- Add instance specific functions to the public API
- Small improvements and fixes.
- Fix alignment of buffers accessed with SIMD instructions when building with MSVC.
- Ensure that builds fail if alignment cannot be specified.
- Fix debug build.
- Ensure the memory in Merkle tree helper function is initialized.
- Ensure that public keys stored in private keys are valid.
- Small improvements in memory handling.
- Add the
plain32
implementation of SHAKE and automatically enable it for 32-bit targets. - Avoid
__builtin_cpu_supports
on Apple as support is spotty. - Reduce memory usage.
- Improve matrix multiplication for L5 instances on AVX2. This improves the performance of Picnic-L5-{FS,UR} by up to 5 %.
- Reduce memory usage and initialization cost for trees.
- Make the use of
mzd_local_t
as SIMD registers explicit. This change avoids potential issues when not following the rules for accessing members of unions. - Clear temporary copies of private keys.
- Fix NEON code on M1.
- Ensure SSE2/AVX2/NEON shift intrinsics with immediate operands are used correctly.
- Use Boost.Test as unit test framework.
- Fix build with llvm on ARM with NEON enabled.
- Unbreak x86-32 build.
- Fix build on M1 with NEON enabled.
- Prefix compat function implementations with
picnic_
. - Use OQS instruction set checking functions.
- Use OQS implementations of
aligned_alloc
,aligned_free
,explicit_bzero
, andtimingsafe_bcmp
. - Install cmake configuration files.
- Various changes to improve OQS integration.
- Require cmake version 3.10.
- Reduce size of global parameters for instance specification to 12 bytes per instance.
- Provide compat implementation of
clz
on MSVC using_BitScanReverse
. - Do not assume that
aligned_alloc
is available on MSVC.
- Update SHAKE3 implementation.
- Fix build with GCC 11.
- Reduce memory consumption slightly.
- Expose
picnic_get_{private,public}key_size
as part of the public API. - Add
picnic_get_{private,public}_key_param
to retrieve a key's parameter set.
- Slightly improve memory consumption.
- Initial work to support PQClean integration in the future.
- Add cmake options to control availability of specific LowMC instances.
- Fix
explicit_bzero
fallback implementation. - Remove some unused code.
- Update SHAKE3 implementation.
- Add support to check constant time implementation with TIMECOP.
- Slightly reduce memory consumption.
- Add support for BSD variants.
- Expose
picnic_sk_to_pk
as part of the public API. - Add
picnic_clear_private_key
to clear the private key.
- Implement new Picnic 3 parameter set. This implementation replaces the Picnic 2 parameter set.
- Implement new Picnic instances with full Sbox layer.
- Various small improvements and bug fixes.
- Remove all optimizations for partial LowMC instances except for OLLE.
- Fix Picnic2 implementation on big endian systems.
- Add support for SHA3/SHAKE3 instructions on IBM z.
- Various small improvements and bug fixes.
- Remove LowMC instances with m=1.
- Add options to build with ZKB++- or KKW-based instances only.
- Fix ARM NEON optimizations.
- Slightly reduce heap usage.
- Remove more unused code.
- Various small improvements and bug fixes.
- Remove M4RM-based implementation.
- Fix input size in Picnic2's commitment implementation.
- Additional improvements and optimizations of the Picnic2 code.
- Implement Picnic 2.
- Use 4-times parallel SHAKE3 for faster PRF evaluation, commitment generation, etc.
- Fix size of salts to 32 bytes.
- Reduce heap usage.
- Implement linear layer optimizations to speed up LowMC evaluations. Besides the runtime improvements, this optimization also greatly reduces the memory size of the LowMC instances.
- Provide LowMC instances with m=1 to demonstrate feasibility of those instances.
- Slightly improve internal storage of matrices to require less memory.
- Remove unused code and support for dynamic LowMC instances.
- Implement RRKC optimizations for round constants.
- Compatibility fixes for Mac OS X.
- Reduce memory usage when using Fiat-Shamir and Unruh transform in the same process.
- Fix deviations from specification. The KDF was missing the output length as input and the public key was incorrectly serialized. Note that this change requires an update of the test vectors.
- Update SHA3 implementation and fix endiannes bug on big-endian.
- Record state before Sbox evaluation and drop one branch of XOR computations. This optimization is based based on an idea by Markus Schofnegger.
- Add per-signature salt to random tapes generation. Prevents a seed-guessing attack reported by Itai Dinur.
- Compatibility fixes for Visual Studio, clang and MinGW.
- Various improvements to the SIMD versions of the matrix operations.
- Default to constant-time matrix multiplication algorithms without lookup tables.
- Add option to feed extra randomness to initial seed expansion to counter fault attacks.
- Version submitted for inclusion in SUPERCOP.
- Initial release.
- Version submitted to the NIST PQC project.