Skip to content

Latest commit

 

History

History
53 lines (32 loc) · 2.15 KB

changelog.md

File metadata and controls

53 lines (32 loc) · 2.15 KB
Raw

2024-06-06

  • Use NTT form for BGV Ciphertexts. (Microsoft SEAL 4.1.0)

2024-06-12

  • Use AES-CTR (inspired by torchcsprng) to generate randomness. In the previous versions, we used the cuRAND/BLAKE2 to generate randomness for GPU/CPU respectively. But cuRAND's default XORWOW is not cryptographic secure.

2024-06-20

  • Implement CKKS matrix multiplication.

2024-06-21

  • Add examples.
  • Provide an unsafe implementation of the Memory Pool.

2024-06-24

  • Implement ciphertext-plaintext support for Ring2k BFV.
    • The user may use plain modulus $t = 2^k$ for $k \leq 128$, with PolynomialEncoderRing2k<T>, where T = uint32_t, uint64_t, uint128_t, and the given k must be greater than half of the type T's bitwidth.
    • scale_up, centralize are used for encoding, while scale_down is for decoding. See the evaluator unit tests for some usages.
    • Matrix multiplication support for Ring2k-BFV.
    • Pybind11 encapsulation. This only includes uint32_t and uint64_t versions, since native 128-bit support is missing in pybind11/python/numpy.

2024-06-26

  • Implement invariant_noist_budget in Decryptor, available for BFV and BGV.

2024-06-29

  • Allow user to create MemoryPoolHandles and supply them to API calls.
    • Add multithread tests and benchmark.
    • Update pybind11 encapsulation and test folder structure.

2024-07-03

  • Move the implementation of MemoryPool to different input files. Directly using macros in the header may lead to user's on including troy wrong with defines not given.

2024-07-14

  • Allow scaled up polynomial plaintexts to store only partial coefficients, to save device memory when only a small degree polynomial is encoded.

2024-08-09

  • Update kernels for multiple APIs in evaluator and encryptor, removing redundant cuda memory copies and memsets.
  • All unittests using device will be skipped if no device is detected on the machine.
  • Examples will run on host if no device is detected on the machine.

2024-09-06

  • Use cudaMemcpyAsync instead of cudaMemcpy when copying from host to device, and device to device.
  • Use cudaMemsetAsync instead of cudaMemset when setting device memory.