Performing gate split operation using cuTensorNet adopts a very similar workflow as QR example and SVD example. Here, we here highlight the notable differences between the two APIs. The full code can be found in the NVIDIA/cuQuantum repository (here).

Define tensor operands

As with QR/SVD decomposition, we first define all the tensor operands by specifying the data type, modes partition, and the extents. In this process, we choose to perform fixed extent truncation to a size of 16.

Execution

Similar to SVD example, we can specify the SVD options in cutensornetTensorSVDConfig_t. Workspace size query can be achieved by calling cutensornetWorkspaceComputeGateSplitSizes() with the provided cutensornetGateSplitAlgo_t. Finally, we can execute the gate split computation by calling cutensornetGateSplit().

Note

Like in cutensornetTensorSVD(), since we turned on weighted truncation options in this example, the tensor descriptors for outputs A and B need to be restored if we wish to perform the same computation multiple times.

After the computation, we always need to free up all resources.