baracuda-cutensor-sys 0.0.1-alpha.68

Raw FFI bindings and dynamic loader for NVIDIA cuTENSOR (tensor contraction).
Documentation

baracuda-cutensor-sys

Raw FFI bindings + dynamic loader for NVIDIA cuTENSOR — high-performance tensor primitives (contraction, reduction, elementwise, permutation) with arbitrary index permutations.

Symbols resolve lazily via libloading; no link-time dependency on libcutensor.so / cutensor64_*.dll.

Most users want baracuda-cutensor — that crate exposes typed Handle, TensorDescriptor, OperationDescriptor, ComputeDescriptor, PlanPreference, and Plan builders, plus the full op catalog (Contraction, Reduction, ElementwiseBinary, ElementwiseTrinary, Permutation, BlockSparseContraction, TrinaryContraction).

What's exposed

  • All cutensor*Descriptor_t types.
  • Plan creation + execution.
  • Op functions: cutensorContract, cutensorReduce, cutensorElementwiseBinaryExecute, cutensorElementwiseTrinaryExecute, cutensorPermute.
  • Plan-cache I/O.

Part of the baracuda workspace.

License

Dual MIT / Apache-2.0.