed Buffer or a sequence; with cuda.bindings 12.8+ the N>1 case dispatches to the corresponding cuMem*BatchAsync driver entry point, addressing the managed-memory portion of #1333. Locations are expressed via the typed Location dataclass (with classmethod constructors device, host, host_numa, and host_numa_current); Device and int values are still accepted for ergonomic compatibility.

Fixes and enhancements

• StridedMemoryView now provides a fast path for torch.Tensor objects via PyTorch’s AOT Inductor (AOTI) stable C ABI. When a torch.Tensor is passed to any from_* classmethod (from_dlpack, from_cuda_array_interface, from_array_interface, or from_any_interface), tensor metadata is read directly from the underlying C struct, bypassing the DLPack and CUDA Array Interface protocol overhead. This yields ~7-20x faster StridedMemoryView construction for PyTorch tensors (depending on whether stream ordering is required). Proper CUDA stream ordering is established between PyTorch’s current stream and the consumer stream, matching the DLPack synchronization contract. Requires PyTorch >= 2.3. (#749)