This library provides four constant-time implementations of scalar multiplication on Curve25519:

1. PORTABLE is a C-language implementation based on public domain "ref10" code by Daniel J. Bernstein. It uses 26-bit limbs.
2. ARM64 is an optimized 100% assembly implementation for 64-bit ARM CPUs written by Emil Lenngren. It uses 26-bit limbs and can run 3 field operations in parallel by interleaving A64 and NEON instructions.
3. AMD64 is an optimized 100% assembly implementation for 64-bit x86 CPUs. It uses 64-bit limbs.
4. AMD64X is an optimized 100% assembly implementation targeting AMD Ryzen, Intel Broadwell and newer CPUs. It uses the mulx, adcx and adox instructions with 64-bit limbs.

A specific implementation can be selected at runtime using the API. Passing the MX25519_TYPE_AUTO flag automatically selects the fastest implementation supported by the current machine.

Important note: This library does not perform internal clamping of the private key as specified in RFC 7748. This is a deliberate design choice to allow use-cases that require unclamped private keys (for example Carrot). It is the caller's responsibility to clamp the private key as needed before calling mx25519_scmul_base_unclamped or mx25519_scmul_key_unclamped.

Build

git clone https://github.com/tevador/mx25519.git
cd mx25519
mkdir build
cd build
cmake ..
make

./mx25519-tests
./mx25519-bench

On Windows, building with Visual Studio is also supported.

Performance