Tensor arithmetic support

Hello,

newcomer to OpenFHE here, long-time user of TenSEAL. In TenSEAL a user can handle tensors natively. For example, CKKSTensors can participate in native matmul, matvec operations (I guess TenSEAL internally handles necessary packing or reordering or rotations for efficient multiplications). I have used openfhe-numpy which offers a similar interface for the basic operations but I have failed to achieve the performance of TenSEAL especially when working with tensors (because in order to achieve this in openfhe-numpy you still need to iterated over rows-columns). Have I missed something when it comes to tensor support in OpenFHE or is this indeed an internal limitation of the library?

Hi - Please quantify your results regarding the performance gap with TenSEAL? We need to understand your specific use case to help.

  1. What specific workload or operation are you trying to perform?
  2. Can you confirm if you are working with single-block tensors, that is, 2D tensors that fit in a single ciphertext?